A model is still running

A model is still running or downloading. Leaving will stop the current operation. Stay on this page to let it finish.

Skip to content
LocalAI.toolsBeta

Join the waitlist

Get an email when the new apps and model runs launch.

No account needed.

KittenTTS 2

Generate local preset speech with the exact KittenTTS 2 Mini · Multilingual checkpoint, complete 24 kHz WAV output and genuine incremental audio.

Desktop inferencePrivate · Runs on your computer
About this model

Generate local preset speech with the exact KittenTTS 2 Mini · Multilingual checkpoint, complete 24 kHz WAV output and genuine incremental audio.

Created by
KittenML
License
Apache-2.0
Model released
Unknown
Source repository created
Nov 13, 2026
Model weights · FP32
327.1 MB

Best for

  • Local narration and accessibility drafts with review
  • Preset text-to-speech with selectable voice and speed

Keep in mind

  • Preset compact model only; no voice cloning, emotional tags or flagship capabilities.
  • Names, unfamiliar text, numbers and unsupported scripts can be mispronounced, omitted or rejected. Review output before use.
  • Controlled authored intelligibility/signal fixtures, not broad perceptual, voice-identity or real-world language certification. Voice choices identify upstream presets; every speaker is not independently benchmarked.
  • FP32 PyTorch and Q8 ONNX are precisions, not checkpoint variants. Q8 uses int8 stored weights with float32 arithmetic and CPU execution.
  • GPU neural execution is partial acceleration: preprocessing and WAV encoding remain on CPU.
  • Only current Apple M3 Max/macOS 27.2 is eligible for measured support. Windows, Intel Mac and other hardware/OS remain untested.
  • Uploads/public sample publication, full backup and forced source recovery excluded and pending.
  • Synthetic speech can mislead listeners. Disclose generated speech when appropriate; do not use unreviewed output for critical alerts or identity claims.

Processed on your deviceYour input is processed locally in the desktop app. Model and runtime files download from external hosts.

Target platforms · verification required

  • macos · aarch64, x86_64
  • windows · x86_64

Recommended configuration

FP32 · mps (verified platform only)

Sources & technical details
Original model
https://huggingface.co/KittenML/kitten-tts-2-mini
Release verification
Unknown
Release evidence
Exact original checkpoint release day not established by bounded official model-card/repository review; repository creation and artifact update are not substituted.
Source updated
Nov 13, 2026
Native model files
https://huggingface.co/KittenML/kitten-tts-2-mini
Files updated
Nov 13, 2026
Artifact revision
c01797caec3967b400da457656e011eb81f530a4

Desktop app

Sample result

KittenTTS 2 Mini · Multilingual Desktop

Run KittenTTS 2 on your computer

Download the desktop app to load this model, provide input and view results locally.

Public desktop downloads are coming soon.

Device requirements

macOS Apple SiliconCPU / Q8Tested8.0 GiBRAM estimate
macOS Apple SiliconCPU / FP32Tested8.0 GiBRAM estimate
macOS Apple SiliconMPS / FP32Tested8.0 GiBRAM estimate
Windows Intel / AMDCPU / Q8Unverified8.0 GiBRAM estimate
macOS Intel / AMDCPU / Q8Unverified8.0 GiBRAM estimate
If your device falls short

MemoryAdvisory only: close other apps, reduce image/tile/token/batch settings or choose a smaller verified model/precision. CPU can need more RAM than GPU.

GPUIf the selected GPU runtime fails, existing CPU fallback may recover when enabled, but CPU has its own memory/runtime needs.

PlatformUse a compatible OS/architecture/runtime or another model. These estimates do not add hardware detection or block runs.

Benchmark evidence

Saved desktop measurements apply to the recorded hardware and runtime. Fallback cases report the device that actually ran; they do not prove GPU support on the requested device.

KittenTTS 2 Mini · Multilingual Desktop

119 configurations tested
Passed119
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64162 ms / 19087 ms0.048 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64165 ms / 19087 ms0.049 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64166 ms / 19087 ms0.049 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64170 ms / 18854 ms0.051 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64170 ms / 19087 ms0.051 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64176 ms / 20234 ms0.052 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64177 ms / 18854 ms0.053 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64178 ms / 18854 ms0.053 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64178 ms / 18854 ms0.053 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64178 ms / 20234 ms0.054 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64259 ms / 20234 ms0.067 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64296 ms / 20234 ms0.068 RTF
FP32mpsAcceleratedPassedFastmacos · aarch643210 ms / 20234 ms0.140 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641055 ms / 20234 ms0.242 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641171 ms / 20234 ms0.248 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641122 ms / 20234 ms0.251 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641158 ms / 20234 ms0.272 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641143 ms / 20234 ms0.277 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641138 ms / 20234 ms0.286 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641161 ms / 20234 ms0.287 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641109 ms / 20234 ms0.296 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641149 ms / 20234 ms0.302 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641204 ms / 20234 ms0.313 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641084 ms / 20234 ms0.329 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641095 ms / 20234 ms0.337 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641109 ms / 20234 ms0.355 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641022 ms / 20234 ms0.356 RTF
FP32cpuPassedVery fastmacos · aarch642072 ms / 18496 ms0.091 RTF
FP32cpuPassedVery fastmacos · aarch64359 ms / 18496 ms0.093 RTF
FP32cpuPassedVery fastmacos · aarch64412 ms / 18496 ms0.095 RTF
FP32cpuPassedVery fastmacos · aarch64362 ms / 18496 ms0.095 RTF
FP32cpuPassedVery fastmacos · aarch64427 ms / 18496 ms0.098 RTF
FP32cpuPassedVery fastmacos · aarch64325 ms / 18496 ms0.098 RTF
FP32cpuPassedVery fastmacos · aarch64401 ms / 18496 ms0.099 RTF
FP32cpuPassedVery fastmacos · aarch64472 ms / 18496 ms0.100 RTF
FP32cpuPassedFastmacos · aarch64419 ms / 18496 ms0.101 RTF
FP32cpuPassedFastmacos · aarch64455 ms / 18496 ms0.102 RTF
FP32cpuPassedFastmacos · aarch64342 ms / 19787 ms0.102 RTF
FP32cpuPassedFastmacos · aarch64342 ms / 19787 ms0.102 RTF
FP32cpuPassedFastmacos · aarch64334 ms / 18496 ms0.102 RTF
FP32cpuPassedFastmacos · aarch64343 ms / 18740 ms0.102 RTF
FP32cpuPassedFastmacos · aarch64438 ms / 18496 ms0.103 RTF
FP32cpuPassedFastmacos · aarch64346 ms / 18740 ms0.103 RTF
FP32cpuPassedFastmacos · aarch64347 ms / 18496 ms0.104 RTF
FP32cpuPassedFastmacos · aarch64400 ms / 18496 ms0.104 RTF
FP32cpuPassedFastmacos · aarch64348 ms / 19787 ms0.104 RTF
FP32cpuPassedFastmacos · aarch64300 ms / 18496 ms0.104 RTF
FP32cpuPassedFastmacos · aarch64350 ms / 18740 ms0.105 RTF
FP32cpuPassedFastmacos · aarch64393 ms / 18496 ms0.105 RTF
FP32cpuPassedFastmacos · aarch64354 ms / 18740 ms0.106 RTF
FP32cpuPassedFastmacos · aarch64355 ms / 19787 ms0.106 RTF
FP32cpuPassedFastmacos · aarch64421 ms / 18496 ms0.106 RTF
FP32cpuPassedFastmacos · aarch64341 ms / 18496 ms0.109 RTF
FP32cpuPassedFastmacos · aarch64360 ms / 18496 ms0.111 RTF
FP32cpuPassedFastmacos · aarch64750 ms / 20508 ms0.172 RTF
FP32cpuPassedFastmacos · aarch643980 ms / 20508 ms0.174 RTF
FP32cpuPassedFastmacos · aarch64674 ms / 20508 ms0.175 RTF
FP32cpuPassedFastmacos · aarch64576 ms / 20508 ms0.176 RTF
FP32cpuPassedFastmacos · aarch64723 ms / 20508 ms0.178 RTF
FP32cpuPassedFastmacos · aarch64597 ms / 20508 ms0.181 RTF
FP32cpuPassedFastmacos · aarch64687 ms / 20508 ms0.181 RTF
FP32cpuPassedFastmacos · aarch64818 ms / 20508 ms0.183 RTF
FP32cpuPassedFastmacos · aarch64803 ms / 20508 ms0.185 RTF
FP32cpuPassedFastmacos · aarch64874 ms / 20508 ms0.185 RTF
FP32cpuPassedFastmacos · aarch64776 ms / 20508 ms0.188 RTF
FP32cpuPassedFastmacos · aarch64753 ms / 20508 ms0.189 RTF
FP32cpuPassedFastmacos · aarch64548 ms / 20508 ms0.190 RTF
FP32cpuPassedFastmacos · aarch64810 ms / 20508 ms0.190 RTF
FP32cpuPassedFastmacos · aarch64736 ms / 20508 ms0.191 RTF
FP32cpuPassedFastmacos · aarch64658 ms / 20508 ms0.196 RTF
Q8cpuPassedFastmacos · aarch64863 ms / 21075 ms0.198 RTF
Q8cpuPassedFastmacos · aarch64772 ms / 21075 ms0.199 RTF
FP32cpuPassedFastmacos · aarch64623 ms / 20508 ms0.199 RTF
Q8cpuPassedFastmacos · aarch644572 ms / 21075 ms0.200 RTF
Q8cpuPassedFastmacos · aarch64672 ms / 22398 ms0.201 RTF
Q8cpuPassedFastmacos · aarch64767 ms / 21075 ms0.202 RTF
FP32cpuPassedFastmacos · aarch64656 ms / 20508 ms0.202 RTF
Q8cpuPassedFastmacos · aarch64818 ms / 21075 ms0.202 RTF
Q8cpuPassedFastmacos · aarch64683 ms / 22398 ms0.204 RTF
Q8cpuPassedFastmacos · aarch64587 ms / 21075 ms0.204 RTF
FP32cpuPassedFastmacos · aarch64771 ms / 20508 ms0.206 RTF
Q8cpuPassedFastmacos · aarch64675 ms / 21075 ms0.206 RTF
Q8cpuPassedFastmacos · aarch64692 ms / 21075 ms0.210 RTF
Q8cpuPassedFastmacos · aarch64706 ms / 21075 ms0.211 RTF
Q8cpuPassedFastmacos · aarch64917 ms / 21075 ms0.211 RTF
Q8cpuPassedFastmacos · aarch64711 ms / 20297 ms0.212 RTF
Q8cpuPassedFastmacos · aarch64712 ms / 20297 ms0.213 RTF
Q8cpuPassedFastmacos · aarch64717 ms / 20297 ms0.214 RTF
Q8cpuPassedFastmacos · aarch641017 ms / 21075 ms0.215 RTF
Q8cpuPassedFastmacos · aarch64968 ms / 21075 ms0.216 RTF
Q8cpuPassedFastmacos · aarch64865 ms / 21075 ms0.218 RTF
Q8cpuPassedFastmacos · aarch64897 ms / 21075 ms0.219 RTF
Q8cpuPassedFastmacos · aarch64822 ms / 21075 ms0.219 RTF
Q8cpuPassedFastmacos · aarch64741 ms / 22398 ms0.221 RTF
Q8cpuPassedFastmacos · aarch64716 ms / 21075 ms0.229 RTF
Q8cpuPassedFastmacos · aarch64745 ms / 21075 ms0.229 RTF
Q8cpuPassedFastmacos · aarch64784 ms / 20297 ms0.234 RTF
Q8cpuPassedFastmacos · aarch641026 ms / 21075 ms0.241 RTF
Q8cpuPassedFastmacos · aarch64888 ms / 22398 ms0.265 RTF
Q8cpuPassedFastmacos · aarch641033 ms / 21075 ms0.267 RTF
Q8cpuPassedModeratemacos · aarch642118 ms / 21200 ms0.632 RTF
Q8cpuPassedModeratemacos · aarch641822 ms / 21200 ms0.634 RTF
Q8cpuPassedModeratemacos · aarch6414537 ms / 21200 ms0.635 RTF
Q8cpuPassedModeratemacos · aarch642785 ms / 21200 ms0.640 RTF
Q8cpuPassedModeratemacos · aarch642486 ms / 21200 ms0.642 RTF
Q8cpuPassedModeratemacos · aarch642630 ms / 21200 ms0.649 RTF
Q8cpuPassedModeratemacos · aarch642160 ms / 21200 ms0.655 RTF
Q8cpuPassedModeratemacos · aarch642186 ms / 21200 ms0.667 RTF
Q8cpuPassedModeratemacos · aarch643198 ms / 21200 ms0.677 RTF
Q8cpuPassedModeratemacos · aarch642950 ms / 21200 ms0.678 RTF
Q8cpuPassedModeratemacos · aarch642589 ms / 21200 ms0.681 RTF
Q8cpuPassedModeratemacos · aarch643140 ms / 21200 ms0.702 RTF
Q8cpuPassedModeratemacos · aarch642994 ms / 21200 ms0.704 RTF
Q8cpuPassedModeratemacos · aarch642894 ms / 21200 ms0.706 RTF
Q8cpuPassedModeratemacos · aarch642817 ms / 21200 ms0.709 RTF
Q8cpuPassedModeratemacos · aarch642754 ms / 21200 ms0.711 RTF
Q8cpuPassedModeratemacos · aarch642361 ms / 21200 ms0.727 RTF
Q8cpuPassedModeratemacos · aarch642318 ms / 21200 ms0.742 RTF
Q8cpuPassedModeratemacos · aarch642785 ms / 21200 ms0.743 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:00:28.988475+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
c01797caec3967b400da457656e011eb81f530a4
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 639.5821250043809,
      "inferenceMs": 657.5405829935335,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.1962807710428458,
      "audioSecondsPerSecond": 5.094742570487007,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 633.6397920385934,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1438109129667282,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.18914674120301853,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 633.6415830301121,
      "loadMs": 20508.122791012283,
      "totalMs": 21165.663374005817,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 536.0691249952652,
      "inferenceMs": 547.5420419825241,
      "audioSeconds": 2.875,
      "realTimeFactor": 0.1904494059069649,
      "audioSecondsPerSecond": 5.250738353515804,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 535.4964589932933,
      "voice": "en:martha",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1231195330619812,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.18626005807891488,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 535.4976669768803,
      "loadMs": 20508.122791012283,
      "totalMs": 21055.664832994808,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello. This is a local speech synthesis test. Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 3890.4489999986254,
      "inferenceMs": 3979.5481659821235,
      "audioSeconds": 22.875,
      "realTimeFactor": 0.17396931873145896,
      "audioSecondsPerSecond": 5.74814000130455,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 2964.2722500138916,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.18484967947006226,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.16993020764994818,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 3887.1534999925643,
      "loadMs": 20508.122791012283,
      "totalMs": 24487.670956994407,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 671.8369169975631,
      "inferenceMs": 687.3195829684846,
      "audioSeconds": 3.8,
      "realTimeFactor": 0.18087357446539068,
      "audioSecondsPerSecond": 5.528723601309407,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 671.2178330053575,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1951616406440735,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.1766368092227678,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 671.2198750465177,
      "loadMs": 20508.122791012283,
      "totalMs": 21195.442373980768,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 658.8299999712035,
      "inferenceMs": 673.587708035484,
      "audioSeconds": 3.85,
      "realTimeFactor": 0.17495784624298286,
      "audioSecondsPerSecond": 5.715662495131496,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 658.1283750128932,
      "voice": "de:lena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.10013073682785034,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.1709428680576868,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 658.1300420220941,
      "loadMs": 20508.122791012283,
      "totalMs": 21181.710499047767,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 584.6271670307033,
      "inferenceMs": 596.5055840206333,
      "audioSeconds": 3.3,
      "realTimeFactor": 0.18075926788504038,
      "audioSecondsPerSecond": 5.532219795424165,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 584.0654170024209,
      "voice": "es:lucia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.10276605188846588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.1769900254580672,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 584.0670840116218,
      "loadMs": 20508.122791012283,
      "totalMs": 21104.628375032917,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 735.2314169984311,
      "inferenceMs": 750.2254170249216,
      "audioSeconds": 4.35,
      "realTimeFactor": 0.17246561310917738,
      "audioSecondsPerSecond": 5.798257298786637,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 734.5598330139183,
      "voice": "it:giulia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.17710155248641968,
      "clippedFraction": 0.0002681992337164751,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.16886465517729093,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 734.5612500212155,
      "loadMs": 20508.122791012283,
      "totalMs": 21258.348208037205,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 563.4530840325169,
      "inferenceMs": 575.5778339807875,
      "audioSeconds": 3.275,
      "realTimeFactor": 0.17574895694069848,
      "audioSecondsPerSecond": 5.689934195953275,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 562.9113340401091,
      "voice": "pt:beatriz",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1280466467142105,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.1718819340449253,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 562.9133339971304,
      "loadMs": 20508.122791012283,
      "totalMs": 21083.70062499307,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте! Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 708.7703750003129,
      "inferenceMs": 722.6850829902105,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.17844076123215075,
      "audioSecondsPerSecond": 5.6041007284148705,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 708.1125839613378,
      "voice": "ru:anastasia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.02534087561070919,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.17484307629746143,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 708.1144590047188,
      "loadMs": 20508.122791012283,
      "totalMs": 21230.807874002494,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 857.170709001366,
      "inferenceMs": 873.8691660109907,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.18494585524042131,
      "audioSecondsPerSecond": 5.406987892213344,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 856.3760840333998,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11959778517484665,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.18124400359632634,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 856.377916992642,
      "loadMs": 20508.122791012283,
      "totalMs": 21381.991957023274,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Godmorgen, dette er en lokal test- og talesyntese.",
      "errorMetric": "wer",
      "errorRate": 0.125,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 611.3794579869136,
      "inferenceMs": 622.8500830475241,
      "audioSeconds": 3.125,
      "realTimeFactor": 0.19931202657520772,
      "audioSecondsPerSecond": 5.017258703265774,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 610.7354169944301,
      "voice": "da:freja",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.04724461957812309,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.19543587999418377,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 610.7371249818243,
      "loadMs": 20508.122791012283,
      "totalMs": 21130.972874059808,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hyvää huomenta, tämä on paikallinen puhesynteesin testi.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 794.6980419801548,
      "inferenceMs": 809.5676670200191,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.19048650988706334,
      "audioSecondsPerSecond": 5.249715586646452,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 794.0379169886,
      "voice": "fi:aino",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11310887336730957,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.18683279411155074,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 794.0393749740906,
      "loadMs": 20508.122791012283,
      "totalMs": 21317.690458032303,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 758.1567079760134,
      "inferenceMs": 771.2723329896107,
      "audioSeconds": 3.75,
      "realTimeFactor": 0.20567262213056287,
      "audioSecondsPerSecond": 4.862095837749328,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 757.5527919689193,
      "voice": "fr:camille",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.03975072130560875,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.2020144887889425,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 757.5543329585344,
      "loadMs": 20508.122791012283,
      "totalMs": 21279.395124001894,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "こんにちは、これはこの端末で音声を作るためのテストです。",
      "errorMetric": "cer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 787.9014579812065,
      "inferenceMs": 803.4392079571262,
      "audioSeconds": 4.35,
      "realTimeFactor": 0.1846986684958911,
      "audioSecondsPerSecond": 5.414224196327905,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 787.2014580061659,
      "voice": "ja:yuki",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.058920711278915405,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.18096624528346905,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 787.2031669830903,
      "loadMs": 20508.122791012283,
      "totalMs": 21311.56199896941,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " 안녕하세요 이 기기에서 음성을 만드는 시험입니다.",
      "errorMetric": "cer",
      "errorRate": 0,
      "maximumRepeatedWords": 2,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 644.9542090413161,
      "inferenceMs": 656.1897909850813,
      "audioSeconds": 3.25,
      "realTimeFactor": 0.2019045510723327,
      "audioSecondsPerSecond": 4.952835360515216,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 644.4218750111759,
      "voice": "ko:seoyeon",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.01888292469084263,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.19828417968864626,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 644.4235839881003,
      "loadMs": 20508.122791012283,
      "totalMs": 21164.312581997365,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Goedemorgen, dit is een lokale test voor spraaksynthese.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 739.3482500337996,
      "inferenceMs": 752.9035409679636,
      "audioSeconds": 3.975,
      "realTimeFactor": 0.18940969584099715,
      "audioSecondsPerSecond": 5.279560772007497,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 738.7327920296229,
      "voice": "nl:sanne",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.0826149433851242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.18584507346188403,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 738.734167010989,
      "loadMs": 20508.122791012283,
      "totalMs": 21261.026331980247,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " God morgon, det här är ett lokalt test av talsyntes.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 762.2310830047354,
      "inferenceMs": 776.0610000113957,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.1881360000027626,
      "audioSecondsPerSecond": 5.31530382268846,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 761.5897910436615,
      "voice": "sv:astrid",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.12222650647163391,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.18462816146061276,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 761.5911660250276,
      "loadMs": 20508.122791012283,
      "totalMs": 21284.18379102368,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Habari za asubuhi hili ni jaribio la sauti kwenye kifahiki.",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 802.3986669722944,
      "inferenceMs": 817.7563750068657,
      "audioSeconds": 4.475,
      "realTimeFactor": 0.1827388547501376,
      "audioSecondsPerSecond": 5.472289959173267,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 801.706500002183,
      "voice": "sw:zawadi",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.10982336848974228,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.17915265362803412,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 801.7081249854527,
      "loadMs": 20508.122791012283,
      "totalMs": 21325.87916601915,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Günaydın bu cihazda yerel bir konuşma testi yapıyoruz.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 722.3745000083,
      "inferenceMs": 735.7268750201911,
      "audioSeconds": 3.85,
      "realTimeFactor": 0.19109788961563404,
      "audioSecondsPerSecond": 5.2329201646933745,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 721.7610829975456,
      "voice": "tr:elif",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.01869593933224678,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.18747079013786339,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 627.2507909452543,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 721.7625420307741,
      "loadMs": 20508.122791012283,
      "totalMs": 21243.849666032474,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 333.0321250250563,
      "inferenceMs": 346.842041995842,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.10353493790920656,
      "audioSecondsPerSecond": 9.658575358174602,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 328.3678340376355,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1438109129667282,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0980210946282066,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 328.37066700449213,
      "loadMs": 18495.520999946166,
      "totalMs": 18842.363041942008,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 287.69445803482085,
      "inferenceMs": 299.8005829867907,
      "audioSeconds": 2.875,
      "realTimeFactor": 0.10427846364757937,
      "audioSecondsPerSecond": 9.589707836314226,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 287.18966699671,
      "voice": "en:martha",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1231195330619812,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09989272452809889,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 287.1915830182843,
      "loadMs": 18495.520999946166,
      "totalMs": 18795.321582932957,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello. This is a local speech synthesis test. Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1985.3584170341492,
      "inferenceMs": 2072.3377090180293,
      "audioSeconds": 22.875,
      "realTimeFactor": 0.09059399820843844,
      "audioSecondsPerSecond": 11.038258822612097,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1491.6066669975407,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.18484967947006226,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.08665340800153834,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1982.1967080351897,
      "loadMs": 18495.520999946166,
      "totalMs": 20567.858708964195,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 346.98966599535197,
      "inferenceMs": 362.32945800293237,
      "audioSeconds": 3.8,
      "realTimeFactor": 0.09534985736919273,
      "audioSecondsPerSecond": 10.487692667730169,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 346.3846659869887,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1951616406440735,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09115415553280495,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 346.38579102465883,
      "loadMs": 18495.520999946166,
      "totalMs": 18857.8504579491,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 344.9927499750629,
      "inferenceMs": 358.91483403975144,
      "audioSeconds": 3.85,
      "realTimeFactor": 0.09322463221811726,
      "audioSecondsPerSecond": 10.726778708660436,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 344.38691596733406,
      "voice": "de:lena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.10013073682785034,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.08945157999390518,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 344.38858297653496,
      "loadMs": 18495.520999946166,
      "totalMs": 18854.435833985917,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 312.9183749551885,
      "inferenceMs": 324.7330830199644,
      "audioSeconds": 3.3,
      "realTimeFactor": 0.09840396455150437,
      "audioSecondsPerSecond": 10.162192189691735,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 312.37349996808916,
      "voice": "es:lucia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.10276605188846588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0946589899919645,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 312.3746669734828,
      "loadMs": 18495.520999946166,
      "totalMs": 18820.25408296613,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 395.69554198533297,
      "inferenceMs": 411.7299999925308,
      "audioSeconds": 4.35,
      "realTimeFactor": 0.09465057471092662,
      "audioSecondsPerSecond": 10.565176207900599,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 394.8635419947095,
      "voice": "it:giulia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.17710155248641968,
      "clippedFraction": 0.0002681992337164751,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09077364000780828,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 394.86533403396606,
      "loadMs": 18495.520999946166,
      "totalMs": 18907.250999938697,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 322.21274997573346,
      "inferenceMs": 334.10966699011624,
      "audioSeconds": 3.275,
      "realTimeFactor": 0.10201821892827977,
      "audioSecondsPerSecond": 9.802170734847017,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 321.61074999021366,
      "voice": "pt:beatriz",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1280466467142105,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0982023282367594,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 321.612624975387,
      "loadMs": 18495.520999946166,
      "totalMs": 18829.630666936282,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте! Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 386.4195829955861,
      "inferenceMs": 400.7188330288045,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.09894292173550728,
      "audioSecondsPerSecond": 10.106837179047378,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 385.72262501111254,
      "voice": "ru:anastasia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.02534087561070919,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09524082296447438,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 385.7253330061212,
      "loadMs": 18495.520999946166,
      "totalMs": 18896.23983297497,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 455.9892080142163,
      "inferenceMs": 472.12508297525346,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.09992065248153513,
      "audioSecondsPerSecond": 10.007941052874884,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 455.21383301820606,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11959778517484665,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09634184296287242,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 455.21520799957216,
      "loadMs": 18495.520999946166,
      "totalMs": 18967.64608292142,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Godmorgen, dette er en lokal test- og talesyntese.",
      "errorMetric": "wer",
      "errorRate": 0.125,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 329.93800000986084,
      "inferenceMs": 341.30645799450576,
      "audioSeconds": 3.125,
      "realTimeFactor": 0.10921806655824184,
      "audioSecondsPerSecond": 9.155994347022597,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 329.34629201190546,
      "voice": "da:freja",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.04724461957812309,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10539133343845605,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 329.3479169951752,
      "loadMs": 18495.520999946166,
      "totalMs": 18836.82745794067,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hyvää huomenta, tämä on paikallinen puhesynteesin testi.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 423.6235420103185,
      "inferenceMs": 438.4701249655336,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.10316944116836085,
      "audioSecondsPerSecond": 9.692792639713083,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 422.96233295928687,
      "voice": "fi:aino",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11310887336730957,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09952096069943817,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 422.96408297261223,
      "loadMs": 18495.520999946166,
      "totalMs": 18933.9911249117,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 379.58262499887496,
      "inferenceMs": 392.54450000589713,
      "audioSeconds": 3.75,
      "realTimeFactor": 0.1046785333349059,
      "audioSecondsPerSecond": 9.553057041796954,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 378.98762500844896,
      "voice": "fr:camille",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.03975072130560875,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.1010637778788805,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 378.9891670458019,
      "loadMs": 18495.520999946166,
      "totalMs": 18888.065499952063,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "こんにちは、これはこの端末で音声を作るためのテストです。",
      "errorMetric": "cer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 411.96604195283726,
      "inferenceMs": 427.2225830354728,
      "audioSeconds": 4.35,
      "realTimeFactor": 0.09821208805413167,
      "audioSecondsPerSecond": 10.18204601707306,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 411.23937495285645,
      "voice": "ja:yuki",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.058920711278915405,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09453823769882579,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 411.24133398989215,
      "loadMs": 18495.520999946166,
      "totalMs": 18922.74358298164,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " 안녕하세요 이 기기에서 음성을 만드는 시험입니다.",
      "errorMetric": "cer",
      "errorRate": 0,
      "maximumRepeatedWords": 2,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 348.3195410226472,
      "inferenceMs": 359.5802499912679,
      "audioSeconds": 3.25,
      "realTimeFactor": 0.11064007692039013,
      "audioSecondsPerSecond": 9.038316203626097,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 347.77508303523064,
      "voice": "ko:seoyeon",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.01888292469084263,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10700825632263261,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 347.776833048556,
      "loadMs": 18495.520999946166,
      "totalMs": 18855.101249937434,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Goedemorgen, dit is een lokale test voor spraaksynthese.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 407.3639159905724,
      "inferenceMs": 421.31504201097414,
      "audioSeconds": 3.975,
      "realTimeFactor": 0.10599120553735199,
      "audioSecondsPerSecond": 9.434745033139505,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 406.62629099097103,
      "voice": "nl:sanne",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.0826149433851242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10229638364728329,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 406.6281249979511,
      "loadMs": 18495.520999946166,
      "totalMs": 18916.83604195714,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " God morgon, det här är ett lokalt test av talsyntes.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 404.3862909893505,
      "inferenceMs": 418.5368330217898,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.10146347467194904,
      "audioSecondsPerSecond": 9.855763398929444,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 403.6997910006903,
      "voice": "sv:astrid",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.12222650647163391,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0978669796392999,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 403.70129101211205,
      "loadMs": 18495.520999946166,
      "totalMs": 18914.057832967956,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Habari za asubuhi hili ni jaribio la sauti kwenye kifahiki.",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 439.8255000123754,
      "inferenceMs": 455.2871249616146,
      "audioSeconds": 4.475,
      "realTimeFactor": 0.10174013965622673,
      "audioSecondsPerSecond": 9.828962328722318,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 438.9846249832772,
      "voice": "sw:zawadi",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.10982336848974228,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09809754190539847,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 438.9865000266582,
      "loadMs": 18495.520999946166,
      "totalMs": 18950.80812490778,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Günaydın bu cihazda yerel bir konuşma testi yapıyoruz.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 386.42837502993643,
      "inferenceMs": 399.5834169909358,
      "audioSeconds": 3.85,
      "realTimeFactor": 0.10378790051712618,
      "audioSecondsPerSecond": 9.635034479139392,
      "placement": {
        "cpu/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 385.80091699259356,
      "voice": "tr:elif",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.01869593933224678,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10020840908521665,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 347.4934999831021,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 385.8023749780841,
      "loadMs": 18495.520999946166,
      "totalMs": 18895.1044169371,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 163.04470802424476,
      "inferenceMs": 175.65595800988376,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.05243461433130858,
      "audioSecondsPerSecond": 19.071371321270544,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 158.95541600184515,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1413072645664215,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1356316672,
      "synthesisRealTimeFactor": 0.04744996269817339,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 158.95737503888085,
      "loadMs": 20234.01700000977,
      "totalMs": 20409.672958019655,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1011.0926670022309,
      "inferenceMs": 1022.4228750448674,
      "audioSeconds": 2.875,
      "realTimeFactor": 0.35562534784169303,
      "audioSecondsPerSecond": 2.811948040456191,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1010.5805000057444,
      "voice": "en:martha",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.12035790085792542,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1363214336,
      "synthesisRealTimeFactor": 0.3515071161901173,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1010.5829590465873,
      "loadMs": 20234.01700000977,
      "totalMs": 21256.43987505464,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello. This is a local speech synthesis test. Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 3118.6217080103233,
      "inferenceMs": 3209.9349580239505,
      "audioSeconds": 22.875,
      "realTimeFactor": 0.14032502548738582,
      "audioSecondsPerSecond": 7.12631261976783,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1891.6011250112206,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.18116290867328644,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1498939392,
      "synthesisRealTimeFactor": 0.13618756646972374,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 3115.2905829949304,
      "loadMs": 20234.01700000977,
      "totalMs": 23443.951958033722,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1133.2867079763673,
      "inferenceMs": 1149.2303329869173,
      "audioSeconds": 3.8,
      "realTimeFactor": 0.3024290349965572,
      "audioSecondsPerSecond": 3.3065608267783677,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1132.6384160201997,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.19134403765201569,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1511604224,
      "synthesisRealTimeFactor": 0.2980632015724519,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1132.6401659753174,
      "loadMs": 20234.01700000977,
      "totalMs": 21383.24733299669,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 245.28079200536013,
      "inferenceMs": 259.29254200309515,
      "audioSeconds": 3.85,
      "realTimeFactor": 0.06734871220859615,
      "audioSecondsPerSecond": 14.848093856683478,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 244.62533299811184,
      "voice": "de:lena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.09828025102615356,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1515782144,
      "synthesisRealTimeFactor": 0.06353964285271896,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 244.62762498296797,
      "loadMs": 20234.01700000977,
      "totalMs": 20493.309542012867,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1071.9440410030074,
      "inferenceMs": 1084.092083037831,
      "audioSeconds": 3.3,
      "realTimeFactor": 0.3285127524357063,
      "audioSecondsPerSecond": 3.044021860903897,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1071.35633303551,
      "voice": "es:lucia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.098124660551548,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1527726080,
      "synthesisRealTimeFactor": 0.3246538509142049,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1071.3577080168761,
      "loadMs": 20234.01700000977,
      "totalMs": 21318.109083047602,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1038.2649580133148,
      "inferenceMs": 1054.5830000191927,
      "audioSeconds": 4.35,
      "realTimeFactor": 0.2424328735676305,
      "audioSecondsPerSecond": 4.1248531409294795,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1037.5707080238499,
      "voice": "it:giulia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1731383055448532,
      "clippedFraction": 0.0002777777777777778,
      "driverAllocatedMemoryBytes": 1536409600,
      "synthesisRealTimeFactor": 0.2385223563225276,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1037.5722500029951,
      "loadMs": 20234.01700000977,
      "totalMs": 21288.600000028964,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 165.583209018223,
      "inferenceMs": 177.7153750299476,
      "audioSeconds": 3.275,
      "realTimeFactor": 0.054264236650365685,
      "audioSecondsPerSecond": 18.42834363345386,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 165.01412499928847,
      "voice": "pt:beatriz",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.12560097873210907,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1473495040,
      "synthesisRealTimeFactor": 0.0503865905454425,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 165.01608403632417,
      "loadMs": 20234.01700000977,
      "totalMs": 20411.73237503972,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте! Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1146.8575419858098,
      "inferenceMs": 1161.3861250225455,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.2867620061784063,
      "audioSecondsPerSecond": 3.4872123170245204,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1146.1163749918342,
      "voice": "ru:anastasia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.024497410282492638,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1486897152,
      "synthesisRealTimeFactor": 0.2829920987592849,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1146.1179999751039,
      "loadMs": 20234.01700000977,
      "totalMs": 21395.403125032317,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1154.009332996793,
      "inferenceMs": 1170.914750022348,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.24781264550737525,
      "audioSecondsPerSecond": 4.035306583942015,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1153.112665982917,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11513970047235489,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1501642752,
      "synthesisRealTimeFactor": 0.24404539682325863,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1153.114499989897,
      "loadMs": 20234.01700000977,
      "totalMs": 21404.93175003212,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Godmorgen. Dette er en lokal test- og talesyntese.",
      "errorMetric": "wer",
      "errorRate": 0.125,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1097.0660420134664,
      "inferenceMs": 1108.9130839682184,
      "audioSeconds": 3.125,
      "realTimeFactor": 0.3548521868698299,
      "audioSecondsPerSecond": 2.818074784380092,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1096.534584008623,
      "voice": "da:freja",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.044105399399995804,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1517830144,
      "synthesisRealTimeFactor": 0.3508916668780148,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1096.5364589937963,
      "loadMs": 20234.01700000977,
      "totalMs": 21342.93008397799,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hyvää huomenta, tämä on paikallinen puhesynteesin testi.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1143.0912080104463,
      "inferenceMs": 1157.681333948858,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.2723956079879666,
      "audioSecondsPerSecond": 3.6711311440975076,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1142.426708014682,
      "voice": "fi:aino",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11067885905504227,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1531510784,
      "synthesisRealTimeFactor": 0.2688067058813484,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1142.4284999957308,
      "loadMs": 20234.01700000977,
      "totalMs": 21391.69833395863,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1096.549625042826,
      "inferenceMs": 1109.4018750009127,
      "audioSeconds": 3.75,
      "realTimeFactor": 0.29584050000024337,
      "audioSecondsPerSecond": 3.3801998036076104,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1095.8604580373503,
      "voice": "fr:camille",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.038958590477705,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1544224768,
      "synthesisRealTimeFactor": 0.2922299888140211,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1095.8624580525793,
      "loadMs": 20234.01700000977,
      "totalMs": 21343.418875010684,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "こんにちは、これはこの端末で音声を作るためのテストです。",
      "errorMetric": "cer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 280.3138749441132,
      "inferenceMs": 296.4123749989085,
      "audioSeconds": 4.35,
      "realTimeFactor": 0.06814077586181805,
      "audioSecondsPerSecond": 14.675500643372324,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 279.53749999869615,
      "voice": "ja:yuki",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.05764712020754814,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1549205504,
      "synthesisRealTimeFactor": 0.0642618868933541,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 279.53920798609033,
      "loadMs": 20234.01700000977,
      "totalMs": 20530.42937500868,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " 안녕하세요 이 기기에서 음성을 만드는 시험입니다.",
      "errorMetric": "cer",
      "errorRate": 0,
      "maximumRepeatedWords": 2,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1083.6766670108773,
      "inferenceMs": 1095.2037919778377,
      "audioSeconds": 3.25,
      "realTimeFactor": 0.33698578214702696,
      "audioSecondsPerSecond": 2.9674842470466594,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1083.1368330400437,
      "voice": "ko:seoyeon",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.01841161772608757,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1561837568,
      "synthesisRealTimeFactor": 0.33327348707602,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1083.138832997065,
      "loadMs": 20234.01700000977,
      "totalMs": 21329.22079198761,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Goedemorgen, dit is een lokale test voor spraaksynthese.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1124.4621659861878,
      "inferenceMs": 1138.4727090480737,
      "audioSeconds": 3.975,
      "realTimeFactor": 0.2864082286913393,
      "audioSecondsPerSecond": 3.491519795255935,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1123.829875024967,
      "voice": "nl:sanne",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.07924826443195343,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1575190528,
      "synthesisRealTimeFactor": 0.2827249579834488,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1123.831707984209,
      "loadMs": 20234.01700000977,
      "totalMs": 21372.489709057845,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " God morgon, det här är ett lokalt test av talsyntes.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1128.2557089580223,
      "inferenceMs": 1142.8291669581085,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.27704949502014753,
      "audioSecondsPerSecond": 3.609463355734607,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1127.59670894593,
      "voice": "sv:astrid",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.12082409858703613,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1588903936,
      "synthesisRealTimeFactor": 0.27335731321099127,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1127.598916995339,
      "loadMs": 20234.01700000977,
      "totalMs": 21376.84616696788,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Habari za asubuhi hili ni jaribio la sauti kwenye kifahiki.",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1107.069791003596,
      "inferenceMs": 1122.3962499643676,
      "audioSeconds": 4.475,
      "realTimeFactor": 0.2508148044613112,
      "audioSecondsPerSecond": 3.987005480588577,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1106.3670830335468,
      "voice": "sw:zawadi",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.10597290098667145,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1603190784,
      "synthesisRealTimeFactor": 0.24723345430713542,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1106.369708024431,
      "loadMs": 20234.01700000977,
      "totalMs": 21356.41324997414,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Günaydın bu cihazda yerel bir konuşma testi yapıyoruz.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1182.0542910136282,
      "inferenceMs": 1203.927707974799,
      "audioSeconds": 3.85,
      "realTimeFactor": 0.31270849557786984,
      "audioSecondsPerSecond": 3.1978664287711447,
      "placement": {
        "mps:0/torch.float32": 548
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1181.415208033286,
      "voice": "tr:elif",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.018103504553437233,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1611743232,
      "synthesisRealTimeFactor": 0.30686158000479463,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1878.9680839981884,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1181.4170830184594,
      "loadMs": 20234.01700000977,
      "totalMs": 21437.94470798457,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2105.778041994199,
      "inferenceMs": 2117.943749995902,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.6322220149241499,
      "audioSecondsPerSecond": 1.5817228384873212,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2100.824749970343,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1435573697090149,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.6271125993920741,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2100.827207963448,
      "loadMs": 21200.03170799464,
      "totalMs": 23317.975457990542,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1811.5656250156462,
      "inferenceMs": 1822.0594170270488,
      "audioSeconds": 2.875,
      "realTimeFactor": 0.6337597972267995,
      "audioSecondsPerSecond": 1.5778848774816436,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1811.0447500366718,
      "voice": "en:martha",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.12337637692689896,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.6299292900923478,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1811.0467090154998,
      "loadMs": 21200.03170799464,
      "totalMs": 23022.09112502169,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello. This is a local speech synthesis test. Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 14438.076207996346,
      "inferenceMs": 14537.20008401433,
      "audioSeconds": 22.9,
      "realTimeFactor": 0.6348122307429839,
      "audioSecondsPerSecond": 1.5752689560338191,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 11120.24229200324,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.184909388422966,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.6303225036260091,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 14434.385333035607,
      "loadMs": 21200.03170799464,
      "totalMs": 35737.23179200897,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2573.439541971311,
      "inferenceMs": 2589.2879590392113,
      "audioSeconds": 3.8,
      "realTimeFactor": 0.6813915681682134,
      "audioSecondsPerSecond": 1.4675849345895229,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2572.828958975151,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.195246621966362,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.6770608223680603,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2572.831124998629,
      "loadMs": 21200.03170799464,
      "totalMs": 23789.31966703385,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2471.9176669605076,
      "inferenceMs": 2486.0702080186456,
      "audioSeconds": 3.875,
      "realTimeFactor": 0.6415665052951344,
      "audioSecondsPerSecond": 1.5586848623588578,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2471.272249997128,
      "voice": "de:lena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1001310721039772,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.6377481721161354,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2471.2741669500247,
      "loadMs": 21200.03170799464,
      "totalMs": 23686.101916013286,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2149.3863340001553,
      "inferenceMs": 2160.323124960996,
      "audioSeconds": 3.3,
      "realTimeFactor": 0.6546433712003019,
      "audioSecondsPerSecond": 1.5275492642146207,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2148.794667038601,
      "voice": "es:lucia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.10254645347595215,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.6511504800150623,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2148.7965840497054,
      "loadMs": 21200.03170799464,
      "totalMs": 23360.354832955636,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2767.4697089823894,
      "inferenceMs": 2784.94437498739,
      "audioSeconds": 4.35,
      "realTimeFactor": 0.6402170976982505,
      "audioSecondsPerSecond": 1.561970156053726,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2766.6440419852734,
      "voice": "it:giulia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.17735247313976288,
      "clippedFraction": 0.00029693486590038313,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.636010670568794,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2766.646416974254,
      "loadMs": 21200.03170799464,
      "totalMs": 23984.97608298203,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2174.8546669841744,
      "inferenceMs": 2185.732416983228,
      "audioSeconds": 3.275,
      "realTimeFactor": 0.6673992112925887,
      "audioSecondsPerSecond": 1.4983535837017923,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2174.2893339833245,
      "voice": "pt:beatriz",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.127797931432724,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.6639057633694517,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2174.2913750349544,
      "loadMs": 21200.03170799464,
      "totalMs": 23385.764124977868,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте! Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2616.2985409609973,
      "inferenceMs": 2629.884416994173,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.6493541770355983,
      "audioSecondsPerSecond": 1.5399916337878257,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2615.644249948673,
      "voice": "ru:anastasia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.02528490498661995,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.645838600732846,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2615.6463329680264,
      "loadMs": 21200.03170799464,
      "totalMs": 23829.916124988813,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 3181.192416988779,
      "inferenceMs": 3197.562667017337,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.6767328395803888,
      "audioSecondsPerSecond": 1.477688005535618,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 3180.337583995424,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11977250128984451,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.6730877337533803,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 3180.339541984722,
      "loadMs": 21200.03170799464,
      "totalMs": 24397.594375011977,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Godmorgen, dette er en lokal test- og talesyntese.",
      "errorMetric": "wer",
      "errorRate": 0.125,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2307.7302500023507,
      "inferenceMs": 2318.2503329589963,
      "audioSeconds": 3.125,
      "realTimeFactor": 0.7418401065468788,
      "audioSecondsPerSecond": 1.3479993750335302,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2307.093541021459,
      "voice": "da:freja",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.04741906374692917,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.7382704799994826,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2307.095249998383,
      "loadMs": 21200.03170799464,
      "totalMs": 23518.282040953636,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hyvää huomenta, tämä on paikallinen puhesynteesin testi.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2979.339666024316,
      "inferenceMs": 2993.6598329804838,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.7043905489365844,
      "audioSecondsPerSecond": 1.4196669752450484,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2978.566750010941,
      "voice": "fi:aino",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11310196667909622,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.700839744949275,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2978.568916034419,
      "loadMs": 21200.03170799464,
      "totalMs": 24193.691540975124,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2773.0608329875395,
      "inferenceMs": 2785.43820802588,
      "audioSeconds": 3.75,
      "realTimeFactor": 0.7427835221402347,
      "audioSecondsPerSecond": 1.346287269699561,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2772.4616670166142,
      "voice": "fr:camille",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.039808887988328934,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.7393236887951692,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2772.4638329818845,
      "loadMs": 21200.03170799464,
      "totalMs": 23985.46991602052,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "こんにちは、これはこの端末で音声を作るためのテストです。",
      "errorMetric": "cer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2935.751207987778,
      "inferenceMs": 2950.2360000042245,
      "audioSeconds": 4.35,
      "realTimeFactor": 0.6782151724147643,
      "audioSecondsPerSecond": 1.4744583145191676,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2934.9650829681195,
      "voice": "ja:yuki",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.05849319323897362,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.6747049903470338,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2934.966708009597,
      "loadMs": 21200.03170799464,
      "totalMs": 24150.267707998864,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " 안녕하세요 이 기기에서 음성을 만드는 시험입니다.",
      "errorMetric": "cer",
      "errorRate": 0,
      "maximumRepeatedWords": 2,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2350.520542007871,
      "inferenceMs": 2361.3657919922844,
      "audioSeconds": 3.25,
      "realTimeFactor": 0.7265740898437798,
      "audioSecondsPerSecond": 1.3763221314635776,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2349.9610840226524,
      "voice": "ko:seoyeon",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.018892094492912292,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.7230656538516856,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2349.9633750179783,
      "loadMs": 21200.03170799464,
      "totalMs": 23561.397499986924,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Goedemorgen, dit is een lokale test voor spraaksynthese.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2804.1425829869695,
      "inferenceMs": 2817.457499972079,
      "audioSeconds": 3.975,
      "realTimeFactor": 0.7087943396156173,
      "audioSecondsPerSecond": 1.410846481282998,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2803.4160409588367,
      "voice": "nl:sanne",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.082499660551548,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.7052623688055893,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2803.4179160022177,
      "loadMs": 21200.03170799464,
      "totalMs": 24017.48920796672,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " God morgon, det här är ett lokalt test av talsyntes.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2880.7039170060307,
      "inferenceMs": 2894.3566249799915,
      "audioSeconds": 4.1,
      "realTimeFactor": 0.7059406402390223,
      "audioSecondsPerSecond": 1.4165496969566918,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2879.893292032648,
      "voice": "sv:astrid",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.12287326902151108,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.7024135060960472,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2879.8953749937937,
      "loadMs": 21200.03170799464,
      "totalMs": 24094.38833297463,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Habari za asubuhi hili ni jaribio la sauti kwenye kifahiki.",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 3125.132792047225,
      "inferenceMs": 3139.724957989529,
      "audioSeconds": 4.475,
      "realTimeFactor": 0.70161451575185,
      "audioSecondsPerSecond": 1.425284080572934,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 3124.368792050518,
      "voice": "sw:zawadi",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11006675660610199,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.6981834545263281,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 3124.3709590053186,
      "loadMs": 21200.03170799464,
      "totalMs": 24339.75666598417,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Günaydın bu cihazda yerel bir konuşma testi yapıyoruz.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2741.2192500196397,
      "inferenceMs": 2754.14550001733,
      "audioSeconds": 3.875,
      "realTimeFactor": 0.7107472258109239,
      "audioSecondsPerSecond": 1.4069699658117616,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 2740.458791027777,
      "voice": "tr:elif",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.018762262538075447,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.7072156557452774,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 2188.062915985938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2740.4606660129502,
      "loadMs": 21200.03170799464,
      "totalMs": 23954.17720801197,
      "grade": "acceptable",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 693.5138340340927,
      "inferenceMs": 706.0924160177819,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.2107738555276961,
      "audioSecondsPerSecond": 4.7444214439992445,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 689.5466250134632,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.14355479180812836,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.20583563433862778,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 689.549375034403,
      "loadMs": 21075.31841704622,
      "totalMs": 21781.410833064,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 575.7782919681631,
      "inferenceMs": 586.841624986846,
      "audioSeconds": 2.875,
      "realTimeFactor": 0.20411882608238122,
      "audioSecondsPerSecond": 4.899107148482255,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 575.2082499675453,
      "voice": "en:martha",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.12337637692689896,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.2000731596439753,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 575.210333976429,
      "loadMs": 21075.31841704622,
      "totalMs": 21662.160042033065,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. Hello. This is a local speech synthesis test. Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 4485.125125036575,
      "inferenceMs": 4572.255084000062,
      "audioSeconds": 22.9,
      "realTimeFactor": 0.1996617940611381,
      "audioSecondsPerSecond": 5.008469470597824,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 3425.1640000147745,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.184909388422966,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.19569344065451028,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 4481.379790988285,
      "loadMs": 21075.31841704622,
      "totalMs": 25647.57350104628,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 751.449000032153,
      "inferenceMs": 766.8214169680141,
      "audioSeconds": 3.8,
      "realTimeFactor": 0.20179510972842477,
      "audioSecondsPerSecond": 4.955521475945561,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 750.815125007648,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.195246621966362,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.19758355263069174,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 750.8174999966286,
      "loadMs": 21075.31841704622,
      "totalMs": 21842.139834014233,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 759.3216249952093,
      "inferenceMs": 771.8342080479488,
      "audioSeconds": 3.875,
      "realTimeFactor": 0.19918302143172872,
      "audioSecondsPerSecond": 5.020508238162039,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 758.7054999894463,
      "voice": "de:lena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1001310721039772,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.19579556980921375,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 758.7078330107033,
      "loadMs": 21075.31841704622,
      "totalMs": 21847.152625094168,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 680.527999997139,
      "inferenceMs": 691.9834589934908,
      "audioSeconds": 3.3,
      "realTimeFactor": 0.2096919572707548,
      "audioSecondsPerSecond": 4.768900119086577,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 679.8948330106214,
      "voice": "es:lucia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.10254491865634918,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.20602983576033942,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 679.8984580091201,
      "loadMs": 21075.31841704622,
      "totalMs": 21767.30187603971,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 847.97395800706,
      "inferenceMs": 862.6122500281781,
      "audioSeconds": 4.35,
      "realTimeFactor": 0.1983016666731444,
      "audioSecondsPerSecond": 5.042821963005862,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 846.991500002332,
      "voice": "it:giulia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1773497611284256,
      "clippedFraction": 0.00029693486590038313,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.1947117144852106,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 846.9959580106661,
      "loadMs": 21075.31841704622,
      "totalMs": 21937.930667074397,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 664.0800420427695,
      "inferenceMs": 675.458499987144,
      "audioSeconds": 3.275,
      "realTimeFactor": 0.20624687022508215,
      "audioSecondsPerSecond": 4.848558423740812,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 663.5510839987546,
      "voice": "pt:beatriz",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1277921497821808,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.20261164123675865,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 663.5531250503846,
      "loadMs": 21075.31841704622,
      "totalMs": 21750.776917033363,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте! Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 804.5752909965813,
      "inferenceMs": 818.0542910122313,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.20198871383018055,
      "audioSecondsPerSecond": 4.950771659651921,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 803.8530409685336,
      "voice": "ru:anastasia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.025284718722105026,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.19848280863686937,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 803.8553749793209,
      "loadMs": 21075.31841704622,
      "totalMs": 21893.37270805845,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1001.2592079583555,
      "inferenceMs": 1017.0427919947542,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.21524715174492154,
      "audioSecondsPerSecond": 4.645822218289092,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1000.4700419958681,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11977250128984451,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.2117402205199358,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1000.4725419566967,
      "loadMs": 21075.31841704622,
      "totalMs": 22092.361209040973,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Godmorgen, dette er en lokal test- og talesyntese.",
      "errorMetric": "wer",
      "errorRate": 0.125,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 705.228541046381,
      "inferenceMs": 716.0069999517873,
      "audioSeconds": 3.125,
      "realTimeFactor": 0.22912223998457193,
      "audioSecondsPerSecond": 4.364482470437333,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 704.576000047382,
      "voice": "da:freja",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.04741772264242172,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.22546505313366652,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 704.5782910427079,
      "loadMs": 21075.31841704622,
      "totalMs": 21791.325416998006,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hyvää huomenta, tämä on paikallinen puhesynteesin testi.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1012.1127919992432,
      "inferenceMs": 1026.3169999816455,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.24148635293685777,
      "audioSecondsPerSecond": 4.141020756818806,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1011.3437919644639,
      "voice": "fi:aino",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11310196667909622,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.23796393129262416,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1011.3467079936527,
      "loadMs": 21075.31841704622,
      "totalMs": 22101.635417027865,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 810.1767500047572,
      "inferenceMs": 821.9536250107922,
      "audioSeconds": 3.75,
      "realTimeFactor": 0.21918763333621125,
      "audioSecondsPerSecond": 4.562301188161027,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 809.5779579598457,
      "voice": "fr:camille",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.03980860114097595,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.2158880888018757,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 809.5803330070339,
      "loadMs": 21075.31841704622,
      "totalMs": 21897.27204205701,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "こんにちは、これはこの端末で音声を作るためのテストです。",
      "errorMetric": "cer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 903.3161249826662,
      "inferenceMs": 917.2007089946419,
      "audioSeconds": 4.35,
      "realTimeFactor": 0.2108507376999177,
      "audioSecondsPerSecond": 4.742691493084544,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 902.4750419775955,
      "voice": "ja:yuki",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.05849319323897362,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.20746641769609148,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 902.4789169779979,
      "loadMs": 21075.31841704622,
      "totalMs": 21992.51912604086,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " 안녕하세요 이 기기에서 음성을 만드는 시험입니다.",
      "errorMetric": "cer",
      "errorRate": 0,
      "maximumRepeatedWords": 2,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 734.5585830044001,
      "inferenceMs": 745.4175419989042,
      "audioSeconds": 3.25,
      "realTimeFactor": 0.22935924369197053,
      "audioSecondsPerSecond": 4.359972521286302,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 734.0095829567872,
      "voice": "ko:seoyeon",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.018892088904976845,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.22584984614513814,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 734.011999971699,
      "loadMs": 21075.31841704622,
      "totalMs": 21820.735959045123,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Goedemorgen, dit is een lokale test voor spraaksynthese.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 851.2480420176871,
      "inferenceMs": 864.5987079944462,
      "audioSeconds": 3.975,
      "realTimeFactor": 0.2175091089294204,
      "audioSecondsPerSecond": 4.597508605143016,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 850.5147920222953,
      "voice": "nl:sanne",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.082499660551548,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.2139665200023578,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 850.5169170093723,
      "loadMs": 21075.31841704622,
      "totalMs": 21939.917125040665,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " God morgon, det här är ett lokalt test av talsyntes.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 883.1436249893159,
      "inferenceMs": 896.5699169784784,
      "audioSeconds": 4.1,
      "realTimeFactor": 0.21867558950694596,
      "audioSecondsPerSecond": 4.572984128016887,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 882.4865419883281,
      "voice": "sv:astrid",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.12287326902151108,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.2152413009699979,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 882.4893339769915,
      "loadMs": 21075.31841704622,
      "totalMs": 21971.888334024698,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Habari za asubuhi hili ni jaribio la sauti kwenye kifahiki.",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 954.307832987979,
      "inferenceMs": 968.2678340468556,
      "audioSeconds": 4.475,
      "realTimeFactor": 0.21637270034566605,
      "audioSecondsPerSecond": 4.621655127483508,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 953.6127909668721,
      "voice": "sw:zawadi",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.11006385087966919,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.2130984169738003,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 953.6154159577563,
      "loadMs": 21075.31841704622,
      "totalMs": 22043.586251093075,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Günaydın bu cihazda yerel bir konuşma testi yapıyoruz.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1018.1574579910375,
      "inferenceMs": 1032.7982919989154,
      "audioSeconds": 3.875,
      "realTimeFactor": 0.26652859148359104,
      "audioSecondsPerSecond": 3.7519426881508333,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1017.4339580116794,
      "voice": "tr:elif",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.018762750551104546,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.26256443019355497,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 695.6534589990042,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1017.4371670000255,
      "loadMs": 21075.31841704622,
      "totalMs": 22108.116709045134,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 548
      },
      "loadMs": 19786.566792055964,
      "warmupMs": 392.48533302452415,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 342.14020904619247,
      "inferenceMs": 354.5499999891035,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.10583582089226971,
      "audioSecondsPerSecond": 9.44859681315176,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 337.4513750313781,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1438109129667282,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10073217672796281,
      "synthesisMs": 337.4527920386754,
      "totalMs": 20141.116792045068,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1519599616,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 548
      },
      "loadMs": 19786.566792055964,
      "warmupMs": 392.48533302452415,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 335.7902909629047,
      "inferenceMs": 348.3586250222288,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.10398764926036681,
      "audioSecondsPerSecond": 9.616526646315233,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 335.2267909795046,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1438109129667282,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10006818386948486,
      "synthesisMs": 335.2284159627743,
      "totalMs": 20134.925417078193,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1561493504,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 548
      },
      "loadMs": 19786.566792055964,
      "warmupMs": 392.48533302452415,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 328.4427920007147,
      "inferenceMs": 341.6097500012256,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.1019730597018584,
      "audioSecondsPerSecond": 9.806511670079619,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 327.7465000282973,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1438109129667282,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0978352364201321,
      "synthesisMs": 327.74804200744256,
      "totalMs": 20128.17654205719,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1569144832,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 548
      },
      "loadMs": 19786.566792055964,
      "warmupMs": 392.48533302452415,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 328.2691659987904,
      "inferenceMs": 341.6287919972092,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.10197874387976394,
      "audioSecondsPerSecond": 9.805965066396881,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 327.6855410076678,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1438109129667282,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0978170770199608,
      "synthesisMs": 327.6872080168687,
      "totalMs": 20128.195584053174,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1572765696,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 548
      },
      "loadMs": 18739.82779198559,
      "warmupMs": 386.7138749919832,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 340.7518749590963,
      "inferenceMs": 353.735750017222,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.10559276119917076,
      "audioSecondsPerSecond": 9.470346154825746,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 336.47145895520225,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1438109129667282,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10043971402111894,
      "synthesisMs": 336.47304197074845,
      "totalMs": 19093.563542002812,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1502740480,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 548
      },
      "loadMs": 18739.82779198559,
      "warmupMs": 386.7138749919832,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 337.05662505235523,
      "inferenceMs": 350.4075829987414,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.10459927850708699,
      "audioSecondsPerSecond": 9.560295388961467,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 336.39191603288054,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1438109129667282,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10041592031050084,
      "synthesisMs": 336.3933330401778,
      "totalMs": 19090.23537498433,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1507164160,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 548
      },
      "loadMs": 18739.82779198559,
      "warmupMs": 386.7138749919832,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 329.02295800158754,
      "inferenceMs": 342.94387500267476,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.10237130597094769,
      "audioSecondsPerSecond": 9.768362242871,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 328.4663329832256,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1438109129667282,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09805006209635801,
      "synthesisMs": 328.46770802279934,
      "totalMs": 19082.771666988265,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1517666304,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 548
      },
      "loadMs": 18739.82779198559,
      "warmupMs": 386.7138749919832,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 332.11237500654534,
      "inferenceMs": 345.84770898800343,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.10323812208597118,
      "audioSecondsPerSecond": 9.686344344458858,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 331.5464170300402,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1438109129667282,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09896945284762934,
      "synthesisMs": 331.5476670395583,
      "totalMs": 19085.675500973593,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1546551296,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 548
      },
      "loadMs": 19086.956417013425,
      "warmupMs": 726.3214169652201,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 153.02820899523795,
      "inferenceMs": 164.8879999993369,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.04922029850726475,
      "audioSecondsPerSecond": 20.31682111502033,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 149.09158396767452,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1413072645664215,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1356316672,
      "synthesisRealTimeFactor": 0.04450541045238723,
      "synthesisMs": 149.0931250154972,
      "totalMs": 19251.844417012762,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1330003968,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 548
      },
      "loadMs": 19086.956417013425,
      "warmupMs": 726.3214169652201,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 149.1472499910742,
      "inferenceMs": 161.82708303676918,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.04830659195127438,
      "audioSecondsPerSecond": 20.70110847415347,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 148.62533297855407,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1413072645664215,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1356316672,
      "synthesisRealTimeFactor": 0.044366330735441016,
      "synthesisMs": 148.62720796372741,
      "totalMs": 19248.783500050195,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1330085888,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 548
      },
      "loadMs": 19086.956417013425,
      "warmupMs": 726.3214169652201,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 156.31849999772385,
      "inferenceMs": 170.12137500569224,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.05078250000169918,
      "audioSecondsPerSecond": 19.69182296985261,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 155.7057080208324,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1413072645664215,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1356316672,
      "synthesisRealTimeFactor": 0.04647977611939631,
      "synthesisMs": 155.70724999997765,
      "totalMs": 19257.077792019118,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1330135040,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 548
      },
      "loadMs": 19086.956417013425,
      "warmupMs": 726.3214169652201,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 152.22816698951647,
      "inferenceMs": 165.73520802194253,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.04947319642446046,
      "audioSecondsPerSecond": 20.212965247290583,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 151.6837920062244,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1413072645664215,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1356316672,
      "synthesisRealTimeFactor": 0.04527950268447288,
      "synthesisMs": 151.68633399298415,
      "totalMs": 19252.691625035368,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1330135040,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 548
      },
      "loadMs": 18853.784917038865,
      "warmupMs": 656.6189170116559,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 163.69516699342057,
      "inferenceMs": 176.546040980611,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.05270031074048089,
      "audioSecondsPerSecond": 18.975220182750576,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 159.17383297346532,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1413072645664215,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1356316672,
      "synthesisRealTimeFactor": 0.04751507462169141,
      "synthesisMs": 159.17549998266622,
      "totalMs": 19030.330958019476,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1346994176,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 548
      },
      "loadMs": 18853.784917038865,
      "warmupMs": 656.6189170116559,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 155.94108297955245,
      "inferenceMs": 170.038582989946,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.05075778596714806,
      "audioSecondsPerSecond": 19.70141094505638,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 155.27524997014552,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1413072645664215,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1356316672,
      "synthesisRealTimeFactor": 0.04635149253351586,
      "synthesisMs": 155.27749998727813,
      "totalMs": 19023.82350002881,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1347092480,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 548
      },
      "loadMs": 18853.784917038865,
      "warmupMs": 656.6189170116559,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 164.2196670291014,
      "inferenceMs": 178.2123749726452,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.0531977238724314,
      "audioSecondsPerSecond": 18.79779673277016,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 163.6468750075437,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1413072645664215,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1356316672,
      "synthesisRealTimeFactor": 0.048850447759588266,
      "synthesisMs": 163.64899999462068,
      "totalMs": 19031.99729201151,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1347108864,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 548
      },
      "loadMs": 18853.784917038865,
      "warmupMs": 656.6189170116559,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 163.903082953766,
      "inferenceMs": 177.61637503281236,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.053019813442630556,
      "audioSecondsPerSecond": 18.86087360684582,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 163.34637498948723,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.1413072645664215,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1356316672,
      "synthesisRealTimeFactor": 0.04876063432919779,
      "synthesisMs": 163.3481250028126,
      "totalMs": 19031.401292071678,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1347108864,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 20297.179499990307,
      "warmupMs": 712.2441250248812,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 771.3913340121508,
      "inferenceMs": 783.7169169797562,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.23394534835216604,
      "audioSecondsPerSecond": 4.274502600900897,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 767.1252919826657,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.14355479180812836,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.22899319671455826,
      "synthesisMs": 767.1272089937702,
      "totalMs": 21080.896416970063,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1008762880,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 20297.179499990307,
      "warmupMs": 712.2441250248812,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 703.8443749770522,
      "inferenceMs": 716.7942500091158,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.21396843283854203,
      "audioSecondsPerSecond": 4.673586597489303,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 703.2126249978319,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.14369092881679535,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.20991493760149427,
      "synthesisMs": 703.2150409650058,
      "totalMs": 21013.973749999423,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1010089984,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 20297.179499990307,
      "warmupMs": 712.2441250248812,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 698.4597079572268,
      "inferenceMs": 711.9290830451064,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.21251614419256906,
      "audioSecondsPerSecond": 4.705524861649388,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 697.8714159922674,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.14382585883140564,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.20832057192629733,
      "synthesisMs": 697.873915953096,
      "totalMs": 21009.108583035413,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1010827264,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 20297.179499990307,
      "warmupMs": 712.2441250248812,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 698.4698330052197,
      "inferenceMs": 711.0099160345271,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.21224176598045585,
      "audioSecondsPerSecond": 4.711607988090734,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 697.8672909899615,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.143954336643219,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.20831942777218881,
      "synthesisMs": 697.8700830368325,
      "totalMs": 21008.189416024834,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 1011630080,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 22398.29658297822,
      "warmupMs": 931.6847920417786,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 726.9188330392353,
      "inferenceMs": 740.8796669915318,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.221158109549711,
      "audioSecondsPerSecond": 4.521651962191439,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 722.9152910294943,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.14355479180812836,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.21579625613209028,
      "synthesisMs": 722.9174580425024,
      "totalMs": 23139.17624996975,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 504545280,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 22398.29658297822,
      "warmupMs": 931.6847920417786,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 874.9068330507725,
      "inferenceMs": 888.0382499773987,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.26508604476937275,
      "audioSecondsPerSecond": 3.7723600307591036,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 874.2235000245273,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.14369092881679535,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.26096289791514293,
      "synthesisMs": 874.2257080157287,
      "totalMs": 23286.334832955617,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 506675200,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 22398.29658297822,
      "warmupMs": 931.6847920417786,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 670.2185409958474,
      "inferenceMs": 683.1329999840818,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.20392029850271098,
      "audioSecondsPerSecond": 4.903876697624124,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 669.4448749767616,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.14382585883140564,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.19983482566217323,
      "synthesisMs": 669.4466659682803,
      "totalMs": 23081.4295829623,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 507396096,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 22398.29658297822,
      "warmupMs": 931.6847920417786,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en",
        "de",
        "es",
        "it",
        "pt",
        "ru",
        "hi",
        "da",
        "fi",
        "fr",
        "ja",
        "ko",
        "nl",
        "sv",
        "sw",
        "tr"
      ],
      "sampleRate": 24000,
      "durationMs": 659.6483750035986,
      "inferenceMs": 672.2297919914126,
      "audioSeconds": 3.35,
      "realTimeFactor": 0.2006656095496754,
      "audioSecondsPerSecond": 4.983414957072885,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 659.0711250319146,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v2",
      "rms": 0.143954336643219,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.19673814686056948,
      "synthesisMs": 659.0727919829078,
      "totalMs": 23070.52637496963,
      "grade": "fast",
      "artifactRevision": "c01797caec3967b400da457656e011eb81f530a4",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 508510208,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Nano · German Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64109 ms / 3704 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64110 ms / 3622 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64110 ms / 3704 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64110 ms / 3704 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64111 ms / 3622 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64112 ms / 3622 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64114 ms / 3622 ms0.031 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64117 ms / 3704 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64118 ms / 3887 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64125 ms / 3887 ms0.034 RTF
FP32mpsAcceleratedPassedFastmacos · aarch642704 ms / 3887 ms0.101 RTF
FP32cpuPassedVery fastmacos · aarch64883 ms / 3362 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64130 ms / 3173 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64130 ms / 3173 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64131 ms / 3178 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64132 ms / 3178 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64132 ms / 3173 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64132 ms / 3173 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64134 ms / 3178 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64134 ms / 3178 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64140 ms / 3362 ms0.038 RTF
FP32cpuPassedVery fastmacos · aarch64142 ms / 3362 ms0.038 RTF
Q8cpuPassedVery fastmacos · aarch641130 ms / 2722 ms0.042 RTF
Q8cpuPassedVery fastmacos · aarch64166 ms / 2649 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64168 ms / 2649 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64169 ms / 2649 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64170 ms / 2547 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64171 ms / 2547 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64172 ms / 2547 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64172 ms / 2722 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64175 ms / 2722 ms0.047 RTF
Q8cpuPassedVery fastmacos · aarch64176 ms / 2649 ms0.048 RTF
Q8cpuPassedVery fastmacos · aarch64178 ms / 2547 ms0.048 RTF
FP32cpuPassedVery fastmacos · aarch641490 ms / 3410 ms0.055 RTF
FP32cpuPassedVery fastmacos · aarch64208 ms / 3410 ms0.056 RTF
FP32cpuPassedVery fastmacos · aarch64211 ms / 3410 ms0.057 RTF
Q8cpuPassedFastmacos · aarch642776 ms / 2990 ms0.103 RTF
Q8cpuPassedFastmacos · aarch64390 ms / 2990 ms0.105 RTF
Q8cpuPassedFastmacos · aarch64400 ms / 2990 ms0.108 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:01:14.997936+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
54526b8d8e45333cd720d3a082c106aacf416064
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 198.36416700854897,
      "inferenceMs": 210.97570902202278,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.057020461897843995,
      "audioSecondsPerSecond": 17.53756400275339,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 195.72766701458022,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16126641631126404,
      "clippedFraction": 0.0005968468468468468,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05289976351353264,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 206.98200003243983,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 195.72912500007078,
      "loadMs": 3409.843000001274,
      "totalMs": 3620.818709023297,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 195.42674999684095,
      "inferenceMs": 208.31808302318677,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.05630218460086129,
      "audioSecondsPerSecond": 17.761300153612552,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 194.82187496032566,
      "voice": "de:mia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08109333366155624,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05265497756694015,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 206.98200003243983,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 194.82341699767858,
      "loadMs": 3409.843000001274,
      "totalMs": 3618.161083024461,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest. Guten Morgen, Dies ist ein lokaler Sprachtest. Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen. Dies ist ein lokaler Sprachtest. Guten Morgen. Dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1403.1265000230633,
      "inferenceMs": 1489.5071660284884,
      "audioSeconds": 26.9,
      "realTimeFactor": 0.055372013607007005,
      "audioSecondsPerSecond": 18.05966470891454,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 927.642458002083,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19866396486759186,
      "clippedFraction": 0.001290272614622057,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.052014394350813835,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 206.98200003243983,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1399.1872080368921,
      "loadMs": 3409.843000001274,
      "totalMs": 4899.3501660297625,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 129.37358301132917,
      "inferenceMs": 142.31966697843745,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03846477485903715,
      "audioSecondsPerSecond": 25.997812379371144,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 126.80899997940287,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16126641631126404,
      "clippedFraction": 0.0005968468468468468,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.034273119463988054,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 137.6939169713296,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 126.81054201675579,
      "loadMs": 3362.026999995578,
      "totalMs": 3504.3466669740155,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 126.4383330126293,
      "inferenceMs": 139.53133299946785,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03771117108093726,
      "audioSecondsPerSecond": 26.517341449136097,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 125.64283300889656,
      "voice": "de:mia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08109333366155624,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.033958153232831405,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 137.6939169713296,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 125.6451669614762,
      "loadMs": 3362.026999995578,
      "totalMs": 3501.558332995046,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest. Guten Morgen, Dies ist ein lokaler Sprachtest. Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen. Dies ist ein lokaler Sprachtest. Guten Morgen. Dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 795.3454999951646,
      "inferenceMs": 883.1517079961486,
      "audioSeconds": 26.9,
      "realTimeFactor": 0.03283091851286798,
      "audioSecondsPerSecond": 30.45909299211513,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 494.40579203655943,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19866396486759186,
      "clippedFraction": 0.001290272614622057,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02943314282573261,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 137.6939169713296,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 791.7515420122072,
      "loadMs": 3362.026999995578,
      "totalMs": 4245.178707991727,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 111.1924159922637,
      "inferenceMs": 125.06779201794416,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03380210595079572,
      "audioSecondsPerSecond": 29.583955551635075,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 108.2930410047993,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16098365187644958,
      "clippedFraction": 0.0005067567567567568,
      "driverAllocatedMemoryBytes": 205373440,
      "synthesisRealTimeFactor": 0.02926887377995897,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 560.5636670370586,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 108.29483298584819,
      "loadMs": 3887.3814159887843,
      "totalMs": 4012.4492080067284,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 105.12937500607222,
      "inferenceMs": 118.34000004455447,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03198378379582553,
      "audioSecondsPerSecond": 31.265844166021353,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 104.5305419829674,
      "voice": "de:mia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08095905929803848,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 205373440,
      "synthesisRealTimeFactor": 0.028252083512425825,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 560.5636670370586,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 104.53270899597555,
      "loadMs": 3887.3814159887843,
      "totalMs": 4005.7214160333388,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest. Guten Morgen, Dies ist ein lokaler Sprachtest. Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen. Dies ist ein lokaler Sprachtest. Guten Morgen. Dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2616.6327919927426,
      "inferenceMs": 2704.175499966368,
      "audioSeconds": 26.9,
      "realTimeFactor": 0.1005269702589728,
      "audioSecondsPerSecond": 9.94757921604369,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1411.4836669759825,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1985754817724228,
      "clippedFraction": 0.001293370508054523,
      "driverAllocatedMemoryBytes": 1379041280,
      "synthesisRealTimeFactor": 0.09711562732250177,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 560.5636670370586,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2612.4103749752976,
      "loadMs": 3887.3814159887843,
      "totalMs": 6591.556915955152,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 387.727957975585,
      "inferenceMs": 399.97916598804295,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.10810247729406566,
      "audioSecondsPerSecond": 9.250481811621679,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 385.79520798521116,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1614098697900772,
      "clippedFraction": 0.0005630630630630631,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10426966215458673,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 389.0660419710912,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 385.7977499719709,
      "loadMs": 2990.298166987486,
      "totalMs": 3390.277332975529,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 377.64104100642726,
      "inferenceMs": 389.7954170242883,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.10535011270926711,
      "audioSecondsPerSecond": 9.492158805370078,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 376.95620802696794,
      "voice": "de:mia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08144448697566986,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10188065297788361,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 389.0660419710912,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 376.9584160181694,
      "loadMs": 2990.298166987486,
      "totalMs": 3380.0935840117745,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest. Guten Morgen. dies ist ein lokaler Sprachtest. Guten Morgen. Dies ist ein lokaler Sprachtest, Guten Morgen. Dies ist ein lokaler Sprachtest. Guten Morgen, Dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2685.849374975078,
      "inferenceMs": 2775.6397079792805,
      "audioSeconds": 26.9,
      "realTimeFactor": 0.10318363226688775,
      "audioSecondsPerSecond": 9.691459566120605,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1800.7832909934223,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19849854707717896,
      "clippedFraction": 0.0012345105328376703,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09968241479561134,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 389.0660419710912,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2681.456958001945,
      "loadMs": 2990.298166987486,
      "totalMs": 5765.937874966767,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 162.37679199548438,
      "inferenceMs": 174.5080419932492,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.047164335673851136,
      "audioSecondsPerSecond": 21.2024612604566,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 160.6995000038296,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1614079475402832,
      "clippedFraction": 0.0005630630630630631,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04343292811205862,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 167.1094159828499,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 160.7018340146169,
      "loadMs": 2721.6602920088917,
      "totalMs": 2896.168334002141,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 159.86054099630564,
      "inferenceMs": 171.91933299181983,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.04646468459238374,
      "audioSecondsPerSecond": 21.521721470243556,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 159.1804579948075,
      "voice": "de:mia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08144822716712952,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0430221959470293,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 167.1094159828499,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 159.1821250040084,
      "loadMs": 2721.6602920088917,
      "totalMs": 2893.5796250007115,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest. Guten Morgen. dies ist ein lokaler Sprachtest. Guten Morgen. Dies ist ein lokaler Sprachtest, Guten Morgen. Dies ist ein lokaler Sprachtest. Guten Morgen, Dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1041.1939169862308,
      "inferenceMs": 1130.1197499851696,
      "audioSeconds": 26.9,
      "realTimeFactor": 0.04201188661654905,
      "audioSecondsPerSecond": 23.802787271307313,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 665.1517919963226,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1984977275133133,
      "clippedFraction": 0.0012345105328376703,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03854535163473684,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 167.1094159828499,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1036.869958974421,
      "loadMs": 2721.6602920088917,
      "totalMs": 3851.7800419940613,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3173.3121250290424,
      "warmupMs": 123.79441602388397,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 119.4994580000639,
      "inferenceMs": 132.11649999720976,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03570716216140804,
      "audioSecondsPerSecond": 28.00558597963269,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 116.77983298432082,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16126641631126404,
      "clippedFraction": 0.0005968468468468468,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03156246622155949,
      "synthesisMs": 116.78112501977012,
      "totalMs": 3305.428625026252,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 618496000,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3173.3121250290424,
      "warmupMs": 123.79441602388397,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 119.13208401529118,
      "inferenceMs": 131.94374996237457,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03566047296280394,
      "audioSecondsPerSecond": 28.042252861958993,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 118.46604204038158,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16126641631126404,
      "clippedFraction": 0.0005968468468468468,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.032018310818317776,
      "synthesisMs": 118.46775002777576,
      "totalMs": 3305.255874991417,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 652558336,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3173.3121250290424,
      "warmupMs": 123.79441602388397,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 117.54058301448822,
      "inferenceMs": 130.3284169989638,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.035223896486206434,
      "audioSecondsPerSecond": 28.389817702070435,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 116.90804100362584,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16126641631126404,
      "clippedFraction": 0.0005968468468468468,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.031597162172122786,
      "synthesisMs": 116.9095000368543,
      "totalMs": 3303.640542028006,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 657276928,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3173.3121250290424,
      "warmupMs": 123.79441602388397,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 116.25183396972716,
      "inferenceMs": 129.7370829852298,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03506407648249454,
      "audioSecondsPerSecond": 28.51921682578014,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 115.63974997261539,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16126641631126404,
      "clippedFraction": 0.0005968468468468468,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03125436945403045,
      "synthesisMs": 115.64116697991267,
      "totalMs": 3303.049208014272,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 657276928,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3178.4711250220425,
      "warmupMs": 127.1962079918012,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 120.90854102279991,
      "inferenceMs": 133.71937500778586,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03614037162372591,
      "audioSecondsPerSecond": 27.66988702859676,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 118.01066604675725,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16126641631126404,
      "clippedFraction": 0.0005968468468468468,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.031895270284793865,
      "synthesisMs": 118.01250005373731,
      "totalMs": 3312.1905000298284,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 616677376,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3178.4711250220425,
      "warmupMs": 127.1962079918012,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 118.55883302632719,
      "inferenceMs": 131.73983298474923,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03560536026614844,
      "audioSecondsPerSecond": 28.08565880319833,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 117.91979102417827,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16126641631126404,
      "clippedFraction": 0.0005968468468468468,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03187068676689287,
      "synthesisMs": 117.92154103750363,
      "totalMs": 3310.2109580067918,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 617562112,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3178.4711250220425,
      "warmupMs": 127.1962079918012,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 119.90575003437698,
      "inferenceMs": 133.9420000440441,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03620054055244435,
      "audioSecondsPerSecond": 27.62389690151954,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 119.29366702679545,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16126641631126404,
      "clippedFraction": 0.0005968468468468468,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03224200460543806,
      "synthesisMs": 119.29541704012081,
      "totalMs": 3312.4131250660866,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 617578496,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3178.4711250220425,
      "warmupMs": 127.1962079918012,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 117.83133295830339,
      "inferenceMs": 131.08529103919864,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03542845703762125,
      "audioSecondsPerSecond": 28.22589758673674,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 117.22366599133238,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16126641631126404,
      "clippedFraction": 0.0005968468468468468,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03168255620875169,
      "synthesisMs": 117.22545797238126,
      "totalMs": 3309.556416061241,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 617742336,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3703.518082969822,
      "warmupMs": 453.7773330230266,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 104.1494170203805,
      "inferenceMs": 117.0345830032602,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.031630968379259515,
      "audioSecondsPerSecond": 31.614586945611883,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 101.66441701585427,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16098365187644958,
      "clippedFraction": 0.0005067567567567568,
      "driverAllocatedMemoryBytes": 205373440,
      "synthesisRealTimeFactor": 0.027477353515506193,
      "synthesisMs": 101.66620800737292,
      "totalMs": 3820.552665973082,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 698187776,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3703.518082969822,
      "warmupMs": 453.7773330230266,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 97.9695410351269,
      "inferenceMs": 110.32937502022833,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.029818750005467114,
      "audioSecondsPerSecond": 33.535946336337204,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.26912504993379,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16098365187644958,
      "clippedFraction": 0.0005067567567567568,
      "driverAllocatedMemoryBytes": 205373440,
      "synthesisRealTimeFactor": 0.026289481902495027,
      "synthesisMs": 97.2710830392316,
      "totalMs": 3813.8474579900503,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 698220544,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3703.518082969822,
      "warmupMs": 453.7773330230266,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 96.64929198333994,
      "inferenceMs": 109.11891603609547,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.02949159892867445,
      "audioSecondsPerSecond": 33.90796146450058,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 96.09662496950477,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16098365187644958,
      "clippedFraction": 0.0005067567567567568,
      "driverAllocatedMemoryBytes": 205373440,
      "synthesisRealTimeFactor": 0.025972477569420997,
      "synthesisMs": 96.0981670068577,
      "totalMs": 3812.6369990059175,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 698220544,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3703.518082969822,
      "warmupMs": 453.7773330230266,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 97.36112499376759,
      "inferenceMs": 110.45300000114366,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.02985216216247126,
      "audioSecondsPerSecond": 33.49841108853258,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 96.81812499184161,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16098365187644958,
      "clippedFraction": 0.0005067567567567568,
      "driverAllocatedMemoryBytes": 205357056,
      "synthesisRealTimeFactor": 0.026167421082563296,
      "synthesisMs": 96.8194580054842,
      "totalMs": 3813.9710829709657,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 698220544,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3622.4955830257386,
      "warmupMs": 454.38512502005324,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 102.14866703609005,
      "inferenceMs": 114.0391249791719,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03082138512950592,
      "audioSecondsPerSecond": 32.44500517410817,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 99.71887501887977,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16098365187644958,
      "clippedFraction": 0.0005067567567567568,
      "driverAllocatedMemoryBytes": 205373440,
      "synthesisRealTimeFactor": 0.026951441619973128,
      "synthesisMs": 99.72033399390057,
      "totalMs": 3736.5347080049105,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695025664,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3622.4955830257386,
      "warmupMs": 454.38512502005324,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 97.53825003281236,
      "inferenceMs": 109.96591695584357,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.029720518096173938,
      "audioSecondsPerSecond": 33.646788954487796,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 96.98158298851922,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16098365187644958,
      "clippedFraction": 0.0005067567567567568,
      "driverAllocatedMemoryBytes": 205373440,
      "synthesisRealTimeFactor": 0.026211801891189976,
      "synthesisMs": 96.9836669974029,
      "totalMs": 3732.461499981582,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695042048,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3622.4955830257386,
      "warmupMs": 454.38512502005324,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 99.19699997408316,
      "inferenceMs": 111.56525003025308,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.03015277027844678,
      "audioSecondsPerSecond": 33.1644485984361,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 98.65962498588488,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16098365187644958,
      "clippedFraction": 0.0005067567567567568,
      "driverAllocatedMemoryBytes": 205373440,
      "synthesisRealTimeFactor": 0.026665225395353865,
      "synthesisMs": 98.6613339628093,
      "totalMs": 3734.0608330559917,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695042048,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3622.4955830257386,
      "warmupMs": 454.38512502005324,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 97.99804200883955,
      "inferenceMs": 110.93166697537526,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.02998153161496629,
      "audioSecondsPerSecond": 33.35386640156891,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.45674999430776,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16098365187644958,
      "clippedFraction": 0.0005067567567567568,
      "driverAllocatedMemoryBytes": 205357056,
      "synthesisRealTimeFactor": 0.026339988925846648,
      "synthesisMs": 97.45795902563259,
      "totalMs": 3733.427250001114,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695042048,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2547.309999994468,
      "warmupMs": 165.07208300754428,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 165.75745795853436,
      "inferenceMs": 178.34720795508474,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.04820194809596885,
      "audioSecondsPerSecond": 20.746049475200163,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 164.20533298514783,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1614079475402832,
      "clippedFraction": 0.0005630630630630631,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04438043918108215,
      "synthesisMs": 164.20762497000396,
      "totalMs": 2725.6572079495527,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 531087360,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2547.309999994468,
      "warmupMs": 165.07208300754428,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 159.21083296416327,
      "inferenceMs": 171.63695901399478,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.04638836730107967,
      "audioSecondsPerSecond": 21.557128611782925,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 158.50254200631753,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16151919960975647,
      "clippedFraction": 0.0005292792792792793,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04283917783067335,
      "synthesisMs": 158.5049579734914,
      "totalMs": 2718.9469590084627,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 533069824,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2547.309999994468,
      "warmupMs": 165.07208300754428,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 158.2192910136655,
      "inferenceMs": 170.86091596866027,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.04617862593747575,
      "audioSecondsPerSecond": 21.655040177114955,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 157.49404102098197,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16142995655536652,
      "clippedFraction": 0.0005292792792792793,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04256652027834207,
      "synthesisMs": 157.49612502986565,
      "totalMs": 2718.170915963128,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 534937600,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2547.309999994468,
      "warmupMs": 165.07208300754428,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 157.21579198725522,
      "inferenceMs": 170.2335420413874,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.046009065416591194,
      "audioSecondsPerSecond": 21.73484705558468,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 156.57983301207423,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1615028977394104,
      "clippedFraction": 0.0005518018018018018,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04231953837694852,
      "synthesisMs": 156.58229199470952,
      "totalMs": 2717.5435420358554,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 534937600,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2648.840499983635,
      "warmupMs": 163.38799998629838,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 163.64454198628664,
      "inferenceMs": 175.96470803255215,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.047558029197987066,
      "audioSecondsPerSecond": 21.026943648924917,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 162.08199999528006,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1614079475402832,
      "clippedFraction": 0.0005630630630630631,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04380650890714212,
      "synthesisMs": 162.08408295642585,
      "totalMs": 2824.805208016187,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 531120128,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2648.840499983635,
      "warmupMs": 163.38799998629838,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 156.7866670084186,
      "inferenceMs": 168.8902079476975,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.045646002148026346,
      "audioSecondsPerSecond": 21.907723632774665,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 156.12150001106784,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16151919960975647,
      "clippedFraction": 0.0005292792792792793,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04219557432382292,
      "synthesisMs": 156.1236249981448,
      "totalMs": 2817.7307079313323,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 531775488,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2648.840499983635,
      "warmupMs": 163.38799998629838,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 153.63112499471754,
      "inferenceMs": 165.5064999940805,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.044731486484886625,
      "audioSecondsPerSecond": 22.355617453890535,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 153.02258299198002,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16142995655536652,
      "clippedFraction": 0.0005292792792792793,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04135789404736479,
      "synthesisMs": 153.02420797524974,
      "totalMs": 2814.3469999777153,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 531775488,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2648.840499983635,
      "warmupMs": 163.38799998629838,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 155.15145903918892,
      "inferenceMs": 167.89587499806657,
      "audioSeconds": 3.7,
      "realTimeFactor": 0.04537726351299096,
      "audioSecondsPerSecond": 22.037468163185117,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 154.5425840304233,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1615028977394104,
      "clippedFraction": 0.0005518018018018018,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04176876136429004,
      "synthesisMs": 154.54441704787314,
      "totalMs": 2816.7363749817014,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 531775488,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Nano · English Desktop

44 configurations tested
Passed44
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64109 ms / 3568 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64109 ms / 3872 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64110 ms / 3568 ms0.031 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64111 ms / 3568 ms0.031 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64111 ms / 3872 ms0.031 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64113 ms / 3568 ms0.031 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64114 ms / 3872 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64117 ms / 3872 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64123 ms / 4679 ms0.034 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64305 ms / 4679 ms0.094 RTF
FP32mpsAcceleratedPassedFastmacos · aarch642499 ms / 4679 ms0.100 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641056 ms / 4679 ms0.250 RTF
FP32cpuPassedVery fastmacos · aarch64126 ms / 2761 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64126 ms / 2761 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64127 ms / 2761 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64129 ms / 2761 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64922 ms / 3049 ms0.037 RTF
FP32cpuPassedVery fastmacos · aarch64134 ms / 2984 ms0.037 RTF
FP32cpuPassedVery fastmacos · aarch64135 ms / 2984 ms0.037 RTF
FP32cpuPassedVery fastmacos · aarch64135 ms / 3049 ms0.038 RTF
FP32cpuPassedVery fastmacos · aarch64160 ms / 3049 ms0.038 RTF
FP32cpuPassedVery fastmacos · aarch64137 ms / 2984 ms0.038 RTF
FP32cpuPassedVery fastmacos · aarch64139 ms / 2984 ms0.038 RTF
Q8cpuPassedVery fastmacos · aarch641059 ms / 3447 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64195 ms / 3447 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64167 ms / 3331 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64169 ms / 3331 ms0.047 RTF
Q8cpuPassedVery fastmacos · aarch64170 ms / 3331 ms0.047 RTF
FP32cpuPassedVery fastmacos · aarch64154 ms / 3049 ms0.048 RTF
Q8cpuPassedVery fastmacos · aarch64171 ms / 3461 ms0.048 RTF
Q8cpuPassedVery fastmacos · aarch64174 ms / 3447 ms0.048 RTF
Q8cpuPassedVery fastmacos · aarch64177 ms / 3331 ms0.049 RTF
Q8cpuPassedVery fastmacos · aarch64160 ms / 3447 ms0.049 RTF
Q8cpuPassedVery fastmacos · aarch64180 ms / 3461 ms0.050 RTF
Q8cpuPassedVery fastmacos · aarch64183 ms / 3461 ms0.051 RTF
Q8cpuPassedVery fastmacos · aarch64192 ms / 3461 ms0.053 RTF
FP32cpuPassedVery fastmacos · aarch64232 ms / 3096 ms0.055 RTF
FP32cpuPassedVery fastmacos · aarch641404 ms / 3096 ms0.056 RTF
FP32cpuPassedVery fastmacos · aarch64205 ms / 3096 ms0.057 RTF
FP32cpuPassedVery fastmacos · aarch64186 ms / 3096 ms0.057 RTF
Q8cpuPassedFastmacos · aarch642607 ms / 3636 ms0.105 RTF
Q8cpuPassedFastmacos · aarch64387 ms / 3636 ms0.107 RTF
Q8cpuPassedFastmacos · aarch64350 ms / 3636 ms0.108 RTF
Q8cpuPassedFastmacos · aarch64482 ms / 3636 ms0.113 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:00:52.601768+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
54526b8d8e45333cd720d3a082c106aacf416064
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 191.4728750125505,
      "inferenceMs": 204.87454097019508,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.05690959471394308,
      "audioSecondsPerSecond": 17.5717294250032,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 188.18837497383356,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485864222049713,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05227491916027955,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 201.71320799272507,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 188.18970897700638,
      "loadMs": 3095.6527910311706,
      "totalMs": 3300.5273320013657,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 173.5555420164019,
      "inferenceMs": 186.15579197648913,
      "audioSeconds": 3.25,
      "realTimeFactor": 0.05727870522353511,
      "audioSecondsPerSecond": 17.458495196380806,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 172.83550003776327,
      "voice": "en:martha",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11689023673534393,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05318096154727615,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 201.71320799272507,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 172.83812502864748,
      "loadMs": 3095.6527910311706,
      "totalMs": 3281.8085830076598,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello. This is a local speech synthesis test. Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1305.5379589786753,
      "inferenceMs": 1404.4584999792278,
      "audioSeconds": 24.875,
      "realTimeFactor": 0.05646064321524534,
      "audioSecondsPerSecond": 17.711452492450228,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 955.1194169907831,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19161154329776764,
      "clippedFraction": 0.000018425460636515914,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05233651091433575,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 201.71320799272507,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1301.8707089941017,
      "loadMs": 3095.6527910311706,
      "totalMs": 4500.111291010398,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 214.17566697346047,
      "inferenceMs": 231.65254201740026,
      "audioSeconds": 4.225,
      "realTimeFactor": 0.054829004027787044,
      "audioSecondsPerSecond": 18.238522069326763,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 213.46829197136685,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1911122351884842,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.050525394552139315,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 201.71320799272507,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 213.4697919827886,
      "loadMs": 3095.6527910311706,
      "totalMs": 3327.305333048571,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 121.11120799090713,
      "inferenceMs": 135.45891700778157,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.03762747694660599,
      "audioSecondsPerSecond": 26.576323504736088,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 117.5980829866603,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485864222049713,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03266668972274703,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 135.80829201964661,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 117.60008300188929,
      "loadMs": 3048.7783330027014,
      "totalMs": 3184.237250010483,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 136.65499998023733,
      "inferenceMs": 154.43329099798575,
      "audioSeconds": 3.25,
      "realTimeFactor": 0.047517935691687926,
      "audioSecondsPerSecond": 21.044685242396273,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 136.01974997436628,
      "voice": "en:martha",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11689023673534393,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.041852833237499,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 135.80829201964661,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 136.02170802187175,
      "loadMs": 3048.7783330027014,
      "totalMs": 3203.211624000687,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello. This is a local speech synthesis test. Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 822.7804169873707,
      "inferenceMs": 921.5490419883281,
      "audioSeconds": 24.875,
      "realTimeFactor": 0.037047197667872483,
      "audioSecondsPerSecond": 26.992594931605446,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 581.9439589977264,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19161154329776764,
      "clippedFraction": 0.000018425460636515914,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03291377222081225,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 135.80829201964661,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 818.7300839927047,
      "loadMs": 3048.7783330027014,
      "totalMs": 3970.3273749910295,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 142.3557089874521,
      "inferenceMs": 160.0617499789223,
      "audioSeconds": 4.225,
      "realTimeFactor": 0.037884437864833684,
      "audioSecondsPerSecond": 26.396062773000846,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 141.66633394779637,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1911122351884842,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03353087786728373,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 135.80829201964661,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 141.66795898927376,
      "loadMs": 3048.7783330027014,
      "totalMs": 3208.8400829816237,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 108.32216602284461,
      "inferenceMs": 122.96716700075194,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.034157546389097765,
      "audioSecondsPerSecond": 29.2761074993131,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 104.56775000784546,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1484956443309784,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 205144064,
      "synthesisRealTimeFactor": 0.029047280000264034,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1647.3652920103632,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 104.57020800095052,
      "loadMs": 4679.302083968651,
      "totalMs": 4802.2692509694025,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 291.7435000417754,
      "inferenceMs": 304.64499996742234,
      "audioSeconds": 3.25,
      "realTimeFactor": 0.09373692306689918,
      "audioSecondsPerSecond": 10.668154738622146,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 291.1927080131136,
      "voice": "en:martha",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11694340407848358,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 217202688,
      "synthesisRealTimeFactor": 0.08959848707756744,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1647.3652920103632,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 291.1950830020942,
      "loadMs": 4679.302083968651,
      "totalMs": 4983.947083936073,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello. This is a local speech synthesis test. Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2402.8678339673206,
      "inferenceMs": 2499.242875026539,
      "audioSeconds": 24.875,
      "realTimeFactor": 0.10047207537795132,
      "audioSecondsPerSecond": 9.953014270266094,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1340.1554589509033,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19166968762874603,
      "clippedFraction": 0.000018425460636515914,
      "driverAllocatedMemoryBytes": 1390362624,
      "synthesisRealTimeFactor": 0.09639880739517473,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1647.3652920103632,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2397.9203339549713,
      "loadMs": 4679.302083968651,
      "totalMs": 7178.5449589951895,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1039.1089170007035,
      "inferenceMs": 1056.4267080044374,
      "audioSeconds": 4.225,
      "realTimeFactor": 0.2500418243797485,
      "audioSecondsPerSecond": 3.999330921859137,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 1038.40154199861,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19111861288547516,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1396736000,
      "synthesisRealTimeFactor": 0.24577586793136844,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1647.3652920103632,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1038.4030420100316,
      "loadMs": 4679.302083968651,
      "totalMs": 5735.728791973088,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 372.6966250105761,
      "inferenceMs": 386.8408749694936,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.10745579860263711,
      "audioSecondsPerSecond": 9.306152045809268,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 369.10462501691654,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14868292212486267,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10252987249663824,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 377.9639999847859,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 369.10754098789766,
      "loadMs": 3635.5145419947803,
      "totalMs": 4022.355416964274,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 337.846249982249,
      "inferenceMs": 350.12470796937123,
      "audioSeconds": 3.25,
      "realTimeFactor": 0.10773067937519115,
      "audioSecondsPerSecond": 9.282406885389845,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 337.19270897563547,
      "voice": "en:martha",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11665821075439453,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10375219230683377,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 377.9639999847859,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 337.19462499720976,
      "loadMs": 3635.5145419947803,
      "totalMs": 3985.6392499641515,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test, hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello. This is a local speech synthesis test. hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2511.0633749864064,
      "inferenceMs": 2607.1908330195583,
      "audioSeconds": 24.85,
      "realTimeFactor": 0.10491713613760797,
      "audioSecondsPerSecond": 9.53133145655456,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1866.6595409740694,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19199766218662262,
      "clippedFraction": 0.000021797451374916164,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10088966128751785,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 377.9639999847859,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2507.1080829948187,
      "loadMs": 3635.5145419947803,
      "totalMs": 6242.705375014339,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 464.63216602569446,
      "inferenceMs": 481.82145901955664,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.11336975506342509,
      "audioSecondsPerSecond": 8.820694720920466,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 463.8169159879908,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1904372125864029,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10913383317015626,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 377.9639999847859,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 463.81879097316414,
      "loadMs": 3635.5145419947803,
      "totalMs": 4117.336001014337,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 160.7734999852255,
      "inferenceMs": 174.08195900497958,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.04835609972360544,
      "audioSecondsPerSecond": 20.679914337918397,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 157.29470801306888,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14868450164794922,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.043693668881638184,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 164.99345895135775,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 157.29720797389746,
      "loadMs": 3447.3844169988297,
      "totalMs": 3621.4663760038093,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 147.27091696113348,
      "inferenceMs": 160.17237497726455,
      "audioSeconds": 3.25,
      "realTimeFactor": 0.04928380768531217,
      "audioSecondsPerSecond": 20.29064000868637,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 146.68087498284876,
      "voice": "en:martha",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11665967851877213,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04513337169415676,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 164.99345895135775,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 146.68345800600946,
      "loadMs": 3447.3844169988297,
      "totalMs": 3607.5567919760942,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello, this is a local speech synthesis test, hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. Hello, this is a local speech synthesis test. hello, this is a local speech synthesis test. hello. This is a local speech synthesis test. hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 962.5922499690205,
      "inferenceMs": 1059.0397919877432,
      "audioSeconds": 24.85,
      "realTimeFactor": 0.04261729545222307,
      "audioSecondsPerSecond": 23.464651836507766,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 688.4307920117863,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1919984221458435,
      "clippedFraction": 0.000021797451374916164,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03858519786682122,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 164.99345895135775,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 958.8421669905074,
      "loadMs": 3447.3844169988297,
      "totalMs": 4506.424208986573,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 178.98008303018287,
      "inferenceMs": 195.20570896565914,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.04593075505074333,
      "audioSecondsPerSecond": 21.771904226160036,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 178.2799999928102,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19043511152267456,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04194874494500896,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 164.99345895135775,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 178.2821660162881,
      "loadMs": 3447.3844169988297,
      "totalMs": 3642.590125964489,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 2983.8578749913722,
      "warmupMs": 128.26529098674655,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 120.55929197231308,
      "inferenceMs": 134.45300003513694,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.03734805556531581,
      "audioSecondsPerSecond": 26.775155623594884,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 117.05208296189085,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485864222049713,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03251486110255226,
      "synthesisMs": 117.05349996918812,
      "totalMs": 3118.310875026509,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 780353536,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 2983.8578749913722,
      "warmupMs": 128.26529098674655,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 120.7298330264166,
      "inferenceMs": 134.86658304464072,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.03746293973462242,
      "audioSecondsPerSecond": 26.693046703855494,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 120.12345803668723,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485864222049713,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03336803250325223,
      "synthesisMs": 120.12491701170802,
      "totalMs": 3118.724458036013,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 782106624,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 2983.8578749913722,
      "warmupMs": 128.26529098674655,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 124.11412497749552,
      "inferenceMs": 138.59062502160668,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.03849739583933519,
      "audioSecondsPerSecond": 25.975782989929872,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 123.44933295389637,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485864222049713,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.034291944434193686,
      "synthesisMs": 123.45099996309727,
      "totalMs": 3122.448500012979,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 790724608,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 2983.8578749913722,
      "warmupMs": 128.26529098674655,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 123.12233401462436,
      "inferenceMs": 137.40866596344858,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.03816907387873572,
      "audioSecondsPerSecond": 26.199220949846197,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 122.48858396196738,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485864222049713,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03402499999437067,
      "synthesisMs": 122.48999997973442,
      "totalMs": 3121.266540954821,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 791298048,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 2761.3737499923445,
      "warmupMs": 122.60174995753914,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 116.1290830350481,
      "inferenceMs": 129.41712501924485,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.03594920139423468,
      "audioSecondsPerSecond": 27.817029620034177,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 112.87179199280217,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485864222049713,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03135364583512354,
      "synthesisMs": 112.87312500644475,
      "totalMs": 2890.7908750115894,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 779632640,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 2761.3737499923445,
      "warmupMs": 122.60174995753914,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 112.67020797822624,
      "inferenceMs": 126.17629097076133,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.03504896971410037,
      "audioSecondsPerSecond": 28.53150914726304,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 112.06962500000373,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485864222049713,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.031130891115430534,
      "synthesisMs": 112.07120801554993,
      "totalMs": 2887.550040963106,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 780746752,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 2761.3737499923445,
      "warmupMs": 122.60174995753914,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 112.47845797333866,
      "inferenceMs": 126.05754198739305,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.03501598388538696,
      "audioSecondsPerSecond": 28.558386457829187,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 111.84166697785258,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485864222049713,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.031067511655338522,
      "synthesisMs": 111.84304195921868,
      "totalMs": 2887.4312919797376,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 780812288,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 2761.3737499923445,
      "warmupMs": 122.60174995753914,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 113.61525004031137,
      "inferenceMs": 127.04908294836059,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.035291411930100165,
      "audioSecondsPerSecond": 28.335505589310145,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 113.02066699136049,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485864222049713,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03139502305809098,
      "synthesisMs": 113.02208300912753,
      "totalMs": 2888.422832940705,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 783859712,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3871.924958017189,
      "warmupMs": 615.2774170041084,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 102.51512500690296,
      "inferenceMs": 116.60033301450312,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.032388981392917536,
      "audioSecondsPerSecond": 30.87469741233261,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 99.21637497609481,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1484956443309784,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 205160448,
      "synthesisRealTimeFactor": 0.027560752221486635,
      "synthesisMs": 99.21870799735188,
      "totalMs": 3988.525291031692,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 867500032,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3871.924958017189,
      "warmupMs": 615.2774170041084,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 99.80445797555149,
      "inferenceMs": 114.26483298419043,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.031740231384497344,
      "audioSecondsPerSecond": 31.505756460503402,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 99.16183300083503,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1484956443309784,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 205160448,
      "synthesisRealTimeFactor": 0.02754552082882987,
      "synthesisMs": 99.16387498378754,
      "totalMs": 3986.1897910013795,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 867532800,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3871.924958017189,
      "warmupMs": 615.2774170041084,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 95.61224997742102,
      "inferenceMs": 109.37504202593118,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.030381956118314218,
      "audioSecondsPerSecond": 32.914273067401375,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 95.0162920053117,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1484956443309784,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 205144064,
      "synthesisRealTimeFactor": 0.026393911943563983,
      "synthesisMs": 95.01808299683034,
      "totalMs": 3981.3000000431202,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 867565568,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3871.924958017189,
      "warmupMs": 615.2774170041084,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 96.44245798699558,
      "inferenceMs": 110.84312503226101,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.03078975695340584,
      "audioSecondsPerSecond": 32.47833367159412,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 95.84962500957772,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1484956443309784,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 213532672,
      "synthesisRealTimeFactor": 0.02662535860306687,
      "synthesisMs": 95.85129097104073,
      "totalMs": 3982.76808304945,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 867581952,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3567.6017500227317,
      "warmupMs": 527.9414999531582,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 99.42108299583197,
      "inferenceMs": 112.77637502644211,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.031326770840678364,
      "audioSecondsPerSecond": 31.921579312652373,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 95.86020797723904,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1484956443309784,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 205160448,
      "synthesisRealTimeFactor": 0.026628437497290887,
      "synthesisMs": 95.86237499024719,
      "totalMs": 3680.378125049174,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 867188736,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3567.6017500227317,
      "warmupMs": 527.9414999531582,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 94.81304098153487,
      "inferenceMs": 108.7322499952279,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.030203402776452195,
      "audioSecondsPerSecond": 33.10885225090071,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 94.10766599467024,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1484956443309784,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 205160448,
      "synthesisRealTimeFactor": 0.026141504446665447,
      "synthesisMs": 94.1094160079956,
      "totalMs": 3676.3340000179596,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 870187008,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3567.6017500227317,
      "warmupMs": 527.9414999531582,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 96.12891601864249,
      "inferenceMs": 110.17254198668525,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.030603483885190345,
      "audioSecondsPerSecond": 32.67601831711456,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 95.44900001492351,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1484956443309784,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 205160448,
      "synthesisRealTimeFactor": 0.026514027785095904,
      "synthesisMs": 95.45050002634525,
      "totalMs": 3677.774292009417,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 871464960,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3567.6017500227317,
      "warmupMs": 527.9414999531582,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 96.96745802648365,
      "inferenceMs": 110.59791699517518,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.030721643609770883,
      "audioSecondsPerSecond": 32.55034179492774,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 96.28716600127518,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1484956443309784,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 213532672,
      "synthesisRealTimeFactor": 0.026747141123956278,
      "synthesisMs": 96.2897080462426,
      "totalMs": 3678.199667017907,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 871464960,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3461.4747909945436,
      "warmupMs": 172.52529202960432,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 165.94887501560152,
      "inferenceMs": 179.5263749663718,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.049868437490658835,
      "audioSecondsPerSecond": 20.052763838596633,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 162.26804198231548,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14868450164794922,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04507522000090426,
      "synthesisMs": 162.27079200325534,
      "totalMs": 3641.0011659609154,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 858128384,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3461.4747909945436,
      "warmupMs": 172.52529202960432,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 157.8044579946436,
      "inferenceMs": 171.48187500424683,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.047633854167846344,
      "audioSecondsPerSecond": 20.993472341673687,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 156.98370797326788,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485687494277954,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04360749999048292,
      "synthesisMs": 156.9869999657385,
      "totalMs": 3632.9566659987904,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 858963968,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3461.4747909945436,
      "warmupMs": 172.52529202960432,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 176.90470797242597,
      "inferenceMs": 191.633207956329,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.05323144665453583,
      "audioSecondsPerSecond": 18.78588809524286,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 176.10612499993294,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1486850082874298,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0489191088935412,
      "synthesisMs": 176.10879201674834,
      "totalMs": 3653.1079989508726,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 860356608,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3461.4747909945436,
      "warmupMs": 172.52529202960432,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 169.32408401044086,
      "inferenceMs": 183.21320903487504,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.050892558065243065,
      "audioSecondsPerSecond": 19.649238277982086,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 168.6834170250222,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14861047267913818,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04685716444833411,
      "synthesisMs": 168.6857920140028,
      "totalMs": 3644.6880000294186,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 861159424,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3331.167624972295,
      "warmupMs": 167.15470800409093,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 164.08612503437325,
      "inferenceMs": 177.36670799786225,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.04926852999940618,
      "audioSecondsPerSecond": 20.29693193631011,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 160.32354201888666,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14868450164794922,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04453500000333103,
      "synthesisMs": 160.3260000119917,
      "totalMs": 3508.5343329701573,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 854343680,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3331.167624972295,
      "warmupMs": 167.15470800409093,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 153.48491596523672,
      "inferenceMs": 167.29270899668336,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.046470196943523154,
      "audioSecondsPerSecond": 21.519168537532448,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 152.71787496749312,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485687494277954,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.042422245282472834,
      "synthesisMs": 152.7200830169022,
      "totalMs": 3498.4603339689784,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 856834048,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3331.167624972295,
      "warmupMs": 167.15470800409093,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 154.9963330035098,
      "inferenceMs": 170.14283401658759,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.047261898337941,
      "audioSecondsPerSecond": 21.15869305226824,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 154.36937502818182,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1486850082874298,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.042881088059705995,
      "synthesisMs": 154.37191701494157,
      "totalMs": 3501.3104589888826,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 856834048,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3331.167624972295,
      "warmupMs": 167.15470800409093,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 154.65837495867163,
      "inferenceMs": 168.56258298503235,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.04682293971806454,
      "audioSecondsPerSecond": 21.35705288948773,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 154.05845799250528,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14861047267913818,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0427946874858915,
      "synthesisMs": 154.0608749492094,
      "totalMs": 3499.7302079573274,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 856834048,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Nano · Spanish Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64130 ms / 4477 ms0.033 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64149 ms / 3817 ms0.038 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64158 ms / 9600 ms0.041 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64169 ms / 9600 ms0.043 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64173 ms / 9600 ms0.044 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64178 ms / 9600 ms0.046 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64210 ms / 3817 ms0.054 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64218 ms / 3817 ms0.056 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64273 ms / 3817 ms0.070 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64307 ms / 4477 ms0.076 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642608 ms / 4477 ms0.094 RTF
FP32cpuPassedVery fastmacos · aarch64917 ms / 3351 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64134 ms / 3173 ms0.034 RTF
FP32cpuPassedVery fastmacos · aarch64135 ms / 3173 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64136 ms / 3173 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64138 ms / 3150 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64138 ms / 3173 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64139 ms / 3150 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64141 ms / 3150 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64149 ms / 3351 ms0.037 RTF
FP32cpuPassedVery fastmacos · aarch64144 ms / 3150 ms0.037 RTF
FP32cpuPassedVery fastmacos · aarch64144 ms / 3351 ms0.037 RTF
Q8cpuPassedVery fastmacos · aarch641209 ms / 2698 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64180 ms / 2698 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64181 ms / 2698 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64181 ms / 2698 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64183 ms / 2763 ms0.047 RTF
Q8cpuPassedVery fastmacos · aarch64184 ms / 2763 ms0.047 RTF
Q8cpuPassedVery fastmacos · aarch64184 ms / 2763 ms0.047 RTF
Q8cpuPassedVery fastmacos · aarch64193 ms / 2698 ms0.048 RTF
Q8cpuPassedVery fastmacos · aarch64191 ms / 2763 ms0.049 RTF
Q8cpuPassedVery fastmacos · aarch64192 ms / 2698 ms0.049 RTF
Q8cpuPassedVery fastmacos · aarch64204 ms / 2698 ms0.052 RTF
FP32cpuPassedVery fastmacos · aarch64234 ms / 3422 ms0.058 RTF
FP32cpuPassedVery fastmacos · aarch64226 ms / 3422 ms0.058 RTF
FP32cpuPassedVery fastmacos · aarch641615 ms / 3422 ms0.058 RTF
Q8cpuPassedFastmacos · aarch642976 ms / 2948 ms0.107 RTF
Q8cpuPassedFastmacos · aarch64446 ms / 2948 ms0.110 RTF
Q8cpuPassedFastmacos · aarch64436 ms / 2948 ms0.112 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:01:44.631897+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
54526b8d8e45333cd720d3a082c106aacf416064
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 210.60874999966472,
      "inferenceMs": 225.5701250396669,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.05783849359991459,
      "audioSecondsPerSecond": 17.2895235985447,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 207.80904201092198,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1115528866648674,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05328511769990795,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 217.8856250247918,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 207.811959029641,
      "loadMs": 3421.680749976076,
      "totalMs": 3647.250875015743,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 220.1515409979038,
      "inferenceMs": 234.0273750014603,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.05778453703739761,
      "audioSecondsPerSecond": 17.305667766323186,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 219.4302079733461,
      "voice": "es:elena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06298033893108368,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05418088468205598,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 217.8856250247918,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 219.4325829623267,
      "loadMs": 3421.680749976076,
      "totalMs": 3655.7081249775365,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hola, esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola. Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1526.0692079900764,
      "inferenceMs": 1615.094541979488,
      "audioSeconds": 27.675,
      "realTimeFactor": 0.0583593330435226,
      "audioSecondsPerSecond": 17.135219815727343,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 866.1538749583997,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13557256758213043,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05499615927724052,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 217.8856250247918,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1522.0187079976313,
      "loadMs": 3421.680749976076,
      "totalMs": 5036.775291955564,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 130.2829590276815,
      "inferenceMs": 144.3309159949422,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.037007927178190306,
      "audioSecondsPerSecond": 27.021237779275705,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 127.49470898415893,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1115528866648674,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03269135692300132,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 136.02312497096136,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 127.49629199970514,
      "loadMs": 3350.9173750062473,
      "totalMs": 3495.2482910011895,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 134.80850000632927,
      "inferenceMs": 148.63141597015783,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.03669911505435996,
      "audioSecondsPerSecond": 27.24861344800185,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 134.06829198356718,
      "voice": "es:elena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06298033893108368,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03310403309440539,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 136.02312497096136,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 134.07133403234184,
      "loadMs": 3350.9173750062473,
      "totalMs": 3499.548790976405,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hola, esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola. Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 829.7151669976301,
      "inferenceMs": 916.781457955949,
      "audioSeconds": 27.675,
      "realTimeFactor": 0.033126701281154435,
      "audioSecondsPerSecond": 30.187128851519347,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 439.22745797317475,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13557256758213043,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029834769647265596,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 136.02312497096136,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 825.6772499880753,
      "loadMs": 3350.9173750062473,
      "totalMs": 4267.698832962196,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 115.32895802520216,
      "inferenceMs": 130.11766597628593,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.03336350409648357,
      "audioSecondsPerSecond": 29.972870868362936,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 112.62495798291638,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11163366585969925,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 205766656,
      "synthesisRealTimeFactor": 0.028878621792850587,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1207.2284590103664,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 112.62662499211729,
      "loadMs": 4476.654416997917,
      "totalMs": 4606.772082974203,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 292.4599580001086,
      "inferenceMs": 306.7551660351455,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.07574201630497421,
      "audioSecondsPerSecond": 13.202711635950033,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 291.7972500436008,
      "voice": "es:elena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0629269927740097,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 252018688,
      "synthesisRealTimeFactor": 0.07204911530730718,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1207.2284590103664,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 291.79891699459404,
      "loadMs": 4476.654416997917,
      "totalMs": 4783.4095830330625,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hola, esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola. Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2518.9675000146963,
      "inferenceMs": 2607.9479589825496,
      "audioSeconds": 27.675,
      "realTimeFactor": 0.09423479526585544,
      "audioSecondsPerSecond": 10.611791506298681,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1229.5184999820776,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13561519980430603,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1432256512,
      "synthesisRealTimeFactor": 0.09085531616978879,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1207.2284590103664,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2514.4208749989048,
      "loadMs": 4476.654416997917,
      "totalMs": 7084.602375980467,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 422.0500829978846,
      "inferenceMs": 436.2238749745302,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.11185227563449492,
      "audioSecondsPerSecond": 8.940363477876376,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 420.36941699916497,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11137199401855469,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10778762821036462,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 424.7924169758335,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 420.37175002042204,
      "loadMs": 2947.813374979887,
      "totalMs": 3384.0372499544173,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 433.26404195977375,
      "inferenceMs": 446.21104100951925,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.11017556568136277,
      "audioSecondsPerSecond": 9.076422651571276,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 432.528542005457,
      "voice": "es:elena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06277069449424744,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10679773654340317,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 424.7924169758335,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 432.5308330007829,
      "loadMs": 2947.813374979887,
      "totalMs": 3394.0244159894064,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hola, esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola. Esta es una prueba de voz en este dispositivo. Hola. Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2887.4048329889774,
      "inferenceMs": 2975.8242499665357,
      "audioSeconds": 27.7,
      "realTimeFactor": 0.10743047833814208,
      "audioSecondsPerSecond": 9.308345410624131,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1601.4000829891302,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13525360822677612,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10407692386301416,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 424.7924169758335,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2882.930791005492,
      "loadMs": 2947.813374979887,
      "totalMs": 5923.637624946423,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 179.2523330077529,
      "inferenceMs": 192.1087499940768,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.04925865384463508,
      "audioSecondsPerSecond": 20.30100138655968,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 177.56704200292006,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11137428134679794,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04553055563846078,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 178.8776249741204,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 177.56916698999703,
      "loadMs": 2697.8437090292573,
      "totalMs": 2889.952459023334,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 179.27429196424782,
      "inferenceMs": 192.9966249736026,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.04765348764780311,
      "audioSecondsPerSecond": 20.984822923996443,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 178.5532499779947,
      "voice": "es:elena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06276874244213104,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04408770567093642,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 178.8776249741204,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 178.55520796729252,
      "loadMs": 2697.8437090292573,
      "totalMs": 2890.84033400286,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hola, esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola. Esta es una prueba de voz en este dispositivo. Hola. Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo. Hola, Esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1116.0095839877613,
      "inferenceMs": 1209.2391250189394,
      "audioSeconds": 27.7,
      "realTimeFactor": 0.04365484205844546,
      "audioSecondsPerSecond": 22.906966394728716,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 602.2912089829333,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13525159657001495,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.040125330939085394,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 178.8776249741204,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1111.4716670126654,
      "loadMs": 2697.8437090292573,
      "totalMs": 3907.0828340481967,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3149.784000008367,
      "warmupMs": 132.60670902673155,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 129.95695899007842,
      "inferenceMs": 143.56524997856468,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.03681160255860633,
      "audioSecondsPerSecond": 27.165348164561397,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 126.77954201353714,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1115528866648674,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03250813050470196,
      "synthesisMs": 126.78170896833763,
      "totalMs": 3293.3492499869317,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 648019968,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3149.784000008367,
      "warmupMs": 132.60670902673155,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 127.41858296794817,
      "inferenceMs": 140.87883400497958,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.036122777949994765,
      "audioSecondsPerSecond": 27.683363704317347,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 126.7202909803018,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1115528866648674,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0324927776819095,
      "synthesisMs": 126.72183295944706,
      "totalMs": 3290.6628340133466,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 660832256,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3149.784000008367,
      "warmupMs": 132.60670902673155,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 125.58487500064075,
      "inferenceMs": 139.06974997371435,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.03565891024967035,
      "audioSecondsPerSecond": 28.043481783329163,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 124.88716701045632,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1115528866648674,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03202282051997594,
      "synthesisMs": 124.88900002790615,
      "totalMs": 3288.8537499820814,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 661159936,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3149.784000008367,
      "warmupMs": 132.60670902673155,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 123.91658400883898,
      "inferenceMs": 137.68266700208187,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.03530324794925176,
      "audioSecondsPerSecond": 28.32600562524714,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 123.13341698609293,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1115528866648674,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03157298076145638,
      "synthesisMs": 123.13462496967986,
      "totalMs": 3287.466667010449,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 661159936,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3172.5710409809835,
      "warmupMs": 133.52479203604162,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 125.99850003607571,
      "inferenceMs": 138.3553750347346,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.03547573718839349,
      "audioSecondsPerSecond": 28.188279631498894,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 123.27895901398733,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1115528866648674,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03161039539218809,
      "synthesisMs": 123.28054202953354,
      "totalMs": 3310.926416015718,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 625770496,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3172.5710409809835,
      "warmupMs": 133.52479203604162,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 121.77933304337785,
      "inferenceMs": 134.8577500320971,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.03457891026464028,
      "audioSecondsPerSecond": 28.919361320144912,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 121.15854205330834,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1115528866648674,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0310666666766151,
      "synthesisMs": 121.1600000387989,
      "totalMs": 3307.4287910130806,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 630505472,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3172.5710409809835,
      "warmupMs": 133.52479203604162,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 122.68550001317635,
      "inferenceMs": 135.997082979884,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.03487104691791897,
      "audioSecondsPerSecond": 28.67708567379249,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 122.05704202642664,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1115528866648674,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.031297062060986765,
      "synthesisMs": 122.05854203784838,
      "totalMs": 3308.5681239608675,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 658489344,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3172.5710409809835,
      "warmupMs": 133.52479203604162,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 120.72466703830287,
      "inferenceMs": 134.09549999050796,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.034383461536027685,
      "audioSecondsPerSecond": 29.083750016041286,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 120.11437502223998,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1115528866648674,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.030798953077278268,
      "synthesisMs": 120.11591700138524,
      "totalMs": 3306.6665409714915,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 666206208,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3817.355833016336,
      "warmupMs": 553.045624983497,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 122.09229200379923,
      "inferenceMs": 149.37004097737372,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.0383000105070189,
      "audioSecondsPerSecond": 26.10965341162867,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 119.41637500422075,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11163366585969925,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 205783040,
      "synthesisRealTimeFactor": 0.030620106919788014,
      "synthesisMs": 119.41841698717326,
      "totalMs": 3966.7258739937097,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 697139200,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3817.355833016336,
      "warmupMs": 553.045624983497,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 192.95554200652987,
      "inferenceMs": 218.10945903416723,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.055925502316453136,
      "audioSecondsPerSecond": 17.880930140627502,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 191.0894580068998,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11163366585969925,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214155264,
      "synthesisRealTimeFactor": 0.04899810896649097,
      "synthesisMs": 191.09262496931478,
      "totalMs": 4035.465292050503,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 699547648,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3817.355833016336,
      "warmupMs": 553.045624983497,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 252.8969999984838,
      "inferenceMs": 272.5340839824639,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.06988053435447793,
      "audioSecondsPerSecond": 14.310136710280037,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 251.6102499794215,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11163366585969925,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214155264,
      "synthesisRealTimeFactor": 0.06451614307036671,
      "synthesisMs": 251.61295797443017,
      "totalMs": 4089.8899169988,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 699580416,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3817.355833016336,
      "warmupMs": 553.045624983497,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 192.4840830033645,
      "inferenceMs": 210.38079203572124,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.05394379282967211,
      "audioSecondsPerSecond": 18.537814038354824,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 189.68366598710418,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11163366585969925,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214155264,
      "synthesisRealTimeFactor": 0.0486376923091041,
      "synthesisMs": 189.68700000550598,
      "totalMs": 4027.736625052057,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 699645952,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 9599.615250015631,
      "warmupMs": 1147.8517500218004,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 163.67412504041567,
      "inferenceMs": 177.9273750144057,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.045622403849847615,
      "audioSecondsPerSecond": 21.919055455543255,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 160.54137505125254,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11163366585969925,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 205766656,
      "synthesisRealTimeFactor": 0.041165032053891666,
      "synthesisMs": 160.5436250101775,
      "totalMs": 9777.542625030037,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 631619584,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 9599.615250015631,
      "warmupMs": 1147.8517500218004,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 156.05608298210427,
      "inferenceMs": 173.31779201049358,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.044440459489870146,
      "audioSecondsPerSecond": 22.50201756415102,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 154.24454200547189,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11163366585969925,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214155264,
      "synthesisRealTimeFactor": 0.03955061974314352,
      "synthesisMs": 154.24741699825972,
      "totalMs": 9772.933042026125,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 631685120,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 9599.615250015631,
      "warmupMs": 1147.8517500218004,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 153.74995802994817,
      "inferenceMs": 169.2930000135675,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.043408461541940384,
      "audioSecondsPerSecond": 23.03698321659753,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 153.09137501753867,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11163366585969925,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214155264,
      "synthesisRealTimeFactor": 0.03925468999510392,
      "synthesisMs": 153.0932909809053,
      "totalMs": 9768.908250029199,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 631685120,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 9599.615250015631,
      "warmupMs": 1147.8517500218004,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 143.0773339816369,
      "inferenceMs": 158.39799999957904,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.040614871794763856,
      "audioSecondsPerSecond": 24.621522999093198,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 142.2857089783065,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11163366585969925,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214155264,
      "synthesisRealTimeFactor": 0.03648404923016922,
      "synthesisMs": 142.28779199765995,
      "totalMs": 9758.01325001521,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 631701504,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2697.647124994546,
      "warmupMs": 187.81249999301508,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 191.60087505588308,
      "inferenceMs": 204.45100002689287,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.05242333334022894,
      "audioSecondsPerSecond": 19.07547529475036,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 190.0186249986291,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11137428134679794,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04872332282292728,
      "synthesisMs": 190.0209590094164,
      "totalMs": 2902.098125021439,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 535298048,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2697.647124994546,
      "warmupMs": 187.81249999301508,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 167.72808402311057,
      "inferenceMs": 180.94462499720976,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.046396057691592246,
      "audioSecondsPerSecond": 21.553555404368268,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 167.0249589951709,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11133327335119247,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.042827510773443066,
      "synthesisMs": 167.02729201642796,
      "totalMs": 2878.591749991756,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 537903104,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2697.647124994546,
      "warmupMs": 187.81249999301508,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 167.01299999840558,
      "inferenceMs": 180.45800004620105,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.04627128206312847,
      "audioSecondsPerSecond": 21.611676949769574,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 166.3588749943301,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11142002791166306,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04265674152590621,
      "synthesisMs": 166.36129195103422,
      "totalMs": 2878.105125040747,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 538116096,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2697.647124994546,
      "warmupMs": 187.81249999301508,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 167.7747499779798,
      "inferenceMs": 180.56099995737895,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.04629769229676384,
      "audioSecondsPerSecond": 21.599348701660862,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 167.13041596813127,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11133036762475967,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04285443359269546,
      "synthesisMs": 167.13229101151228,
      "totalMs": 2878.208124951925,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 538116096,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2763.38245900115,
      "warmupMs": 179.6330829965882,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 178.98991599213332,
      "inferenceMs": 191.19162496645004,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.04902349358114103,
      "audioSecondsPerSecond": 20.398383039447282,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 177.3667910019867,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11137428134679794,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04547919870581096,
      "synthesisMs": 177.36887495266274,
      "totalMs": 2954.5740839676,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 534757376,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2763.38245900115,
      "warmupMs": 179.6330829965882,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 170.4272500355728,
      "inferenceMs": 183.5459999856539,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.047063076919398435,
      "audioSecondsPerSecond": 21.24807950216745,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 169.69637502916157,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11133327335119247,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04351248924369709,
      "synthesisMs": 169.69870805041865,
      "totalMs": 2946.928458986804,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 537329664,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2763.38245900115,
      "warmupMs": 179.6330829965882,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 170.19299999810755,
      "inferenceMs": 183.09945799410343,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.04694857897284704,
      "audioSecondsPerSecond": 21.299899206285996,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 169.5016670273617,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11142002791166306,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.043462542816996574,
      "synthesisMs": 169.50391698628664,
      "totalMs": 2946.4819169952534,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 538083328,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2763.38245900115,
      "warmupMs": 179.6330829965882,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 170.57283298345283,
      "inferenceMs": 184.1187080135569,
      "audioSeconds": 3.9,
      "realTimeFactor": 0.047209925131681256,
      "audioSecondsPerSecond": 21.18198656767045,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 169.8569169966504,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11133036762475967,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04355358974769329,
      "synthesisMs": 169.85900001600385,
      "totalMs": 2947.501167014707,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 539983872,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Nano · French Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64122 ms / 3900 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64122 ms / 3900 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64123 ms / 3900 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64123 ms / 4016 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64125 ms / 3900 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64127 ms / 4016 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64127 ms / 4016 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64128 ms / 3655 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64129 ms / 4016 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64332 ms / 3655 ms0.078 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642502 ms / 3655 ms0.084 RTF
FP32cpuPassedVery fastmacos · aarch64953 ms / 3048 ms0.032 RTF
FP32cpuPassedVery fastmacos · aarch64148 ms / 3517 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64148 ms / 3517 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64149 ms / 3048 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64148 ms / 3517 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64149 ms / 3048 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64149 ms / 3517 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64152 ms / 3304 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64156 ms / 3304 ms0.037 RTF
FP32cpuPassedVery fastmacos · aarch64157 ms / 3304 ms0.037 RTF
FP32cpuPassedVery fastmacos · aarch64170 ms / 3304 ms0.040 RTF
Q8cpuPassedVery fastmacos · aarch641203 ms / 2640 ms0.040 RTF
Q8cpuPassedVery fastmacos · aarch64185 ms / 2725 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64186 ms / 2640 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64186 ms / 2648 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64186 ms / 2725 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64186 ms / 2648 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64187 ms / 2640 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64187 ms / 2725 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64190 ms / 2648 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64194 ms / 2648 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64195 ms / 2725 ms0.046 RTF
FP32cpuPassedVery fastmacos · aarch641611 ms / 3182 ms0.054 RTF
FP32cpuPassedVery fastmacos · aarch64232 ms / 3182 ms0.054 RTF
FP32cpuPassedVery fastmacos · aarch64237 ms / 3182 ms0.056 RTF
Q8cpuPassedVery fastmacos · aarch64419 ms / 2949 ms0.099 RTF
Q8cpuPassedVery fastmacos · aarch64422 ms / 2949 ms0.099 RTF
Q8cpuPassedVery fastmacos · aarch642959 ms / 2949 ms0.099 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:03:45.267172+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
54526b8d8e45333cd720d3a082c106aacf416064
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 220.0872499961406,
      "inferenceMs": 236.58725002314895,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.05566758824074093,
      "audioSecondsPerSecond": 17.963774461997243,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 217.3696249956265,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0587020106613636,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05114618610755047,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 224.76283297874033,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 217.3712909570895,
      "loadMs": 3182.432208035607,
      "totalMs": 3419.019458058756,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 217.04645798308775,
      "inferenceMs": 231.5622079768218,
      "audioSeconds": 4.275,
      "realTimeFactor": 0.0541665983571513,
      "audioSecondsPerSecond": 18.46156174339081,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 216.38737502507865,
      "voice": "fr:gabriel",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11565744131803513,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05061729052730384,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 224.76283297874033,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 216.3889170042239,
      "loadMs": 3182.432208035607,
      "totalMs": 3413.9944160124287,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. bonjour, ceci est un test de synthèse vocale locale bonjour, ceci est un test de synthèse vocale locale. bonjour. Ceci est un test de synthèse vocale locale, bonjour, ceci est un test de synthèse vocale locale, bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1514.8795410059392,
      "inferenceMs": 1611.0151660395786,
      "audioSeconds": 29.875,
      "realTimeFactor": 0.053925193842328994,
      "audioSecondsPerSecond": 18.544207795040737,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 978.747375018429,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07802803814411163,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.050569934427551144,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 224.76283297874033,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1510.7767910230905,
      "loadMs": 3182.432208035607,
      "totalMs": 4793.4473740751855,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 133.85908404598013,
      "inferenceMs": 148.6467090435326,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.034975696245537084,
      "audioSecondsPerSecond": 28.591282157180803,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 131.12358399666846,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0587020106613636,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0308529705987038,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 135.06358402082697,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 131.12512504449114,
      "loadMs": 3048.1467499630526,
      "totalMs": 3196.793459006585,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 134.24191699596122,
      "inferenceMs": 149.22379201743752,
      "audioSeconds": 4.275,
      "realTimeFactor": 0.03490615017951755,
      "audioSecondsPerSecond": 28.648246651582514,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 133.50512500619516,
      "voice": "fr:gabriel",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11565744131803513,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03122972726157447,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 135.06358402082697,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 133.50708404323086,
      "loadMs": 3048.1467499630526,
      "totalMs": 3197.37054198049,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. bonjour, ceci est un test de synthèse vocale locale bonjour, ceci est un test de synthèse vocale locale. bonjour. Ceci est un test de synthèse vocale locale, bonjour, ceci est un test de synthèse vocale locale, bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 857.4907080037519,
      "inferenceMs": 952.593624999281,
      "audioSeconds": 29.875,
      "realTimeFactor": 0.03188597907947384,
      "audioSecondsPerSecond": 31.36174672596885,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 523.6119999899529,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07802803814411163,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.028560638760906257,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 135.06358402082697,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 853.2490829820745,
      "loadMs": 3048.1467499630526,
      "totalMs": 4000.7403749623336,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 112.83350002486259,
      "inferenceMs": 127.91091698454693,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.03009668634930516,
      "audioSecondsPerSecond": 33.22624917553713,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 109.69175002537668,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0586746484041214,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214073344,
      "synthesisRealTimeFactor": 0.025810431291842285,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 478.7454580073245,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 109.69433299032971,
      "loadMs": 3654.9472920014523,
      "totalMs": 3782.858208985999,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 316.41308299731463,
      "inferenceMs": 332.12987502338365,
      "audioSeconds": 4.275,
      "realTimeFactor": 0.07769119883587922,
      "audioSecondsPerSecond": 12.871470835615188,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 315.69658301305026,
      "voice": "fr:gabriel",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11582837998867035,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 251936768,
      "synthesisRealTimeFactor": 0.07384761216184288,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 478.7454580073245,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 315.6985419918783,
      "loadMs": 3654.9472920014523,
      "totalMs": 3987.077167024836,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. bonjour, ceci est un test de synthèse vocale locale bonjour, ceci est un test de synthèse vocale locale. bonjour. Ceci est un test de synthèse vocale locale, bonjour, ceci est un test de synthèse vocale locale, bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2404.830041981768,
      "inferenceMs": 2501.9288749899715,
      "audioSeconds": 29.875,
      "realTimeFactor": 0.08374657322142164,
      "audioSecondsPerSecond": 11.940787085771872,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1334.3704579747282,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07802622020244598,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1432076288,
      "synthesisRealTimeFactor": 0.08032325522927372,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 478.7454580073245,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2399.6572499745525,
      "loadMs": 3654.9472920014523,
      "totalMs": 6156.876166991424,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 404.1353749926202,
      "inferenceMs": 418.7761669745669,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.09853556869989809,
      "audioSecondsPerSecond": 10.14861956138519,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 402.4527500150725,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.058627430349588394,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09469530399998322,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 408.3685420337133,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 402.45504199992865,
      "loadMs": 2948.908541991841,
      "totalMs": 3367.684708966408,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 408.24312501354143,
      "inferenceMs": 421.9186659902334,
      "audioSeconds": 4.275,
      "realTimeFactor": 0.09869442479303706,
      "audioSecondsPerSecond": 10.132284595578803,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 407.5232499744743,
      "voice": "fr:gabriel",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11584766209125519,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09532750479343255,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 408.3685420337133,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 407.52508299192414,
      "loadMs": 2948.908541991841,
      "totalMs": 3370.8272079820745,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. bonjour ceci est un test de synthèse vocale locale. bonjour, ceci est un test de synthèse vocale locale bonjour, ceci est un test de synthèse vocale locale. bonjour. Ceci est un test de synthèse vocale locale, bonjour, ceci est un test de synthèse vocale locale, bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2865.587999986019,
      "inferenceMs": 2959.0676249936223,
      "audioSeconds": 29.875,
      "realTimeFactor": 0.09904828870271538,
      "audioSecondsPerSecond": 10.096085587116107,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1904.5882080099545,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07799085229635239,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09576443514580132,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 408.3685420337133,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2860.9624999808148,
      "loadMs": 2948.908541991841,
      "totalMs": 5907.976166985463,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 172.96429100679234,
      "inferenceMs": 186.5211660042405,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.04388733317746835,
      "audioSecondsPerSecond": 22.785617799018997,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 171.26245802501217,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.058627303689718246,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.040297548944020975,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 176.2200419907458,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 171.26458301208913,
      "loadMs": 2639.8392079863697,
      "totalMs": 2826.3603739906102,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 172.05550003563985,
      "inferenceMs": 185.90687500545755,
      "audioSeconds": 4.275,
      "realTimeFactor": 0.043486988305370185,
      "audioSecondsPerSecond": 22.99538411301089,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 171.33383301552385,
      "voice": "fr:gabriel",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11583586782217026,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0400786451472525,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 176.2200419907458,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 171.33620800450444,
      "loadMs": 2639.8392079863697,
      "totalMs": 2825.7460829918273,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. Bonjour, ceci est un test de synthèse vocale locale. bonjour ceci est un test de synthèse vocale locale. bonjour, ceci est un test de synthèse vocale locale bonjour, ceci est un test de synthèse vocale locale. bonjour. Ceci est un test de synthèse vocale locale, bonjour, ceci est un test de synthèse vocale locale, bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1110.486292047426,
      "inferenceMs": 1203.3066670410335,
      "audioSeconds": 29.875,
      "realTimeFactor": 0.04027804743233585,
      "audioSecondsPerSecond": 24.82741999050292,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 701.7667920445092,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07798999547958374,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03701624267824213,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 176.2200419907458,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1105.8602500124834,
      "loadMs": 2639.8392079863697,
      "totalMs": 3843.1458750274032,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3516.6706669842824,
      "warmupMs": 140.78337501268834,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 134.84870898537338,
      "inferenceMs": 149.2687090067193,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.0351220491780516,
      "audioSecondsPerSecond": 28.472142810645515,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 131.90966699039564,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0587020106613636,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.031038009876604464,
      "synthesisMs": 131.91154197556898,
      "totalMs": 3665.9393759910017,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 632455168,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3516.6706669842824,
      "warmupMs": 140.78337501268834,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 133.14283301588148,
      "inferenceMs": 147.94537500711158,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.03481067647226155,
      "audioSecondsPerSecond": 28.726818934324285,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 132.34441698295996,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0587020106613636,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.031140313648125705,
      "synthesisMs": 132.34633300453424,
      "totalMs": 3664.616041991394,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 638058496,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3516.6706669842824,
      "warmupMs": 140.78337501268834,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 133.1774169811979,
      "inferenceMs": 148.45062501262873,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.03492955882650088,
      "audioSecondsPerSecond": 28.629047534413893,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 132.49699998414144,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0587020106613636,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.031176186358446583,
      "synthesisMs": 132.49879202339798,
      "totalMs": 3665.121291996911,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642334720,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3516.6706669842824,
      "warmupMs": 140.78337501268834,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 132.14733300264925,
      "inferenceMs": 148.33795896265656,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.03490304916768389,
      "audioSecondsPerSecond": 28.65079194644925,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 131.42712495755404,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0587020106613636,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03092436258242849,
      "synthesisMs": 131.42854097532108,
      "totalMs": 3665.008625946939,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642793472,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3303.957042051479,
      "warmupMs": 157.05466602230445,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 154.3890000320971,
      "inferenceMs": 169.9815000174567,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.03999564706293099,
      "audioSecondsPerSecond": 25.002720881763807,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 151.4767500339076,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0587020106613636,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.035642009885872114,
      "synthesisMs": 151.47854201495647,
      "totalMs": 3473.938542068936,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 639238144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3303.957042051479,
      "warmupMs": 157.05466602230445,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 142.16254197526723,
      "inferenceMs": 157.23758295644075,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.03699707834269194,
      "audioSecondsPerSecond": 27.02916134991321,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 141.45391702186316,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0587020106613636,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03328366658431204,
      "synthesisMs": 141.45558298332617,
      "totalMs": 3461.19462500792,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643481600,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3303.957042051479,
      "warmupMs": 157.05466602230445,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 136.48383296094835,
      "inferenceMs": 151.54429204994813,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.035657480482340735,
      "audioSecondsPerSecond": 28.044606250159685,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 135.69220795761794,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0587020106613636,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.031928009874023056,
      "synthesisMs": 135.694041964598,
      "totalMs": 3455.5013341014273,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 647757824,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3303.957042051479,
      "warmupMs": 157.05466602230445,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 140.51462500356138,
      "inferenceMs": 156.43729094881564,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.0368087743408978,
      "audioSecondsPerSecond": 27.16743542555047,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 139.83954099239781,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0587020106613636,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03290379411705276,
      "synthesisMs": 139.84112499747425,
      "totalMs": 3460.394333000295,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 648282112,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 4016.479458019603,
      "warmupMs": 571.1968330433592,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 114.12870802450925,
      "inferenceMs": 129.3098750174977,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.030425852945293572,
      "audioSecondsPerSecond": 32.8667860782087,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 110.86529103340581,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0586746484041214,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214073344,
      "synthesisRealTimeFactor": 0.026086411767584437,
      "synthesisMs": 110.86725001223385,
      "totalMs": 4145.789333037101,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694714368,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 4016.479458019603,
      "warmupMs": 571.1968330433592,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 112.02583403792232,
      "inferenceMs": 127.49929202254862,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.029999833417070264,
      "audioSecondsPerSecond": 33.333518426505265,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 111.31925001973286,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0586746484041214,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214073344,
      "synthesisRealTimeFactor": 0.026193235302343965,
      "synthesisMs": 111.32125003496185,
      "totalMs": 4143.978750042152,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694845440,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 4016.479458019603,
      "warmupMs": 571.1968330433592,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 111.26624996541068,
      "inferenceMs": 127.13562499266118,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.02991426470415557,
      "audioSecondsPerSecond": 33.42886779567355,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 110.58237496763468,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0586746484041214,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214056960,
      "synthesisRealTimeFactor": 0.02601974517764414,
      "synthesisMs": 110.5839170049876,
      "totalMs": 4143.615083012264,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694927360,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 4016.479458019603,
      "warmupMs": 571.1968330433592,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 107.98437497578561,
      "inferenceMs": 123.47012496320531,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.029051794108989483,
      "audioSecondsPerSecond": 34.421282081527984,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 106.92395799560472,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0586746484041214,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214056960,
      "synthesisRealTimeFactor": 0.025159137172843602,
      "synthesisMs": 106.92633298458531,
      "totalMs": 4139.949582982808,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694960128,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3900.3607910126448,
      "warmupMs": 546.2246659444645,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 107.78287495486438,
      "inferenceMs": 121.98616599198431,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.0287026272922316,
      "audioSecondsPerSecond": 34.84001620543814,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 104.99149997485802,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0586746484041214,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214073344,
      "synthesisRealTimeFactor": 0.024704352938844,
      "synthesisMs": 104.993499990087,
      "totalMs": 4022.346957004629,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695320576,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3900.3607910126448,
      "warmupMs": 546.2246659444645,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 107.16679197503254,
      "inferenceMs": 122.51474999357015,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.028826999998487094,
      "audioSecondsPerSecond": 34.68970062970418,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 106.42316698795184,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0586746484041214,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214073344,
      "synthesisRealTimeFactor": 0.025041205876999918,
      "synthesisMs": 106.42512497724965,
      "totalMs": 4022.875541006215,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695418880,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3900.3607910126448,
      "warmupMs": 546.2246659444645,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 106.11795802833512,
      "inferenceMs": 121.90204201033339,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.028682833414196092,
      "audioSecondsPerSecond": 34.86405912412637,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 105.43412499828264,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0586746484041214,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214056960,
      "synthesisRealTimeFactor": 0.02480846070958411,
      "synthesisMs": 105.43595801573247,
      "totalMs": 4022.262833022978,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695435264,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3900.3607910126448,
      "warmupMs": 546.2246659444645,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 108.87733299750835,
      "inferenceMs": 124.54158300533891,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.029303901883609155,
      "audioSecondsPerSecond": 34.12514838371541,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 108.20341698126867,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0586746484041214,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 214056960,
      "synthesisRealTimeFactor": 0.025460088228368583,
      "synthesisMs": 108.20537497056648,
      "totalMs": 4024.9023740179837,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695451648,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2724.5545419864357,
      "warmupMs": 184.8093749722466,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 181.36216601124033,
      "inferenceMs": 195.35608298610896,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.04596613717320211,
      "audioSecondsPerSecond": 21.75514545048593,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 179.55891601741314,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.058627303689718246,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04224968612577547,
      "synthesisMs": 179.56116603454575,
      "totalMs": 2919.9106249725446,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 532611072,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2724.5545419864357,
      "warmupMs": 184.8093749722466,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 170.53399997530505,
      "inferenceMs": 184.61408402072266,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.04343860800487592,
      "audioSecondsPerSecond": 23.020995513662672,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 169.7633329895325,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05862269178032875,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03994488235398689,
      "synthesisMs": 169.76575000444427,
      "totalMs": 2909.1686260071583,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 534265856,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2724.5545419864357,
      "warmupMs": 184.8093749722466,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 172.29800001950935,
      "inferenceMs": 186.59116700291634,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.0439038040006862,
      "audioSecondsPerSecond": 22.77706961301965,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 171.61008302355185,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.058633241802453995,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0403792842373471,
      "synthesisMs": 171.6119580087252,
      "totalMs": 2911.145708989352,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 534855680,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2724.5545419864357,
      "warmupMs": 184.8093749722466,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 171.34133400395513,
      "inferenceMs": 185.734250000678,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.04370217647074776,
      "audioSecondsPerSecond": 22.88215555280992,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 170.62683403491974,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.058624349534511566,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04014794117606738,
      "synthesisMs": 170.62874999828637,
      "totalMs": 2910.2887919871137,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 535871488,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2647.642583993729,
      "warmupMs": 181.9829999585636,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 180.0958340172656,
      "inferenceMs": 194.08112502424046,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.045666147064527166,
      "audioSecondsPerSecond": 21.8980593783614,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 178.3500419696793,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.058627303689718246,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.041965245173367506,
      "synthesisMs": 178.3522919868119,
      "totalMs": 2841.7237090179697,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 533512192,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2647.642583993729,
      "warmupMs": 181.9829999585636,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 172.0680840080604,
      "inferenceMs": 186.25233304919675,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.043824078364516884,
      "audioSecondsPerSecond": 22.818506111692056,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 171.2953340029344,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05862269178032875,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04030538235297974,
      "synthesisMs": 171.2978750001639,
      "totalMs": 2833.894917042926,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 534331392,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2647.642583993729,
      "warmupMs": 181.9829999585636,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 171.3990420103073,
      "inferenceMs": 185.61462499201298,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.043674029409885406,
      "audioSecondsPerSecond": 22.896902656150495,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 170.50924996146932,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.058633241802453995,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04012040188536048,
      "synthesisMs": 170.51170801278204,
      "totalMs": 2833.257208985742,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 535511040,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale,",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2647.642583993729,
      "warmupMs": 181.9829999585636,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 175.5276249605231,
      "inferenceMs": 189.96966700069606,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.044698745176634365,
      "audioSecondsPerSecond": 22.371992682307685,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 174.84958295244724,
      "voice": "fr:juliette",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.058624349534511566,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04114152940318865,
      "synthesisMs": 174.85149996355176,
      "totalMs": 2837.6122509944253,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 535511040,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Nano · Hindi Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64140 ms / 4340 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64144 ms / 3885 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64144 ms / 3885 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64146 ms / 3885 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64147 ms / 3911 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64148 ms / 3911 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64148 ms / 3885 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64153 ms / 3911 ms0.031 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64155 ms / 3911 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642590 ms / 4340 ms0.072 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64856 ms / 4340 ms0.185 RTF
FP32cpuPassedVery fastmacos · aarch641161 ms / 3271 ms0.032 RTF
FP32cpuPassedVery fastmacos · aarch64171 ms / 3233 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64171 ms / 3233 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64173 ms / 3271 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64164 ms / 3271 ms0.035 RTF
FP32cpuPassedVery fastmacos · aarch64174 ms / 3382 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64177 ms / 3382 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64177 ms / 3382 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64178 ms / 3382 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64183 ms / 3233 ms0.038 RTF
FP32cpuPassedVery fastmacos · aarch64184 ms / 3233 ms0.038 RTF
Q8cpuPassedVery fastmacos · aarch641475 ms / 2820 ms0.041 RTF
Q8cpuPassedVery fastmacos · aarch64216 ms / 2795 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64216 ms / 2786 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64217 ms / 2786 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64218 ms / 2786 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64219 ms / 2795 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64209 ms / 2820 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64221 ms / 2795 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64221 ms / 2820 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64224 ms / 2795 ms0.046 RTF
Q8cpuPassedVery fastmacos · aarch64225 ms / 2786 ms0.046 RTF
FP32cpuPassedVery fastmacos · aarch642004 ms / 3270 ms0.056 RTF
FP32cpuPassedVery fastmacos · aarch64273 ms / 3270 ms0.056 RTF
FP32cpuPassedVery fastmacos · aarch64260 ms / 3270 ms0.056 RTF
Q8cpuPassedFastmacos · aarch643696 ms / 2928 ms0.103 RTF
Q8cpuPassedFastmacos · aarch64501 ms / 2928 ms0.103 RTF
Q8cpuPassedFastmacos · aarch64483 ms / 2928 ms0.104 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:03:21.171608+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
54526b8d8e45333cd720d3a082c106aacf416064
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 256.5366250346415,
      "inferenceMs": 273.0173749732785,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.05600356409708277,
      "audioSecondsPerSecond": 17.856006418921652,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 253.43962502665818,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13052882254123688,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.051988136620284654,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 265.6699169892818,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 253.4421660238877,
      "loadMs": 3269.777624984272,
      "totalMs": 3542.7949999575503,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 244.4225000217557,
      "inferenceMs": 260.18662500428036,
      "audioSeconds": 4.625,
      "realTimeFactor": 0.05625656756849305,
      "audioSecondsPerSecond": 17.77570234413054,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 243.4566660085693,
      "voice": "hi:ananya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16349293291568756,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.052639648658097594,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 265.6699169892818,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 243.45837504370138,
      "loadMs": 3269.777624984272,
      "totalMs": 3529.964249988552,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जांच है नमस्ते.  यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.24242424242424243,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1888.9244580059312,
      "inferenceMs": 2003.548790991772,
      "audioSeconds": 36.025,
      "realTimeFactor": 0.055615511200326775,
      "audioSecondsPerSecond": 17.980595312663862,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 854.6359579777345,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14850902557373047,
      "clippedFraction": 0.00001272264631043257,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05224923547493045,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 265.6699169892818,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1882.2787079843692,
      "loadMs": 3269.777624984272,
      "totalMs": 5273.326415976044,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 155.6845839950256,
      "inferenceMs": 172.5579170160927,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.03539649579817286,
      "audioSecondsPerSecond": 28.25138413988481,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 152.74995897198096,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13052882254123688,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03133367527371798,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 163.41491602361202,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 152.75166695937514,
      "loadMs": 3271.071499970276,
      "totalMs": 3443.6294169863686,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 148.32991699222475,
      "inferenceMs": 164.01754197431728,
      "audioSeconds": 4.625,
      "realTimeFactor": 0.0354632523187713,
      "audioSecondsPerSecond": 28.19820334049517,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 147.57191704120487,
      "voice": "hi:ananya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16349293291568756,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0319078559987247,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 163.41491602361202,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 147.57383399410173,
      "loadMs": 3271.071499970276,
      "totalMs": 3435.089041944593,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जांच है नमस्ते.  यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.24242424242424243,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1047.6458750199527,
      "inferenceMs": 1161.3107500015758,
      "audioSeconds": 36.025,
      "realTimeFactor": 0.032236245662777954,
      "audioSecondsPerSecond": 31.02098211004343,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 464.6882499800995,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14850902557373047,
      "clippedFraction": 0.00001272264631043257,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.028938179514787724,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 163.41491602361202,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1042.4979170202278,
      "loadMs": 3271.071499970276,
      "totalMs": 4432.382249971852,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 122.53241601865739,
      "inferenceMs": 139.64212499558926,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.028644538460633695,
      "audioSecondsPerSecond": 34.910668970083215,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 119.7368330322206,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13054725527763367,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242401280,
      "synthesisRealTimeFactor": 0.024561863189610917,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1141.6268749744631,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 119.73908304935321,
      "loadMs": 4340.352082974277,
      "totalMs": 4479.994207969867,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 840.2368340175599,
      "inferenceMs": 856.1016659950837,
      "audioSeconds": 4.625,
      "realTimeFactor": 0.18510306291785594,
      "audioSecondsPerSecond": 5.402395747734194,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 839.5290420157835,
      "voice": "hi:ananya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1636069267988205,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 247087104,
      "synthesisRealTimeFactor": 0.18152021621731487,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1141.6268749744631,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 839.5310000050813,
      "loadMs": 4340.352082974277,
      "totalMs": 5196.453748969361,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जांच है नमस्ते.  यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.24242424242424243,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2474.8358340002596,
      "inferenceMs": 2589.717584021855,
      "audioSeconds": 36.025,
      "realTimeFactor": 0.0718866782518211,
      "audioSecondsPerSecond": 13.910783253845327,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1163.9536250149831,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1485033929347992,
      "clippedFraction": 0.000011566042100393246,
      "driverAllocatedMemoryBytes": 1446363136,
      "synthesisRealTimeFactor": 0.06854010873043097,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1141.6268749744631,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2469.1574170137756,
      "loadMs": 4340.352082974277,
      "totalMs": 6930.069666996133,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरण पर आवाज बनाने की एक जाँच है.",
      "errorMetric": "wer",
      "errorRate": 0.09090909090909091,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 485.5250410037115,
      "inferenceMs": 501.3104169629514,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.10283290604368234,
      "audioSecondsPerSecond": 9.724513664674715,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 483.7487500044517,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13017864525318146,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09923097436340192,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 492.33208398800343,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 483.75100002158433,
      "loadMs": 2928.2437919755466,
      "totalMs": 3429.554208938498,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 468.74158398713917,
      "inferenceMs": 483.2016250002198,
      "audioSeconds": 4.625,
      "realTimeFactor": 0.10447602702707455,
      "audioSecondsPerSecond": 9.571573771089648,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 467.9668750031851,
      "voice": "hi:ananya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16363508999347687,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10118247762423109,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 492.33208398800343,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 467.9689590120688,
      "loadMs": 2928.2437919755466,
      "totalMs": 3411.4454169757664,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरण पर आवाज बनाने की एक जाँच है।  नमस्ते यह इस उकरन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जांच है नमस्ते.  यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते यह इस उकरन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.24242424242424243,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 3582.456458010711,
      "inferenceMs": 3696.494584030006,
      "audioSeconds": 36,
      "realTimeFactor": 0.10268040511194461,
      "audioSecondsPerSecond": 9.73895651180745,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1657.682457996998,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1481834352016449,
      "clippedFraction": 0.000008101851851851852,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09935619444392312,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 492.33208398800343,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 3576.822999981232,
      "loadMs": 2928.2437919755466,
      "totalMs": 6624.738376005553,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरण पर आवाज बनाने की एक जाँच है.",
      "errorMetric": "wer",
      "errorRate": 0.09090909090909091,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 204.38220899086446,
      "inferenceMs": 221.30787500645965,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.04539648718081224,
      "audioSecondsPerSecond": 22.028136142275578,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 202.5216250331141,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13016736507415771,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04154337621413362,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 205.6713750353083,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 202.5239590439014,
      "loadMs": 2820.383124984801,
      "totalMs": 3041.6909999912605,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 194.11674997536466,
      "inferenceMs": 209.09508399199694,
      "audioSeconds": 4.625,
      "realTimeFactor": 0.0452097478901615,
      "audioSecondsPerSecond": 22.11912356665937,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 193.32579098409042,
      "voice": "hi:ananya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16363732516765594,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.041800594599162404,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 205.6713750353083,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 193.32775002112612,
      "loadMs": 2820.383124984801,
      "totalMs": 3029.4782089767978,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरण पर आवाज बनाने की एक जाँच है।  नमस्ते यह इस उकरन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जांच है नमस्ते.  यह इस उपकरन पर आवाज बनाने की एक जाँच है।  नमस्ते यह इस उकरन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जाँच है।  नमस्ते या इस उक्करन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.24242424242424243,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1360.0707910372876,
      "inferenceMs": 1475.0985419959761,
      "audioSeconds": 36,
      "realTimeFactor": 0.040974959499888226,
      "audioSecondsPerSecond": 24.405149198566697,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 606.5106249880046,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1481834203004837,
      "clippedFraction": 0.000008101851851851852,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03757566202774695,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 205.6713750353083,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1352.7238329988904,
      "loadMs": 2820.383124984801,
      "totalMs": 4295.481666980777,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3382.4457909795456,
      "warmupMs": 178.66954096825793,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 158.33333402406424,
      "inferenceMs": 174.18766603805125,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.035730803289856665,
      "audioSecondsPerSecond": 27.987056207154517,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 155.2537499810569,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13052882254123688,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03184733333257146,
      "synthesisMs": 155.25574999628589,
      "totalMs": 3556.633457017597,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 672989184,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3382.4457909795456,
      "warmupMs": 178.66954096825793,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 160.50220798933879,
      "inferenceMs": 177.77737497817725,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.03646715384167738,
      "audioSecondsPerSecond": 27.421937131192436,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 159.72591598983854,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13052882254123688,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.032764658045310244,
      "synthesisMs": 159.72770797088742,
      "totalMs": 3560.223165957723,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 679559168,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3382.4457909795456,
      "warmupMs": 178.66954096825793,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 159.61425000568852,
      "inferenceMs": 176.88333301339298,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.03628376061813189,
      "audioSecondsPerSecond": 27.560539011501337,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 158.81925000576302,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13052882254123688,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03257864964409517,
      "synthesisMs": 158.82091701496392,
      "totalMs": 3559.3291239929385,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 679919616,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3382.4457909795456,
      "warmupMs": 178.66954096825793,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 159.1332089737989,
      "inferenceMs": 176.59637500764802,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.036224897437466264,
      "audioSecondsPerSecond": 27.60532315450345,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 158.30379200633615,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13052882254123688,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03247294871088786,
      "synthesisMs": 158.30562496557832,
      "totalMs": 3559.0421659871936,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 679985152,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3233.476541005075,
      "warmupMs": 176.7718750052154,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 166.89508297713473,
      "inferenceMs": 183.7864590343088,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.03769978646857616,
      "audioSecondsPerSecond": 26.52534917760153,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 163.97875000257045,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13052882254123688,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03363707692672809,
      "synthesisMs": 163.98075001779944,
      "totalMs": 3417.2630000393838,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 672628736,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3233.476541005075,
      "warmupMs": 176.7718750052154,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 165.8956249593757,
      "inferenceMs": 183.09245799900964,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.03755742728184813,
      "audioSecondsPerSecond": 26.625891930657072,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 165.12883396353573,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13052882254123688,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03387289742628733,
      "synthesisMs": 165.13037495315075,
      "totalMs": 3416.5689990040846,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 682295296,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3233.476541005075,
      "warmupMs": 176.7718750052154,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 153.71137502370402,
      "inferenceMs": 170.80299998633564,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.03503651281770987,
      "audioSecondsPerSecond": 28.541653251933536,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 152.94766705483198,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13052882254123688,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03137414544247664,
      "synthesisMs": 152.94895903207362,
      "totalMs": 3404.2795409914106,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 691912704,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3233.476541005075,
      "warmupMs": 176.7718750052154,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 153.36312499130145,
      "inferenceMs": 170.58408301090822,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.03499160677146835,
      "audioSecondsPerSecond": 28.578281830012603,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 152.60462497826666,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13052882254123688,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.031303829131409146,
      "synthesisMs": 152.60616701561958,
      "totalMs": 3404.060624015983,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 692879360,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3911.305750021711,
      "warmupMs": 570.8635419723578,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 139.30533296661451,
      "inferenceMs": 155.4971250006929,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.03189684615398829,
      "audioSecondsPerSecond": 31.351061956793586,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 136.22233300702646,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13054725527763367,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242417664,
      "synthesisRealTimeFactor": 0.02794350419814388,
      "synthesisMs": 136.2245829659514,
      "totalMs": 4066.802875022404,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695517184,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3911.305750021711,
      "warmupMs": 570.8635419723578,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 129.45137498900294,
      "inferenceMs": 147.2322500194423,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.03020148718347534,
      "audioSecondsPerSecond": 33.11095224963449,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 128.5775420255959,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13054725527763367,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242401280,
      "synthesisRealTimeFactor": 0.02637532491905567,
      "synthesisMs": 128.5797089803964,
      "totalMs": 4058.5380000411533,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695533568,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3911.305750021711,
      "warmupMs": 570.8635419723578,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 130.08812500629574,
      "inferenceMs": 147.7108330000192,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.03029965805128599,
      "audioSecondsPerSecond": 33.00367279087355,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 129.28812502650544,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13054725527763367,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242401280,
      "synthesisRealTimeFactor": 0.02652103426412512,
      "synthesisMs": 129.29004203760996,
      "totalMs": 4059.01658302173,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695599104,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3911.305750021711,
      "warmupMs": 570.8635419723578,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 135.56274998700246,
      "inferenceMs": 152.8362919925712,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.031351034254886403,
      "audioSecondsPerSecond": 31.896874338177188,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 134.7405419801362,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13054725527763367,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242401280,
      "synthesisRealTimeFactor": 0.027639760611913144,
      "synthesisMs": 134.74383298307657,
      "totalMs": 4064.142042014282,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695631872,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3884.9585420102812,
      "warmupMs": 580.9888749499805,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 129.63616603519768,
      "inferenceMs": 145.61199996387586,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.029869128197718125,
      "audioSecondsPerSecond": 33.47938357559414,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 126.52470800094306,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13054725527763367,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242417664,
      "synthesisRealTimeFactor": 0.025954196516137857,
      "synthesisMs": 126.52670801617205,
      "totalMs": 4030.570541974157,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695992320,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3884.9585420102812,
      "warmupMs": 580.9888749499805,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 131.1880419962108,
      "inferenceMs": 148.2135829865001,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.03040278625364105,
      "audioSecondsPerSecond": 32.891722214448016,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 130.0192920025438,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13054725527763367,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242401280,
      "synthesisRealTimeFactor": 0.026671059895306826,
      "synthesisMs": 130.02141698962077,
      "totalMs": 4033.1721249967813,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 696041472,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3884.9585420102812,
      "warmupMs": 580.9888749499805,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 126.2351669720374,
      "inferenceMs": 143.6609160155058,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.02946890584933452,
      "audioSecondsPerSecond": 33.93407292122393,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 125.42929197661579,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13054725527763367,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242401280,
      "synthesisRealTimeFactor": 0.025729461537244227,
      "synthesisMs": 125.43112499406561,
      "totalMs": 4028.619458025787,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 696074240,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3884.9585420102812,
      "warmupMs": 580.9888749499805,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 126.05091603472829,
      "inferenceMs": 143.65866698790342,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.029468444510339163,
      "audioSecondsPerSecond": 33.9346041712227,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 125.23341603809968,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13054725527763367,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242401280,
      "synthesisRealTimeFactor": 0.025689273439825345,
      "synthesisMs": 125.23520801914856,
      "totalMs": 4028.6172089981847,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 696139776,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरण पर आवाज बनाने की एक जाँच है.",
      "errorMetric": "wer",
      "errorRate": 0.09090909090909091,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2794.717042008415,
      "warmupMs": 211.64054097607732,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 207.52720802556723,
      "inferenceMs": 223.9719579811208,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.04594296573971709,
      "audioSecondsPerSecond": 21.766117704837527,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 205.64162504160777,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13016736507415771,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04218330769202648,
      "synthesisMs": 205.6436249986291,
      "totalMs": 3018.6889999895357,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 534560768,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2794.717042008415,
      "warmupMs": 211.64054097607732,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 204.57870897371322,
      "inferenceMs": 220.9749580360949,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.045328196520224594,
      "audioSecondsPerSecond": 22.061323343270864,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 203.75120895914733,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13014960289001465,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04179553004602591,
      "synthesisMs": 203.75320897437632,
      "totalMs": 3015.6920000445098,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 538001408,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2794.717042008415,
      "warmupMs": 211.64054097607732,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 199.230250029359,
      "inferenceMs": 215.5495419865474,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.04421529066390716,
      "audioSecondsPerSecond": 22.616610339651068,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 198.46891704946756,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13022445142269135,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.040712034258131795,
      "synthesisMs": 198.4711670083925,
      "totalMs": 3010.2665839949623,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 538001408,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2794.717042008415,
      "warmupMs": 211.64054097607732,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 202.5124579668045,
      "inferenceMs": 218.79179100506008,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.04488036738565335,
      "audioSecondsPerSecond": 22.281457533693548,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 201.59270794829354,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13015802204608917,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.041352709325460285,
      "synthesisMs": 201.5944579616189,
      "totalMs": 3013.508833013475,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 539885568,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरण पर आवाज बनाने की एक जाँच है.",
      "errorMetric": "wer",
      "errorRate": 0.09090909090909091,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2786.220583016984,
      "warmupMs": 208.2662910106592,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 209.27745901281014,
      "inferenceMs": 225.22670903708786,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.04620035057171033,
      "audioSecondsPerSecond": 21.644857400980975,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 207.34445902053267,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13016736507415771,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.042532624008181766,
      "synthesisMs": 207.34654203988612,
      "totalMs": 3011.447292054072,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 532955136,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2786.220583016984,
      "warmupMs": 208.2662910106592,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 199.40062501700595,
      "inferenceMs": 215.98637499846518,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.0443048974355826,
      "audioSecondsPerSecond": 22.570868185711447,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 198.55016696965322,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13014960289001465,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04072867528511546,
      "synthesisMs": 198.55229201493785,
      "totalMs": 3002.2069580154493,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 536772608,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2786.220583016984,
      "warmupMs": 208.2662910106592,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 200.78579196706414,
      "inferenceMs": 217.21508301561698,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.044556940105767585,
      "audioSecondsPerSecond": 22.443192858985327,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 200.00379194971174,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13022445142269135,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.041026820512249686,
      "synthesisMs": 200.0057499972172,
      "totalMs": 3003.435666032601,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 537575424,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जाँच है।",
      "errorMetric": "wer",
      "errorRate": 0.18181818181818182,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2786.220583016984,
      "warmupMs": 208.2662910106592,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 200.96891705179587,
      "inferenceMs": 217.67333301249892,
      "audioSeconds": 4.875,
      "realTimeFactor": 0.04465094010512798,
      "audioSecondsPerSecond": 22.395945027037715,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 200.1850830274634,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13015802204608917,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04106399138720754,
      "synthesisMs": 200.18695801263675,
      "totalMs": 3003.893916029483,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 538886144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Nano · Italian Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64122 ms / 3692 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64123 ms / 3692 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64124 ms / 3692 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64125 ms / 3716 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64125 ms / 3716 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64128 ms / 3716 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64129 ms / 3692 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64130 ms / 3716 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64133 ms / 4453 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642609 ms / 4453 ms0.089 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64910 ms / 4453 ms0.201 RTF
FP32cpuPassedVery fastmacos · aarch64949 ms / 3455 ms0.032 RTF
FP32cpuPassedVery fastmacos · aarch64149 ms / 3455 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64151 ms / 3455 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64151 ms / 4365 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64151 ms / 4365 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64153 ms / 3291 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64153 ms / 4365 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64154 ms / 3291 ms0.034 RTF
FP32cpuPassedVery fastmacos · aarch64155 ms / 3291 ms0.034 RTF
FP32cpuPassedVery fastmacos · aarch64155 ms / 3291 ms0.034 RTF
FP32cpuPassedVery fastmacos · aarch64156 ms / 4365 ms0.034 RTF
Q8cpuPassedVery fastmacos · aarch64194 ms / 2615 ms0.042 RTF
Q8cpuPassedVery fastmacos · aarch641252 ms / 2839 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64196 ms / 2615 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64197 ms / 2759 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64198 ms / 2615 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64199 ms / 2759 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64198 ms / 2839 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64201 ms / 2759 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64204 ms / 2759 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64207 ms / 2615 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64211 ms / 2839 ms0.046 RTF
FP32cpuPassedVery fastmacos · aarch64250 ms / 3323 ms0.055 RTF
FP32cpuPassedVery fastmacos · aarch64248 ms / 3323 ms0.055 RTF
FP32cpuPassedVery fastmacos · aarch641663 ms / 3323 ms0.057 RTF
Q8cpuPassedFastmacos · aarch64471 ms / 3020 ms0.103 RTF
Q8cpuPassedFastmacos · aarch64467 ms / 3020 ms0.103 RTF
Q8cpuPassedFastmacos · aarch643099 ms / 3020 ms0.106 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:02:09.137571+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
54526b8d8e45333cd720d3a082c106aacf416064
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 234.73612504312769,
      "inferenceMs": 250.2760419738479,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.05470514578663342,
      "audioSecondsPerSecond": 18.279816014023652,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 232.02779202256352,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06427944451570511,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.050716821646287304,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 242.2922920086421,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 232.02945903176442,
      "loadMs": 3323.11545900302,
      "totalMs": 3573.391500976868,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 232.85479203332216,
      "inferenceMs": 248.16275003831834,
      "audioSeconds": 4.525,
      "realTimeFactor": 0.054842596693551016,
      "audioSecondsPerSecond": 18.2340016755186,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 232.15070803416893,
      "voice": "it:alessandro",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.209901362657547,
      "clippedFraction": 0.0017587476979742172,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0513044015494129,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 242.2922920086421,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 232.15241701109335,
      "loadMs": 3323.11545900302,
      "totalMs": 3571.2782090413384,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, Buongiorno. questo è un test di sintesi vocale locale. Buongiorno. questo è un test di sintesi vocale locale, Buongiorno. Questo è un test di sintesi vocale locale, buongiorno. questo è un test di sintesi vocale locale. Buongiorno. Questo è un test di sintesi vocale locale. Buongiorno. Questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1552.6862500119023,
      "inferenceMs": 1662.8480410436168,
      "audioSeconds": 29.275,
      "realTimeFactor": 0.056800957849483066,
      "audioSecondsPerSecond": 17.60533691438622,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 840.8784170169383,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06656769663095474,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.052844192997737766,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 242.2922920086421,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1547.013750008773,
      "loadMs": 3323.11545900302,
      "totalMs": 4985.963500046637,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 135.33891597762704,
      "inferenceMs": 151.03524998994544,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.03301316939670939,
      "audioSecondsPerSecond": 30.290942017208316,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 132.5249999645166,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06427944451570511,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.028967513662096257,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 142.11354102008045,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 132.52637500409037,
      "loadMs": 3454.5612080255523,
      "totalMs": 3605.596458015498,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 133.62300000153482,
      "inferenceMs": 148.92837504157797,
      "audioSeconds": 4.525,
      "realTimeFactor": 0.03291234807548684,
      "audioSecondsPerSecond": 30.383733111549134,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 132.8668330097571,
      "voice": "it:alessandro",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.209901362657547,
      "clippedFraction": 0.0017587476979742172,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029363176805991164,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 142.11354102008045,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 132.86837504711002,
      "loadMs": 3454.5612080255523,
      "totalMs": 3603.4895830671303,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, Buongiorno. questo è un test di sintesi vocale locale. Buongiorno. questo è un test di sintesi vocale locale, Buongiorno. Questo è un test di sintesi vocale locale, buongiorno. questo è un test di sintesi vocale locale. Buongiorno. Questo è un test di sintesi vocale locale. Buongiorno. Questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 855.3582499735057,
      "inferenceMs": 949.3937909719534,
      "audioSeconds": 29.275,
      "realTimeFactor": 0.03243018927316664,
      "audioSecondsPerSecond": 30.835466039892008,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 449.23133397242054,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06656769663095474,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029077514943234895,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 142.11354102008045,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 851.2442499632016,
      "loadMs": 3454.5612080255523,
      "totalMs": 4403.954998997506,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 117.47604102129117,
      "inferenceMs": 133.2864590222016,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.02913365224528997,
      "audioSecondsPerSecond": 34.32456705326638,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 114.5849580061622,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06425411254167557,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216612864,
      "synthesisRealTimeFactor": 0.025046229517708223,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1175.3226249711588,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 114.58650004351512,
      "loadMs": 4453.33412499167,
      "totalMs": 4586.620584013872,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 893.8598749809898,
      "inferenceMs": 909.9154580035247,
      "audioSeconds": 4.525,
      "realTimeFactor": 0.20108628906155243,
      "audioSecondsPerSecond": 4.972989479625339,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 893.1490000104532,
      "voice": "it:alessandro",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.21038801968097687,
      "clippedFraction": 0.0017863720073664825,
      "driverAllocatedMemoryBytes": 221134848,
      "synthesisRealTimeFactor": 0.1973814826499856,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1175.3226249711588,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 893.1512089911848,
      "loadMs": 4453.33412499167,
      "totalMs": 5363.249582995195,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, Buongiorno. questo è un test di sintesi vocale locale. Buongiorno. questo è un test di sintesi vocale locale. Buongiorno. Questo è un test di sintesi vocale locale, buongiorno. questo è un test di sintesi vocale locale. Buongiorno. Questo è un test di sintesi vocale locale. Buongiorno. Questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2516.700416977983,
      "inferenceMs": 2608.7955409893766,
      "audioSeconds": 29.275,
      "realTimeFactor": 0.0891134258237191,
      "audioSecondsPerSecond": 11.221653648218656,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1330.3310420014895,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06654387712478638,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1414856704,
      "synthesisRealTimeFactor": 0.0857851537140158,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1175.3226249711588,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2511.360374977812,
      "loadMs": 4453.33412499167,
      "totalMs": 7062.129665981047,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 454.76650004275143,
      "inferenceMs": 470.770874992013,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.10290073770317223,
      "audioSecondsPerSecond": 9.718103313161883,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 452.9842500342056,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06453680992126465,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09901350645267003,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 465.9844169509597,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 452.98679202096537,
      "loadMs": 3019.648541987408,
      "totalMs": 3490.419416979421,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 452.16600003186613,
      "inferenceMs": 467.026749975048,
      "audioSeconds": 4.525,
      "realTimeFactor": 0.10321033148619846,
      "audioSecondsPerSecond": 9.688952506985432,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 451.40187500510365,
      "voice": "it:alessandro",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.21022053062915802,
      "clippedFraction": 0.001721915285451197,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09975777171006765,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 465.9844169509597,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 451.40391698805615,
      "loadMs": 3019.648541987408,
      "totalMs": 3486.675291962456,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, Buongiorno. questo è un test di sintesi vocale locale. Buongiorno. questo è un test di sintesi vocale locale, Buongiorno. Questo è un test di sintesi vocale locale, buongiorno. questo è un test di sintesi vocale locale. Buongiorno. Questo è un test di sintesi vocale locale. Buongiorno. questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 3005.377750028856,
      "inferenceMs": 3098.8701669848524,
      "audioSeconds": 29.25,
      "realTimeFactor": 0.10594427921315734,
      "audioSecondsPerSecond": 9.43892400256954,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1678.781125054229,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06654363870620728,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.1025808091299274,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 465.9844169509597,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 3000.4886670503765,
      "loadMs": 3019.648541987408,
      "totalMs": 6118.51870897226,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 196.25891698524356,
      "inferenceMs": 211.02620800957084,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.04612594710591712,
      "audioSecondsPerSecond": 21.67977164140914,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 194.52166702831164,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06453870236873627,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04251898907613559,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 210.55041695944965,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 194.52437502332032,
      "loadMs": 2839.2971249995753,
      "totalMs": 3050.323333009146,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 183.4862909745425,
      "inferenceMs": 198.23270797496662,
      "audioSeconds": 4.525,
      "realTimeFactor": 0.043808333254136265,
      "audioSecondsPerSecond": 22.826707288745858,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 182.66241595847532,
      "voice": "it:alessandro",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.21022264659404755,
      "clippedFraction": 0.001721915285451197,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04036784529665392,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 210.55041695944965,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 182.664499967359,
      "loadMs": 2839.2971249995753,
      "totalMs": 3037.529832974542,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, Buongiorno. questo è un test di sintesi vocale locale. Buongiorno. questo è un test di sintesi vocale locale, Buongiorno. Questo è un test di sintesi vocale locale, buongiorno. questo è un test di sintesi vocale locale. Buongiorno. Questo è un test di sintesi vocale locale. Buongiorno. questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1153.6636659875512,
      "inferenceMs": 1252.417791984044,
      "audioSeconds": 29.25,
      "realTimeFactor": 0.042817702290052786,
      "audioSecondsPerSecond": 23.354826310526136,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 616.5252079954371,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06654433906078339,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.039278263520672284,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 210.55041695944965,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1148.8892079796642,
      "loadMs": 2839.2971249995753,
      "totalMs": 4091.7149169836193,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 4365.138666005805,
      "warmupMs": 170.53966701496392,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 140.583542000968,
      "inferenceMs": 155.80737503478304,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.034056256838203944,
      "audioSecondsPerSecond": 29.36318000979517,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 137.69241701811552,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06427944451570511,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.030097040001689737,
      "synthesisMs": 137.69395800773054,
      "totalMs": 4520.946041040588,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644923392,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 4365.138666005805,
      "warmupMs": 170.53966701496392,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 135.49849996343255,
      "inferenceMs": 151.1683330172673,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.03304225858300924,
      "audioSecondsPerSecond": 30.264274988581228,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 134.74637496983632,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06427944451570511,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029453133115840087,
      "synthesisMs": 134.7480840049684,
      "totalMs": 4516.306999023072,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 649150464,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 4365.138666005805,
      "warmupMs": 170.53966701496392,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 137.89066701428965,
      "inferenceMs": 153.10629102168605,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.03346585596102427,
      "audioSecondsPerSecond": 29.881201938018307,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 136.8855420150794,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06427944451570511,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029920710383054338,
      "synthesisMs": 136.8872500024736,
      "totalMs": 4518.244957027491,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 673710080,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 4365.138666005805,
      "warmupMs": 170.53966701496392,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 135.5950830038637,
      "inferenceMs": 151.29587502451614,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.03307013661738058,
      "audioSecondsPerSecond": 30.23876228785922,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 134.88108303863555,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06427944451570511,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029482531801497187,
      "synthesisMs": 134.88258299184963,
      "totalMs": 4516.434541030321,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 682360832,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3291.030167019926,
      "warmupMs": 142.91929197497666,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 138.14579200698063,
      "inferenceMs": 153.5774590447545,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.03356884350705017,
      "audioSecondsPerSecond": 29.789527893327005,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 135.28341701021418,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06427944451570511,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02957048283920425,
      "synthesisMs": 135.28495898935944,
      "totalMs": 3444.6076260646805,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644988928,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3291.030167019926,
      "warmupMs": 142.91929197497666,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 137.54016702296212,
      "inferenceMs": 153.08729198295623,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.03346170316567349,
      "audioSecondsPerSecond": 29.88491037198144,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 136.79162500193343,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06427944451570511,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029900118471962996,
      "synthesisMs": 136.7930420092307,
      "totalMs": 3444.1174590028822,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 645431296,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3291.030167019926,
      "warmupMs": 142.91929197497666,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 139.2526249983348,
      "inferenceMs": 154.74179096054286,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.033823342286457454,
      "audioSecondsPerSecond": 29.565380958829444,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 138.42058298178017,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06427944451570511,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.030256338796675043,
      "synthesisMs": 138.42274999478832,
      "totalMs": 3445.771957980469,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 647954432,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3291.030167019926,
      "warmupMs": 142.91929197497666,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 138.56287498492748,
      "inferenceMs": 154.53279099892825,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.03377765923473842,
      "audioSecondsPerSecond": 29.60536705786754,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 137.8015410155058,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06427944451570511,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.030120846998618275,
      "synthesisMs": 137.8028750186786,
      "totalMs": 3445.5629580188543,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 648151040,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3691.965790989343,
      "warmupMs": 519.7630830225535,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 112.89158399449661,
      "inferenceMs": 128.8298750296235,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.028159535525600767,
      "audioSecondsPerSecond": 35.511949374692875,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 110.23158399621025,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06425411254167557,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216629248,
      "synthesisRealTimeFactor": 0.02409481792074572,
      "synthesisMs": 110.23379198741168,
      "totalMs": 3820.7956660189666,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694353920,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3691.965790989343,
      "warmupMs": 519.7630830225535,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 108.87104202993214,
      "inferenceMs": 124.2050000000745,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.027148633879797705,
      "audioSecondsPerSecond": 36.83426593130112,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 108.17104199668393,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06425411254167557,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216629248,
      "synthesisRealTimeFactor": 0.023644380767631236,
      "synthesisMs": 108.17304201191291,
      "totalMs": 3816.1707909894176,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694353920,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3691.965790989343,
      "warmupMs": 519.7630830225535,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 106.39741702470928,
      "inferenceMs": 121.65320798521861,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.026590865133381117,
      "audioSecondsPerSecond": 37.60689977493961,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 105.74620903935283,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06425411254167557,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216612864,
      "synthesisRealTimeFactor": 0.023114408094110223,
      "synthesisMs": 105.74841703055426,
      "totalMs": 3813.6189989745617,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694403072,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3691.965790989343,
      "warmupMs": 519.7630830225535,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 106.9791250047274,
      "inferenceMs": 122.90562503039837,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.02686461749298325,
      "audioSecondsPerSecond": 37.223682796197984,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 106.33379104547203,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06425411254167557,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216612864,
      "synthesisRealTimeFactor": 0.02324270492908163,
      "synthesisMs": 106.33537505054846,
      "totalMs": 3814.8714160197414,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694452224,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3716.481167008169,
      "warmupMs": 542.1991249895655,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 114.91558403940871,
      "inferenceMs": 129.8774169990793,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.028388506447886184,
      "audioSecondsPerSecond": 35.22552346442517,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 112.1990000247024,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06425411254167557,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216629248,
      "synthesisRealTimeFactor": 0.024524817925170472,
      "synthesisMs": 112.2010420076549,
      "totalMs": 3846.3585840072483,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694353920,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3716.481167008169,
      "warmupMs": 542.1991249895655,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 112.2313339728862,
      "inferenceMs": 127.6492080069147,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.027901466231019606,
      "audioSecondsPerSecond": 35.840410382743414,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 111.53433402068913,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06425411254167557,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216629248,
      "synthesisRealTimeFactor": 0.02437953552221014,
      "synthesisMs": 111.5363750141114,
      "totalMs": 3844.1303750150837,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694353920,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3716.481167008169,
      "warmupMs": 542.1991249895655,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 108.78945799777284,
      "inferenceMs": 124.67766599729657,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.027251948851868102,
      "audioSecondsPerSecond": 36.69462339870239,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 108.14545798348263,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06425411254167557,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216612864,
      "synthesisRealTimeFactor": 0.02363877049062014,
      "synthesisMs": 108.14737499458715,
      "totalMs": 3841.1588330054656,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694403072,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3716.481167008169,
      "warmupMs": 542.1991249895655,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 109.06137496931478,
      "inferenceMs": 124.83779201284051,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.02728694907384492,
      "audioSecondsPerSecond": 36.64755621061791,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 108.22200000984594,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06425411254167557,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216612864,
      "synthesisRealTimeFactor": 0.02365546448593908,
      "synthesisMs": 108.2237500231713,
      "totalMs": 3841.3189590210095,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694435840,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2759.2844999744557,
      "warmupMs": 200.53058303892612,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 190.4440830112435,
      "inferenceMs": 204.35245899716392,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.04466720415238556,
      "audioSecondsPerSecond": 22.387790303338086,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 188.69974999688566,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06453870236873627,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04124633880087831,
      "synthesisMs": 188.70200001401827,
      "totalMs": 2963.6369589716196,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 534790144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2759.2844999744557,
      "warmupMs": 200.53058303892612,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 184.87308302428573,
      "inferenceMs": 200.55854099337012,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.04383793245756724,
      "audioSecondsPerSecond": 22.81129478375711,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 184.07045799540356,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06448883563280106,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04023441705150438,
      "synthesisMs": 184.07245801063254,
      "totalMs": 2959.843040967826,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 537427968,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2759.2844999744557,
      "warmupMs": 200.53058303892612,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 181.3712500152178,
      "inferenceMs": 196.68012502370402,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.042990191262011805,
      "audioSecondsPerSecond": 23.261120051904676,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 180.58958399342373,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0644778162240982,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03947353377286854,
      "synthesisMs": 180.59141701087356,
      "totalMs": 2955.9646249981597,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 537772032,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2759.2844999744557,
      "warmupMs": 200.53058303892612,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 183.64745896542445,
      "inferenceMs": 198.82604101439938,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.04345924393757363,
      "audioSecondsPerSecond": 23.010064359067883,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 182.907999958843,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06459538638591766,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.039980318906202995,
      "synthesisMs": 182.9099589958787,
      "totalMs": 2958.110540988855,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 537772032,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2615.4030000325292,
      "warmupMs": 187.33124999562278,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 193.43029201263562,
      "inferenceMs": 207.00875000329688,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.0452478142083709,
      "audioSecondsPerSecond": 22.10051507449389,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 191.6699589928612,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06453870236873627,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.041895582953905636,
      "synthesisMs": 191.67229201411828,
      "totalMs": 2822.411750035826,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 534151168,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2615.4030000325292,
      "warmupMs": 187.33124999562278,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 183.10029100393876,
      "inferenceMs": 197.95658299699426,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.04326919846928836,
      "audioSecondsPerSecond": 23.111128363280883,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 182.21745797200128,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06448883563280106,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.039829353221737945,
      "synthesisMs": 182.2192909894511,
      "totalMs": 2813.3595830295235,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 536559616,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2615.4030000325292,
      "warmupMs": 187.33124999562278,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 181.6774999606423,
      "inferenceMs": 196.2554999627173,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.0428973770410311,
      "audioSecondsPerSecond": 23.311448600773552,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 180.9591249912046,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0644778162240982,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03955439889620318,
      "synthesisMs": 180.96137495012954,
      "totalMs": 2811.6584999952465,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 536576000,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2615.4030000325292,
      "warmupMs": 187.33124999562278,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 179.9816670245491,
      "inferenceMs": 194.42349998280406,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.042496939886951704,
      "audioSecondsPerSecond": 23.5311060669345,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 179.2679579812102,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06459538638591766,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.039184681315463585,
      "synthesisMs": 179.2699170182459,
      "totalMs": 2809.8265000153333,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 537509888,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Nano · Portuguese Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64118 ms / 3609 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64119 ms / 3582 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64119 ms / 3609 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64119 ms / 3582 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64120 ms / 3609 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64120 ms / 3582 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64123 ms / 3582 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64123 ms / 3609 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64127 ms / 3822 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642655 ms / 3822 ms0.085 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64961 ms / 3822 ms0.219 RTF
FP32cpuPassedVery fastmacos · aarch64981 ms / 3309 ms0.031 RTF
FP32cpuPassedVery fastmacos · aarch64151 ms / 3178 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64151 ms / 3178 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64152 ms / 3178 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64153 ms / 3292 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64154 ms / 3292 ms0.034 RTF
FP32cpuPassedVery fastmacos · aarch64155 ms / 3292 ms0.034 RTF
FP32cpuPassedVery fastmacos · aarch64155 ms / 3178 ms0.034 RTF
FP32cpuPassedVery fastmacos · aarch64157 ms / 3292 ms0.034 RTF
FP32cpuPassedVery fastmacos · aarch64158 ms / 3309 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64178 ms / 3309 ms0.039 RTF
Q8cpuPassedVery fastmacos · aarch641240 ms / 2763 ms0.040 RTF
Q8cpuPassedVery fastmacos · aarch64190 ms / 2785 ms0.041 RTF
Q8cpuPassedVery fastmacos · aarch64191 ms / 2785 ms0.041 RTF
Q8cpuPassedVery fastmacos · aarch64192 ms / 2581 ms0.042 RTF
Q8cpuPassedVery fastmacos · aarch64192 ms / 2581 ms0.042 RTF
Q8cpuPassedVery fastmacos · aarch64192 ms / 2581 ms0.042 RTF
Q8cpuPassedVery fastmacos · aarch64197 ms / 2785 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64198 ms / 2581 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64189 ms / 2763 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64201 ms / 2763 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64203 ms / 2785 ms0.044 RTF
FP32cpuPassedVery fastmacos · aarch64250 ms / 3409 ms0.055 RTF
FP32cpuPassedVery fastmacos · aarch641707 ms / 3409 ms0.055 RTF
FP32cpuPassedVery fastmacos · aarch64241 ms / 3409 ms0.055 RTF
Q8cpuPassedVery fastmacos · aarch64432 ms / 2979 ms0.099 RTF
Q8cpuPassedVery fastmacos · aarch643103 ms / 2979 ms0.099 RTF
Q8cpuPassedFastmacos · aarch64465 ms / 2979 ms0.101 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:02:32.133352+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
54526b8d8e45333cd720d3a082c106aacf416064
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 233.3641669829376,
      "inferenceMs": 249.67912497231737,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.054574672125096695,
      "audioSecondsPerSecond": 18.323518237687043,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 230.57754198089242,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06812528520822525,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05039980874239096,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 259.77674999739975,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 230.57912499643862,
      "loadMs": 3409.4259159755893,
      "totalMs": 3659.1050409479067,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 226.389667019248,
      "inferenceMs": 241.43008299870417,
      "audioSeconds": 4.4,
      "realTimeFactor": 0.0548704734087964,
      "audioSecondsPerSecond": 18.224737966989874,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 225.68116697948426,
      "voice": "pt:larissa",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04571033641695976,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05129166681647555,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 259.77674999739975,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 225.6833339924924,
      "loadMs": 3409.4259159755893,
      "totalMs": 3650.8559989742935,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local, Ol olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local, olá. Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local, olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0.011111111111111112,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1604.387958010193,
      "inferenceMs": 1706.9298750138842,
      "audioSeconds": 31.25,
      "realTimeFactor": 0.05462175600044429,
      "audioSecondsPerSecond": 18.30772339124114,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1015.372916997876,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0765983983874321,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05112801865674555,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 259.77674999739975,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1597.7505830232985,
      "loadMs": 3409.4259159755893,
      "totalMs": 5116.3557909894735,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 161.64850001223385,
      "inferenceMs": 178.3538750023581,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.038984453552428,
      "audioSecondsPerSecond": 25.651250918655464,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 158.44825003296137,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06812528520822525,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03463386141954631,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 160.09562503313646,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 158.44991599442437,
      "loadMs": 3308.621709002182,
      "totalMs": 3486.97558400454,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 143.6552500235848,
      "inferenceMs": 158.40133297024295,
      "audioSeconds": 4.4,
      "realTimeFactor": 0.03600030294778249,
      "audioSecondsPerSecond": 27.777544023739864,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 142.95529201626778,
      "voice": "pt:larissa",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04571033641695976,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03249023682490753,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 160.09562503313646,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 142.95704202959314,
      "loadMs": 3308.621709002182,
      "totalMs": 3467.023041972425,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local, Ol olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local, olá. Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local, olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0.011111111111111112,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 880.4049160098657,
      "inferenceMs": 981.3018749700859,
      "audioSeconds": 31.25,
      "realTimeFactor": 0.03140165999904275,
      "audioSecondsPerSecond": 31.845450209654015,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 528.8553329883143,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0765983983874321,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.028027497312054037,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 160.09562503313646,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 875.8592910016887,
      "loadMs": 3308.621709002182,
      "totalMs": 4289.923583972268,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 112.34270798740909,
      "inferenceMs": 127.41062499117106,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.02784931693796089,
      "audioSecondsPerSecond": 35.90752341350672,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 109.60287495981902,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06808102130889893,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216383488,
      "synthesisRealTimeFactor": 0.023957367864299993,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 546.9212919706479,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 109.60495797917247,
      "loadMs": 3821.8412499991246,
      "totalMs": 3949.2518749902956,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 946.618958027102,
      "inferenceMs": 961.4356250385754,
      "audioSeconds": 4.4,
      "realTimeFactor": 0.2185080965996762,
      "audioSecondsPerSecond": 4.5764894553636495,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 945.9074169863015,
      "voice": "pt:larissa",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.045732688158750534,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 220790784,
      "synthesisRealTimeFactor": 0.21497934659815987,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 546.9212919706479,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 945.9091250319034,
      "loadMs": 3821.8412499991246,
      "totalMs": 4783.2768750377,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local, Ol olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local, olá. Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local, olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0.011111111111111112,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2555.5137910414487,
      "inferenceMs": 2654.6945000300184,
      "audioSeconds": 31.25,
      "realTimeFactor": 0.08495022400096058,
      "audioSecondsPerSecond": 11.77159933078802,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1399.711208010558,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07656852155923843,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1419968512,
      "synthesisRealTimeFactor": 0.08160641331225633,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 546.9212919706479,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2550.2004160080105,
      "loadMs": 3821.8412499991246,
      "totalMs": 6476.535750029143,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 449.5260830153711,
      "inferenceMs": 464.9082499672659,
      "audioSeconds": 4.6,
      "realTimeFactor": 0.1010670108624491,
      "audioSecondsPerSecond": 9.894425406139566,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 447.61929102241993,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06788164377212524,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09730904892026244,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 447.0907910144888,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 447.62162503320724,
      "loadMs": 2979.361500008963,
      "totalMs": 3444.269749976229,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 418.64150000037625,
      "inferenceMs": 432.30112496530637,
      "audioSeconds": 4.375,
      "realTimeFactor": 0.09881168570635575,
      "audioSecondsPerSecond": 10.120260502100495,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 417.93804097687826,
      "voice": "pt:larissa",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.045076627284288406,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09552915222676737,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 447.0907910144888,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 417.94004099210724,
      "loadMs": 2979.361500008963,
      "totalMs": 3411.6626249742694,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local, Ol olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local, olá, este é um teste de síntese de voz local, olá. Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local, olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0.011111111111111112,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 3005.8959999587387,
      "inferenceMs": 3103.496666997671,
      "audioSeconds": 31.225,
      "realTimeFactor": 0.09939140646910076,
      "audioSecondsPerSecond": 10.061232007124122,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1962.1965420083143,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07634339481592178,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09610811984009915,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 447.0907910144888,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 3000.976042007096,
      "loadMs": 2979.361500008963,
      "totalMs": 6082.858167006634,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 185.50216598669067,
      "inferenceMs": 201.0707919835113,
      "audioSeconds": 4.6,
      "realTimeFactor": 0.043711041735545936,
      "audioSecondsPerSecond": 22.87751470326541,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 183.7661659810692,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06788214296102524,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03994960130354309,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 183.8497080025263,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 183.7681659962982,
      "loadMs": 2762.7044579712674,
      "totalMs": 2963.7752499547787,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 174.49116701027378,
      "inferenceMs": 188.5746669722721,
      "audioSeconds": 4.375,
      "realTimeFactor": 0.043102781022233624,
      "audioSecondsPerSecond": 23.20035914815269,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 173.69833402335644,
      "voice": "pt:larissa",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04507654905319214,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.039702866745314425,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 183.8497080025263,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 173.70004201075062,
      "loadMs": 2762.7044579712674,
      "totalMs": 2951.2791249435395,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local, Ol olá, este é um teste de síntese de voz local! Olá, este é um teste de síntese de voz local, olá, este é um teste de síntese de voz local, olá. Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local, olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0.011111111111111112,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1142.3353749560192,
      "inferenceMs": 1239.8143339669332,
      "audioSeconds": 31.225,
      "realTimeFactor": 0.039705823345618355,
      "audioSecondsPerSecond": 25.18522261320524,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 711.8428749963641,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07634330540895462,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.036420876701236414,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 183.8497080025263,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1137.241874996107,
      "loadMs": 2762.7044579712674,
      "totalMs": 4002.5187919382006,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3292.175790993497,
      "warmupMs": 146.15350001258776,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 138.16520903492346,
      "inferenceMs": 153.51787500549108,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.033555819673331384,
      "audioSecondsPerSecond": 29.801089937157872,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 135.37512498442084,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06812528520822525,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029590546453748246,
      "synthesisMs": 135.37675002589822,
      "totalMs": 3445.693665998988,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643891200,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3292.175790993497,
      "warmupMs": 146.15350001258776,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 137.49162497697398,
      "inferenceMs": 153.10116700129583,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.0334647359565674,
      "audioSecondsPerSecond": 29.882202008043986,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 136.80804200703278,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06812528520822525,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02990411672120293,
      "synthesisMs": 136.8113339995034,
      "totalMs": 3445.2769579947926,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 648282112,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3292.175790993497,
      "warmupMs": 146.15350001258776,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 139.09412501379848,
      "inferenceMs": 154.76333402330056,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.033828051152634,
      "audioSecondsPerSecond": 29.56126545652736,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 138.36379203712568,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06812528520822525,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.030243834327967442,
      "synthesisMs": 138.36554205045104,
      "totalMs": 3446.9391250167973,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 650641408,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3292.175790993497,
      "warmupMs": 146.15350001258776,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 141.5266660042107,
      "inferenceMs": 157.17724995920435,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.03435568305119221,
      "audioSecondsPerSecond": 29.107265849144518,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 140.76524996198714,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06812528520822525,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0307686428305352,
      "synthesisMs": 140.76654094969854,
      "totalMs": 3449.353040952701,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 651083776,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3178.063333965838,
      "warmupMs": 144.62820801418275,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 139.04724997701123,
      "inferenceMs": 154.9854159820825,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.03387659365728579,
      "audioSecondsPerSecond": 29.518906479103205,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 136.14487496670336,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06812528520822525,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029758806772977927,
      "synthesisMs": 136.14654098637402,
      "totalMs": 3333.0487499479204,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644481024,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3178.063333965838,
      "warmupMs": 144.62820801418275,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 135.88991697179154,
      "inferenceMs": 151.77175001008436,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.03317415300766871,
      "audioSecondsPerSecond": 30.143949711959028,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 135.15262497821823,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06812528520822525,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02954189441691005,
      "synthesisMs": 135.1541669573635,
      "totalMs": 3329.8350839759223,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 649199616,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3178.063333965838,
      "warmupMs": 144.62820801418275,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 134.76341596106067,
      "inferenceMs": 150.91683401260525,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.03298728612297382,
      "audioSecondsPerSecond": 30.314709620915288,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 134.06733301235363,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06812528520822525,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029304644800315464,
      "synthesisMs": 134.06874996144325,
      "totalMs": 3328.980167978443,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 649461760,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3178.063333965838,
      "warmupMs": 144.62820801418275,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 134.7580419969745,
      "inferenceMs": 150.50670900382102,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.032897641312310606,
      "audioSecondsPerSecond": 30.39731604179752,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 134.0390419936739,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06812528520822525,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029298542945294596,
      "synthesisMs": 134.04083397472277,
      "totalMs": 3328.570042969659,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 649494528,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3582.2818329907022,
      "warmupMs": 499.2522079846822,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 104.90962502080947,
      "inferenceMs": 120.18708302639425,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.02627040066150694,
      "audioSecondsPerSecond": 38.06565468433314,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 102.07616700790823,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06808102130889893,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216383488,
      "synthesisRealTimeFactor": 0.02231213115308373,
      "synthesisMs": 102.07800002535805,
      "totalMs": 3702.4689160170965,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 698400768,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3582.2818329907022,
      "warmupMs": 499.2522079846822,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 106.92420898703858,
      "inferenceMs": 122.52491695107892,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.02678140261225769,
      "audioSecondsPerSecond": 37.339343815484334,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 106.1480839853175,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06808102130889893,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216383488,
      "synthesisRealTimeFactor": 0.023202185793856127,
      "synthesisMs": 106.15000000689179,
      "totalMs": 3704.806749941781,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 700170240,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3582.2818329907022,
      "warmupMs": 499.2522079846822,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 103.69858401827514,
      "inferenceMs": 118.79220796981826,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.02596550993875809,
      "audioSecondsPerSecond": 38.512627033267854,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 103.04054198786616,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06808102130889893,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216383488,
      "synthesisRealTimeFactor": 0.022522932684425612,
      "synthesisMs": 103.04241703124717,
      "totalMs": 3701.0740409605205,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 700186624,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3582.2818329907022,
      "warmupMs": 499.2522079846822,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 103.96429098909721,
      "inferenceMs": 119.32945903390646,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.026082941865334747,
      "audioSecondsPerSecond": 38.33923355589882,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 103.282708034385,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06808102130889893,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216367104,
      "synthesisRealTimeFactor": 0.022575765036433285,
      "synthesisMs": 103.28412504168227,
      "totalMs": 3701.6112920246087,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 700203008,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3608.60179096926,
      "warmupMs": 448.0291249928996,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 108.43604104593396,
      "inferenceMs": 123.47287498414516,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.026988606553911512,
      "audioSecondsPerSecond": 37.052672504689504,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 105.63920804997906,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06808102130889893,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216383488,
      "synthesisRealTimeFactor": 0.023090965254486746,
      "synthesisMs": 105.64116603927687,
      "totalMs": 3732.074665953405,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 699416576,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3608.60179096926,
      "warmupMs": 448.0291249928996,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 104.55208300845698,
      "inferenceMs": 119.89358405116946,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.026206247880037038,
      "audioSecondsPerSecond": 38.15883924236874,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 103.7339159520343,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06808102130889893,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216383488,
      "synthesisRealTimeFactor": 0.0226745809825351,
      "synthesisMs": 103.73620799509808,
      "totalMs": 3728.4953750204295,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 699416576,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3608.60179096926,
      "warmupMs": 448.0291249928996,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 102.08683297969401,
      "inferenceMs": 117.71583399968222,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.02573023693982125,
      "audioSecondsPerSecond": 38.86478007718444,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 101.44566599046811,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06808102130889893,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216383488,
      "synthesisRealTimeFactor": 0.022174344255583858,
      "synthesisMs": 101.44762496929616,
      "totalMs": 3726.3176249689423,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 699449344,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3608.60179096926,
      "warmupMs": 448.0291249928996,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 103.7470840383321,
      "inferenceMs": 119.30158297764137,
      "audioSeconds": 4.575,
      "realTimeFactor": 0.02607684873828227,
      "audioSecondsPerSecond": 38.34819191676118,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 103.06554200360551,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06808102130889893,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 216367104,
      "synthesisRealTimeFactor": 0.022528479130215984,
      "synthesisMs": 103.06779202073812,
      "totalMs": 3727.9033739469014,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 699465728,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2785.214208008256,
      "warmupMs": 191.0547079751268,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 188.0163750029169,
      "inferenceMs": 203.0657499562949,
      "audioSeconds": 4.6,
      "realTimeFactor": 0.044144728251368455,
      "audioSecondsPerSecond": 22.65276148730173,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 186.2217080197297,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06788214296102524,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04048341479020841,
      "synthesisMs": 186.2237080349587,
      "totalMs": 2988.2799579645507,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 535314432,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2785.214208008256,
      "warmupMs": 191.0547079751268,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 181.7745840526186,
      "inferenceMs": 196.55583298299462,
      "audioSeconds": 4.6,
      "realTimeFactor": 0.042729528909346656,
      "audioSecondsPerSecond": 23.403019539989828,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 180.9490840532817,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06789079308509827,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03933727370483963,
      "synthesisMs": 180.9514590422623,
      "totalMs": 2981.7700409912504,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 537722880,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2785.214208008256,
      "warmupMs": 191.0547079751268,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 176.14654102362692,
      "inferenceMs": 190.83799998043105,
      "audioSeconds": 4.6,
      "realTimeFactor": 0.04148652173487632,
      "audioSecondsPerSecond": 24.104214047892423,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 175.31466600485146,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0679236650466919,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.038112336959507644,
      "synthesisMs": 175.31675001373515,
      "totalMs": 2976.052207988687,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 538509312,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2785.214208008256,
      "warmupMs": 191.0547079751268,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 175.54033396299928,
      "inferenceMs": 189.9801249965094,
      "audioSeconds": 4.6,
      "realTimeFactor": 0.041300027173154216,
      "audioSecondsPerSecond": 24.213059129656422,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 174.7859999886714,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06786827743053436,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03799745477685381,
      "synthesisMs": 174.78829197352752,
      "totalMs": 2975.194333004765,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 538984448,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2580.646957969293,
      "warmupMs": 182.97629099106416,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 183.80912498105317,
      "inferenceMs": 197.80841702595353,
      "audioSeconds": 4.6,
      "realTimeFactor": 0.04300182978825077,
      "audioSecondsPerSecond": 23.25482438594337,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 182.06233298406005,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06788214296102524,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03957921194935055,
      "synthesisMs": 182.06437496701255,
      "totalMs": 2778.4553749952465,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 534036480,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2580.646957969293,
      "warmupMs": 182.97629099106416,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 177.73791699437425,
      "inferenceMs": 192.43116700090468,
      "audioSeconds": 4.6,
      "realTimeFactor": 0.04183286239150102,
      "audioSecondsPerSecond": 23.904651578495983,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 176.82074999902397,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06789079308509827,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.038439737387386194,
      "synthesisMs": 176.82279198197648,
      "totalMs": 2773.0781249701977,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 536739840,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2580.646957969293,
      "warmupMs": 182.97629099106416,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 177.550541004166,
      "inferenceMs": 192.14537501102313,
      "audioSeconds": 4.6,
      "realTimeFactor": 0.041770733698048505,
      "audioSecondsPerSecond": 23.940206730118298,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 176.75316601525992,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0679236650466919,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03842498173269079,
      "synthesisMs": 176.75491597037762,
      "totalMs": 2772.792332980316,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 539377664,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2580.646957969293,
      "warmupMs": 182.97629099106416,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 177.77008301345631,
      "inferenceMs": 192.44608300505206,
      "audioSeconds": 4.6,
      "realTimeFactor": 0.04183610500109827,
      "audioSecondsPerSecond": 23.902798790034307,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 177.0524160237983,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06786827743053436,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03849010870275695,
      "synthesisMs": 177.05450003268197,
      "totalMs": 2773.093040974345,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 539394048,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Nano · Russian Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64142 ms / 4185 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64143 ms / 3765 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64143 ms / 3845 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64145 ms / 3845 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64147 ms / 3765 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64147 ms / 3845 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64147 ms / 3845 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64159 ms / 3765 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64209 ms / 3765 ms0.042 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch641921 ms / 4185 ms0.054 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64358 ms / 4185 ms0.070 RTF
FP32cpuPassedVery fastmacos · aarch641098 ms / 3357 ms0.031 RTF
FP32cpuPassedVery fastmacos · aarch64162 ms / 3333 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64162 ms / 3333 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64164 ms / 3333 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64170 ms / 3357 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64164 ms / 3333 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64167 ms / 3357 ms0.034 RTF
FP32cpuPassedVery fastmacos · aarch64179 ms / 3242 ms0.036 RTF
FP32cpuPassedVery fastmacos · aarch64184 ms / 3242 ms0.037 RTF
FP32cpuPassedVery fastmacos · aarch64186 ms / 3242 ms0.037 RTF
FP32cpuPassedVery fastmacos · aarch64190 ms / 3242 ms0.038 RTF
Q8cpuPassedVery fastmacos · aarch641439 ms / 2757 ms0.041 RTF
Q8cpuPassedVery fastmacos · aarch64217 ms / 2757 ms0.042 RTF
Q8cpuPassedVery fastmacos · aarch64213 ms / 2813 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64215 ms / 2813 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64215 ms / 2813 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch64220 ms / 2778 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64220 ms / 2778 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64221 ms / 2778 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64222 ms / 2757 ms0.044 RTF
Q8cpuPassedVery fastmacos · aarch64223 ms / 2813 ms0.045 RTF
Q8cpuPassedVery fastmacos · aarch64228 ms / 2778 ms0.046 RTF
FP32cpuPassedVery fastmacos · aarch641925 ms / 3313 ms0.055 RTF
FP32cpuPassedVery fastmacos · aarch64282 ms / 3313 ms0.055 RTF
FP32cpuPassedVery fastmacos · aarch64277 ms / 3313 ms0.056 RTF
Q8cpuPassedFastmacos · aarch64530 ms / 3051 ms0.103 RTF
Q8cpuPassedFastmacos · aarch643761 ms / 3051 ms0.107 RTF
Q8cpuPassedFastmacos · aarch64617 ms / 3051 ms0.123 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:02:56.553265+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
54526b8d8e45333cd720d3a082c106aacf416064
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 260.02849999349564,
      "inferenceMs": 276.8662919988856,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.05565151597967549,
      "audioSecondsPerSecond": 17.96896243339014,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 257.1896250010468,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09775301814079285,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05169670009779646,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 265.52791602443904,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 257.19108298653737,
      "loadMs": 3312.8082909970544,
      "totalMs": 3589.67458299594,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 265.2743330108933,
      "inferenceMs": 282.05812495434657,
      "audioSeconds": 5.15,
      "realTimeFactor": 0.054768567952300305,
      "audioSecondsPerSecond": 18.258647932349298,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 264.48558300035074,
      "voice": "ru:tatiana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.048891473561525345,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05135673126522271,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 265.52791602443904,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 264.48716601589695,
      "loadMs": 3312.8082909970544,
      "totalMs": 3594.866415951401,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте!  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1814.6617500460707,
      "inferenceMs": 1924.7681670240127,
      "audioSeconds": 35.275,
      "realTimeFactor": 0.05456465391988696,
      "audioSecondsPerSecond": 18.326882480886294,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 824.0423750248738,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.10741212218999863,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05130195135447381,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 265.52791602443904,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1809.6763340290636,
      "loadMs": 3312.8082909970544,
      "totalMs": 5237.576458021067,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 150.84133303025737,
      "inferenceMs": 166.90150002250448,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.03354804020552854,
      "audioSecondsPerSecond": 29.808000523237876,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 148.02766602952033,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09775301814079285,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02975466492674758,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 160.11425002943724,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 148.02945801056921,
      "loadMs": 3356.6516670398414,
      "totalMs": 3523.553167062346,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 152.76629099389538,
      "inferenceMs": 169.57974998513237,
      "audioSeconds": 5.15,
      "realTimeFactor": 0.032928106793229585,
      "audioSecondsPerSecond": 30.369192078957056,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 151.98895800858736,
      "voice": "ru:tatiana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.048891473561525345,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02951266990016097,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 160.11425002943724,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 151.990249985829,
      "loadMs": 3356.6516670398414,
      "totalMs": 3526.231417024974,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте!  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 986.457374994643,
      "inferenceMs": 1097.6999580161646,
      "audioSeconds": 35.275,
      "realTimeFactor": 0.03111835458585867,
      "audioSecondsPerSecond": 32.1353751928271,
      "placement": {
        "cpu/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 439.5928329904564,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.10741212218999863,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02782345970222643,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 160.11425002943724,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 981.4725409960374,
      "loadMs": 3356.6516670398414,
      "totalMs": 4454.351625056006,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 124.64058300247416,
      "inferenceMs": 141.77787501830608,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.02849806533031278,
      "audioSecondsPerSecond": 35.090101324749284,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 121.65270803961903,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09770303219556808,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242368512,
      "synthesisRealTimeFactor": 0.024453132265907285,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1104.8628329881467,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 121.65433302288875,
      "loadMs": 4185.445083014201,
      "totalMs": 4327.222958032507,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 341.23166697099805,
      "inferenceMs": 357.9356250120327,
      "audioSeconds": 5.15,
      "realTimeFactor": 0.06950206310913257,
      "audioSecondsPerSecond": 14.388062098671718,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 340.43566696345806,
      "voice": "ru:tatiana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.048878006637096405,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 280821760,
      "synthesisRealTimeFactor": 0.06610439320601233,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1104.8628329881467,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 340.43762501096353,
      "loadMs": 4185.445083014201,
      "totalMs": 4543.380708026234,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте!  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1809.880290995352,
      "inferenceMs": 1920.8542499691248,
      "audioSeconds": 35.275,
      "realTimeFactor": 0.05445369950302267,
      "audioSecondsPerSecond": 18.364225188125022,
      "placement": {
        "mps:0/torch.float32": 568
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 557.1297499700449,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.10741538554430008,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1479688192,
      "synthesisRealTimeFactor": 0.051150538624956686,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1104.8628329881467,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1804.335249995347,
      "loadMs": 4185.445083014201,
      "totalMs": 6106.299332983326,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 599.5426250156015,
      "inferenceMs": 617.2135840170085,
      "audioSeconds": 5,
      "realTimeFactor": 0.12344271680340171,
      "audioSecondsPerSecond": 8.100923455797135,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 597.5708750193007,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09777189791202545,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.11951465000165626,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 511.74362504389137,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 597.5732500082813,
      "loadMs": 3051.20570899453,
      "totalMs": 3668.4192930115387,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 512.9430000088178,
      "inferenceMs": 529.7894579707645,
      "audioSeconds": 5.15,
      "realTimeFactor": 0.10287173941179893,
      "audioSecondsPerSecond": 9.720842728214864,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 511.8491249741055,
      "voice": "ru:tatiana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04896574839949608,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.09938859223677478,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 511.74362504389137,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 511.85125001939014,
      "loadMs": 3051.20570899453,
      "totalMs": 3580.9951669652946,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте!  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 3649.9931670259684,
      "inferenceMs": 3761.3474169629626,
      "audioSeconds": 35.25,
      "realTimeFactor": 0.10670489126136064,
      "audioSecondsPerSecond": 9.371641619976181,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1700.0023750006221,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.10751691460609436,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.10336634870915812,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 511.74362504389137,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 3643.663791997824,
      "loadMs": 3051.20570899453,
      "totalMs": 6812.553125957493,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 205.30450000660494,
      "inferenceMs": 221.78700001677498,
      "audioSeconds": 5,
      "realTimeFactor": 0.044357400003354995,
      "audioSecondsPerSecond": 22.54415272140307,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 203.53166601853445,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09777369350194931,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04070678320713341,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 205.24420798756182,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 203.53391603566706,
      "loadMs": 2757.2384999948554,
      "totalMs": 2979.0255000116304,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 201.2741249636747,
      "inferenceMs": 217.20316703431308,
      "audioSeconds": 5.15,
      "realTimeFactor": 0.042175372239672444,
      "audioSecondsPerSecond": 23.710519834116504,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 200.34683297853917,
      "voice": "ru:tatiana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04896537959575653,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.038902758828644614,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 205.24420798756182,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 200.34920796751976,
      "loadMs": 2757.2384999948554,
      "totalMs": 2974.4416670291685,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте!  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1329.0607500239275,
      "inferenceMs": 1438.9718750026077,
      "audioSeconds": 35.25,
      "realTimeFactor": 0.040821897163194544,
      "audioSecondsPerSecond": 24.496656684089896,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 601.0623750044033,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.10751691460609436,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03750491960317636,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 205.24420798756182,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1322.0484160119668,
      "loadMs": 2757.2384999948554,
      "totalMs": 4196.210374997463,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3333.10245804023,
      "warmupMs": 153.9695409592241,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 147.78133295476437,
      "inferenceMs": 164.12533301627263,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.03299001668668797,
      "audioSecondsPerSecond": 30.312200490747767,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 144.92545800749213,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09775301814079285,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029131046826272603,
      "synthesisMs": 144.9269579607062,
      "totalMs": 3497.2277910565026,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 677560320,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3333.10245804023,
      "warmupMs": 153.9695409592241,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 145.28641704237089,
      "inferenceMs": 161.69770900160074,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.03250205206062327,
      "audioSecondsPerSecond": 30.767288112602447,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 144.20862501719967,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09775301814079285,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02898711055966478,
      "synthesisMs": 144.21087503433228,
      "totalMs": 3494.8001670418307,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 677855232,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3333.10245804023,
      "warmupMs": 153.9695409592241,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 146.085333020892,
      "inferenceMs": 162.39050001604483,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.032641306535888405,
      "audioSecondsPerSecond": 30.63602858238906,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 145.32679104013368,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09775301814079285,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029211608042481855,
      "synthesisMs": 145.32775001134723,
      "totalMs": 3495.492958056275,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 677937152,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3333.10245804023,
      "warmupMs": 153.9695409592241,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 147.40645798156038,
      "inferenceMs": 163.7838330352679,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.032921373474425704,
      "audioSecondsPerSecond": 30.37540340705498,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 146.64945797994733,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09775301814079285,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.029477554375783133,
      "synthesisMs": 146.6508330195211,
      "totalMs": 3496.886291075498,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 678821888,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3242.3962500179186,
      "warmupMs": 187.83183302730322,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 169.84487499576062,
      "inferenceMs": 186.22274999506772,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.037431708541722154,
      "audioSecondsPerSecond": 26.715318080802522,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 167.10441699251533,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09775301814079285,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.033589212666148664,
      "synthesisMs": 167.1063330140896,
      "totalMs": 3428.6190000129864,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 678985728,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3242.3962500179186,
      "warmupMs": 187.83183302730322,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 161.52529197279364,
      "inferenceMs": 178.7223750143312,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.03592409548026758,
      "audioSecondsPerSecond": 27.836469829818846,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 160.69195896852762,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09775301814079285,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03230023456400438,
      "synthesisMs": 160.6936669559218,
      "totalMs": 3421.11862503225,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 681148416,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3242.3962500179186,
      "warmupMs": 187.83183302730322,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 165.98558297846466,
      "inferenceMs": 184.0571669745259,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.036996415472266515,
      "audioSecondsPerSecond": 27.02964563552451,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 165.17662501428276,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09775301814079285,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03320170010688092,
      "synthesisMs": 165.1784580317326,
      "totalMs": 3426.4534169924445,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 694272000,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 568
      },
      "loadMs": 3242.3962500179186,
      "warmupMs": 187.83183302730322,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 172.25766699993983,
      "inferenceMs": 190.4137089732103,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.03827411235642418,
      "audioSecondsPerSecond": 26.12732048982851,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 171.27616703510284,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09775301814079285,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03442757970059113,
      "synthesisMs": 171.27720901044086,
      "totalMs": 3432.809958991129,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 695353344,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3765.236917010043,
      "warmupMs": 519.1607080050744,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 127.10245902417228,
      "inferenceMs": 142.84337498247623,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.028712236177382156,
      "audioSecondsPerSecond": 34.82835658713835,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 124.15516702458262,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09770303219556808,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242384896,
      "synthesisRealTimeFactor": 0.024956273167726382,
      "synthesisMs": 124.15745900943875,
      "totalMs": 3908.080291992519,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 705134592,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3765.236917010043,
      "warmupMs": 519.1607080050744,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 136.459666013252,
      "inferenceMs": 158.7278749793768,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.031905100498367193,
      "audioSecondsPerSecond": 31.342950950779073,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 135.57533296989277,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09770303219556808,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242384896,
      "synthesisRealTimeFactor": 0.027251741907386176,
      "synthesisMs": 135.57741598924622,
      "totalMs": 3923.9647919894196,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 705150976,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3765.236917010043,
      "warmupMs": 519.1607080050744,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 129.4232080108486,
      "inferenceMs": 146.91274997312576,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.02953020099962327,
      "audioSecondsPerSecond": 33.863636756578714,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 128.65308299660683,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09770303219556808,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242368512,
      "synthesisRealTimeFactor": 0.025860267941694912,
      "synthesisMs": 128.6548330099322,
      "totalMs": 3912.1496669831686,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 705232896,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3765.236917010043,
      "warmupMs": 519.1607080050744,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 187.4219580204226,
      "inferenceMs": 209.37250001588836,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.042084924626309216,
      "audioSecondsPerSecond": 23.761477747184895,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 186.60358298802748,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09770303219556808,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242368512,
      "synthesisRealTimeFactor": 0.03750869346754485,
      "synthesisMs": 186.60575000103563,
      "totalMs": 3974.609417025931,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 705314816,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3844.9094999814406,
      "warmupMs": 531.121083011385,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 131.883249967359,
      "inferenceMs": 147.36516698030755,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.029621139091519106,
      "audioSecondsPerSecond": 33.75967402571335,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 128.89483297476545,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09770303219556808,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242384896,
      "synthesisRealTimeFactor": 0.025909036777015008,
      "synthesisMs": 128.89745796564966,
      "totalMs": 3992.274666961748,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 706428928,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3844.9094999814406,
      "warmupMs": 531.121083011385,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 127.84075003582984,
      "inferenceMs": 145.05487499991432,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.029156758793952627,
      "audioSecondsPerSecond": 34.29736504893709,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 127.02362501295283,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09770303219556808,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242384896,
      "synthesisRealTimeFactor": 0.025532722014503264,
      "synthesisMs": 127.02529202215374,
      "totalMs": 3989.964374981355,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 706510848,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3844.9094999814406,
      "warmupMs": 531.121083011385,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 126.40599999576807,
      "inferenceMs": 143.19974998943508,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.028783869344610066,
      "audioSecondsPerSecond": 34.74168076667063,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 125.40854199323803,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09770303219556808,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242368512,
      "synthesisRealTimeFactor": 0.025208090448368286,
      "synthesisMs": 125.41024998063222,
      "totalMs": 3988.1092499708757,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 706543616,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 568
      },
      "loadMs": 3844.9094999814406,
      "warmupMs": 531.121083011385,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 129.61149995680898,
      "inferenceMs": 147.27841597050428,
      "audioSeconds": 4.975,
      "realTimeFactor": 0.029603701702613925,
      "audioSecondsPerSecond": 33.77955939583403,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 128.81062499945983,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09770303219556808,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 242368512,
      "synthesisRealTimeFactor": 0.025891859292628328,
      "synthesisMs": 128.81199998082593,
      "totalMs": 3992.187915951945,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 706592768,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2813.304499955848,
      "warmupMs": 206.61766699049622,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 206.49641595082358,
      "inferenceMs": 222.7069159853272,
      "audioSeconds": 5,
      "realTimeFactor": 0.044541383197065444,
      "audioSecondsPerSecond": 22.451031562618464,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 204.50387499295175,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09777369350194931,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04090128319803625,
      "synthesisMs": 204.50641599018127,
      "totalMs": 3036.011415941175,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 543752192,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2813.304499955848,
      "warmupMs": 206.61766699049622,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 198.35504196817055,
      "inferenceMs": 214.79529200587422,
      "audioSeconds": 5,
      "realTimeFactor": 0.042959058401174845,
      "audioSecondsPerSecond": 23.277977619096326,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 197.4291669903323,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09780501574277878,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03948625839548185,
      "synthesisMs": 197.43129197740927,
      "totalMs": 3028.099791961722,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 546455552,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2813.304499955848,
      "warmupMs": 206.61766699049622,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 198.55108298361301,
      "inferenceMs": 215.17841599415988,
      "audioSeconds": 5,
      "realTimeFactor": 0.04303568319883198,
      "audioSecondsPerSecond": 23.236531307748375,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 197.64970801770687,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09781086444854736,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03953036660095677,
      "synthesisMs": 197.65183300478384,
      "totalMs": 3028.4829159500077,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 549011456,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2813.304499955848,
      "warmupMs": 206.61766699049622,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 196.28350000130013,
      "inferenceMs": 212.66554202884436,
      "audioSeconds": 5,
      "realTimeFactor": 0.04253310840576887,
      "audioSecondsPerSecond": 23.511096119755205,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 195.4523750464432,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09778649359941483,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.039090891601517797,
      "synthesisMs": 195.45445800758898,
      "totalMs": 3025.970041984692,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 551698432,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2777.6523749926127,
      "warmupMs": 208.40858400333673,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 212.05962501699105,
      "inferenceMs": 228.429791983217,
      "audioSeconds": 5,
      "realTimeFactor": 0.0456859583966434,
      "audioSecondsPerSecond": 21.888563468846286,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 210.2078750031069,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09777369350194931,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04204207500442862,
      "synthesisMs": 210.21037502214313,
      "totalMs": 3006.0821669758298,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 543817728,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2777.6523749926127,
      "warmupMs": 208.40858400333673,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 203.04154203040525,
      "inferenceMs": 220.25291697354987,
      "audioSeconds": 5,
      "realTimeFactor": 0.04405058339470998,
      "audioSecondsPerSecond": 22.70117494335137,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 202.07558298716322,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09780501574277878,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04041570840636268,
      "synthesisMs": 202.0785420318134,
      "totalMs": 2997.9052919661626,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 549126144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2777.6523749926127,
      "warmupMs": 208.40858400333673,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 202.09545805118978,
      "inferenceMs": 219.68191699124873,
      "audioSeconds": 5,
      "realTimeFactor": 0.04393638339824975,
      "audioSecondsPerSecond": 22.760180120784273,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 201.29908301169053,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09781086444854736,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04026030000532046,
      "synthesisMs": 201.3015000266023,
      "totalMs": 2997.3342919838615,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 549126144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2777.6523749926127,
      "warmupMs": 208.40858400333673,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 204.01466701878235,
      "inferenceMs": 221.4880830142647,
      "audioSeconds": 5,
      "realTimeFactor": 0.04429761660285294,
      "audioSecondsPerSecond": 22.5745779725674,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 203.22066702647135,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09778649359941483,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.040644591604359445,
      "synthesisMs": 203.22295802179724,
      "totalMs": 2999.1404580068775,
      "grade": "fast",
      "artifactRevision": "54526b8d8e45333cd720d3a082c106aacf416064",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 549126144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Pico · German Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch6495 ms / 3761 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6496 ms / 3711 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6496 ms / 3761 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6497 ms / 3711 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6497 ms / 3761 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 3711 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64100 ms / 3711 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64103 ms / 4456 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64123 ms / 3761 ms0.035 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642270 ms / 4456 ms0.087 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64865 ms / 4456 ms0.246 RTF
FP32cpuPassedVery fastmacos · aarch64635 ms / 3137 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64644 ms / 2628 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch6490 ms / 3068 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch6490 ms / 3068 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch6490 ms / 2670 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch6490 ms / 2650 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch6491 ms / 2628 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch6491 ms / 3068 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch6492 ms / 2670 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch6492 ms / 2670 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch6492 ms / 2650 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch6492 ms / 3017 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch6493 ms / 2628 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch6493 ms / 3017 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch6493 ms / 3017 ms0.027 RTF
Q8cpuPassedVery fastmacos · aarch6494 ms / 2650 ms0.027 RTF
Q8cpuPassedVery fastmacos · aarch6494 ms / 2670 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch6495 ms / 3068 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch6495 ms / 3017 ms0.027 RTF
Q8cpuPassedVery fastmacos · aarch6496 ms / 2650 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch6496 ms / 3137 ms0.028 RTF
FP32cpuPassedVery fastmacos · aarch6499 ms / 3137 ms0.028 RTF
FP32cpuPassedVery fastmacos · aarch641001 ms / 3123 ms0.038 RTF
FP32cpuPassedVery fastmacos · aarch64138 ms / 3123 ms0.039 RTF
FP32cpuPassedVery fastmacos · aarch64140 ms / 3123 ms0.040 RTF
Q8cpuPassedVery fastmacos · aarch641361 ms / 2674 ms0.052 RTF
Q8cpuPassedVery fastmacos · aarch64183 ms / 2674 ms0.052 RTF
Q8cpuPassedVery fastmacos · aarch64185 ms / 2674 ms0.053 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:04:29.220392+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
da8c9d029e16d9302f1827adb303d92316e25f73
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 124.5128340087831,
      "inferenceMs": 137.59408303303644,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.039312595152296126,
      "audioSecondsPerSecond": 25.437140339527883,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 121.78308400325477,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14388658106327057,
      "clippedFraction": 0.00036904761904761905,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03479560714081994,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 128.35229199845344,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 121.7846249928698,
      "loadMs": 3122.7724170312285,
      "totalMs": 3260.366500064265,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 127.4221669882536,
      "inferenceMs": 139.57725005457178,
      "audioSeconds": 3.525,
      "realTimeFactor": 0.03959638299420476,
      "audioSecondsPerSecond": 25.254831991759396,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 126.70195900136605,
      "voice": "de:mia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07092894613742828,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03594426723884035,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 128.35229199845344,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 126.70354201691225,
      "loadMs": 3122.7724170312285,
      "totalMs": 3262.3496670858003,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen. Dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest. Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 918.2623750530183,
      "inferenceMs": 1000.9420830174349,
      "audioSeconds": 26.2,
      "realTimeFactor": 0.03820389629837538,
      "audioSecondsPerSecond": 26.17534065609232,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 600.0368749955669,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1648687869310379,
      "clippedFraction": 0.000031806615776081425,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.034867523856574795,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 128.35229199845344,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 913.5291250422597,
      "loadMs": 3122.7724170312285,
      "totalMs": 4123.7145000486635,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 83.94554199185222,
      "inferenceMs": 96.27945895772427,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.02750841684506408,
      "audioSecondsPerSecond": 36.35251005655141,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 81.31633303128183,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14388658106327057,
      "clippedFraction": 0.00036904761904761905,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02323376201093197,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 85.08929202798754,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 81.31816703826189,
      "loadMs": 3136.711583007127,
      "totalMs": 3232.9910419648513,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 86.21145796496421,
      "inferenceMs": 98.64691703114659,
      "audioSeconds": 3.525,
      "realTimeFactor": 0.027984941001743712,
      "audioSecondsPerSecond": 35.733503956205986,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.59483295539394,
      "voice": "de:mia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07092894613742828,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024282600555621776,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 85.08929202798754,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 85.59616695856676,
      "loadMs": 3136.711583007127,
      "totalMs": 3235.3585000382736,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen. Dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest. Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 550.9330409695394,
      "inferenceMs": 635.0552500225604,
      "audioSeconds": 26.2,
      "realTimeFactor": 0.024238750000861082,
      "audioSecondsPerSecond": 41.256252899364654,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 342.24512497894466,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1648687869310379,
      "clippedFraction": 0.000031806615776081425,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02087807408329597,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 85.08929202798754,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 547.0055409823544,
      "loadMs": 3136.711583007127,
      "totalMs": 3771.7668330296874,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 89.75645794998854,
      "inferenceMs": 102.6960420422256,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.02934172629777874,
      "audioSecondsPerSecond": 34.08115766098272,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 87.07433298695832,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14398956298828125,
      "clippedFraction": 0.00035714285714285714,
      "driverAllocatedMemoryBytes": 134627328,
      "synthesisRealTimeFactor": 0.024878821418886737,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1151.9103329628706,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 87.07587496610358,
      "loadMs": 4455.5978330317885,
      "totalMs": 4558.293875074014,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 853.3096660394222,
      "inferenceMs": 865.4794589965604,
      "audioSeconds": 3.525,
      "realTimeFactor": 0.2455260876585987,
      "audioSecondsPerSecond": 4.072886956886182,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 852.7143330429681,
      "voice": "de:mia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07094516605138779,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 171261952,
      "synthesisRealTimeFactor": 0.2419052363195001,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1151.9103329628706,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 852.7159580262378,
      "loadMs": 4455.5978330317885,
      "totalMs": 5321.077292028349,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen. Dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest. Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2185.764125024434,
      "inferenceMs": 2269.6444589528255,
      "audioSeconds": 26.2,
      "realTimeFactor": 0.08662765110506968,
      "audioSecondsPerSecond": 11.543658257420734,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1132.5944589916617,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16488078236579895,
      "clippedFraction": 0.000031806615776081425,
      "driverAllocatedMemoryBytes": 1333477376,
      "synthesisRealTimeFactor": 0.08326435591607845,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1151.9103329628706,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2181.5261250012554,
      "loadMs": 4455.5978330317885,
      "totalMs": 6725.242291984614,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 174.10325002856553,
      "inferenceMs": 185.38991699460894,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.05296854771274541,
      "audioSecondsPerSecond": 18.87912814644487,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 172.58779203984886,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1439656913280487,
      "clippedFraction": 0.00038095238095238096,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04931140486483595,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 172.2574169980362,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 172.58991702692583,
      "loadMs": 2674.2544579901733,
      "totalMs": 2859.644374984782,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 171.13400000380352,
      "inferenceMs": 182.85008298698813,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.052242880853425176,
      "audioSecondsPerSecond": 19.14136402250944,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 170.50333303632215,
      "voice": "de:mia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0711321011185646,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.048715916578657925,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 172.2574169980362,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 170.50570802530274,
      "loadMs": 2674.2544579901733,
      "totalMs": 2857.1045409771614,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen. Dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest. Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1278.4554589889012,
      "inferenceMs": 1361.4910000469536,
      "audioSeconds": 26.2,
      "realTimeFactor": 0.051965305345303574,
      "audioSecondsPerSecond": 19.24360866072302,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 859.5188339822926,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1648429036140442,
      "clippedFraction": 0.00002703562340966921,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04862938614567486,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 172.2574169980362,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1274.0899170166813,
      "loadMs": 2674.2544579901733,
      "totalMs": 4035.745458037127,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 80.31849999679253,
      "inferenceMs": 92.85858296789229,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.026531023705112084,
      "audioSecondsPerSecond": 37.69172313570836,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 78.72704195324332,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14396387338638306,
      "clippedFraction": 0.00038095238095238096,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022494000000213937,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 80.96162503352389,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 78.72900000074878,
      "loadMs": 2627.6934580528177,
      "totalMs": 2720.55204102071,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 79.45662504062057,
      "inferenceMs": 90.67762497579679,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.025907892850227654,
      "audioSecondsPerSecond": 38.598276045873526,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 78.715292038396,
      "voice": "de:mia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07113310694694519,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02249135713957782,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 80.96162503352389,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 78.71974998852238,
      "loadMs": 2627.6934580528177,
      "totalMs": 2718.3710830286145,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, Guten Morgen, dies ist ein lokaler Sprachtest, guten Morgen. Dies ist ein lokaler Sprachtest, guten Morgen, dies ist ein lokaler Sprachtest. Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 561.5126250195317,
      "inferenceMs": 643.8628750038333,
      "audioSeconds": 26.2,
      "realTimeFactor": 0.024574918893276083,
      "audioSecondsPerSecond": 40.69189421714059,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 355.246499995701,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.16484670341014862,
      "clippedFraction": 0.00002703562340966921,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021276003512276608,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 80.96162503352389,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 557.4312920216471,
      "loadMs": 2627.6934580528177,
      "totalMs": 3271.556333056651,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3068.4853330021724,
      "warmupMs": 81.40645804814994,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 82.2017919854261,
      "inferenceMs": 94.76033400278538,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.027074381143652966,
      "audioSecondsPerSecond": 36.935285600588124,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 79.58195899846032,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14388658106327057,
      "clippedFraction": 0.00036904761904761905,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0227380834320294,
      "synthesisMs": 79.5832920121029,
      "totalMs": 3163.245667004958,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 512655360,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3068.4853330021724,
      "warmupMs": 81.40645804814994,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 77.70912494743243,
      "inferenceMs": 90.02933301962912,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.025722666577036892,
      "audioSecondsPerSecond": 38.87621825696403,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 77.12674996582791,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14388658106327057,
      "clippedFraction": 0.00036904761904761905,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022036654858051667,
      "synthesisMs": 77.12829200318083,
      "totalMs": 3158.5146660218015,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 515686400,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3068.4853330021724,
      "warmupMs": 81.40645804814994,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 78.36395798949525,
      "inferenceMs": 90.68879199912772,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.025911083428322206,
      "audioSecondsPerSecond": 38.59352322207208,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 77.77825003722683,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14388658106327057,
      "clippedFraction": 0.00036904761904761905,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022222821434427584,
      "synthesisMs": 77.77987502049655,
      "totalMs": 3159.1741250013,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 515817472,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3068.4853330021724,
      "warmupMs": 81.40645804814994,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 77.07237498834729,
      "inferenceMs": 89.94025003630668,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.025697214296087623,
      "audioSecondsPerSecond": 38.91472392601906,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 76.49454200873151,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14388658106327057,
      "clippedFraction": 0.00036904761904761905,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0218559525674209,
      "synthesisMs": 76.49583398597315,
      "totalMs": 3158.425583038479,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 516587520,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3016.584375000093,
      "warmupMs": 83.30500003648922,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 81.36650000233203,
      "inferenceMs": 93.28220796305686,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.026652059418016245,
      "audioSecondsPerSecond": 37.52055270160551,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 78.66462500533089,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14388658106327057,
      "clippedFraction": 0.00036904761904761905,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022476083133369684,
      "synthesisMs": 78.6662909667939,
      "totalMs": 3109.86658296315,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 515489792,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3016.584375000093,
      "warmupMs": 83.30500003648922,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 83.05920899147168,
      "inferenceMs": 95.48437502235174,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.02728125000638621,
      "audioSecondsPerSecond": 36.65521190436333,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 82.45045895455405,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14388658106327057,
      "clippedFraction": 0.00036904761904761905,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023557702561707368,
      "synthesisMs": 82.45195896597579,
      "totalMs": 3112.068750022445,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 517505024,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3016.584375000093,
      "warmupMs": 83.30500003648922,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 80.66550001967698,
      "inferenceMs": 92.90229203179479,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.026543512009084225,
      "audioSecondsPerSecond": 37.67398977413995,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 80.07808297406882,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14388658106327057,
      "clippedFraction": 0.00036904761904761905,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022879845146755024,
      "synthesisMs": 80.07945801364258,
      "totalMs": 3109.486667031888,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 520208384,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3016.584375000093,
      "warmupMs": 83.30500003648922,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 79.87658301135525,
      "inferenceMs": 92.14037499623373,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.02632582142749535,
      "audioSecondsPerSecond": 37.98551937891574,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 79.30458296323195,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14388658106327057,
      "clippedFraction": 0.00036904761904761905,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022658809420785735,
      "synthesisMs": 79.30583297275007,
      "totalMs": 3108.724749996327,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 520257536,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3760.9289579559118,
      "warmupMs": 506.72837503952906,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 85.8840830042027,
      "inferenceMs": 97.45995799312368,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.027845702283749624,
      "audioSecondsPerSecond": 35.91218457376047,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 83.11195799615234,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14398956298828125,
      "clippedFraction": 0.00035714285714285714,
      "driverAllocatedMemoryBytes": 134643712,
      "synthesisRealTimeFactor": 0.02374677371699363,
      "synthesisMs": 83.1137080094777,
      "totalMs": 3858.3889159490354,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644612096,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3760.9289579559118,
      "warmupMs": 506.72837503952906,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 82.45599997462705,
      "inferenceMs": 95.15666600782424,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.027187618859378353,
      "audioSecondsPerSecond": 36.78144839282424,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 81.88541699200869,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14398956298828125,
      "clippedFraction": 0.00035714285714285714,
      "driverAllocatedMemoryBytes": 134643712,
      "synthesisRealTimeFactor": 0.023396404859210764,
      "synthesisMs": 81.88741700723767,
      "totalMs": 3856.085623963736,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644710400,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3760.9289579559118,
      "warmupMs": 506.72837503952906,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 82.9157080152072,
      "inferenceMs": 95.87566595291719,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.027393047415119196,
      "audioSecondsPerSecond": 36.50561344438314,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 82.33150001615286,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14398956298828125,
      "clippedFraction": 0.00035714285714285714,
      "driverAllocatedMemoryBytes": 134643712,
      "synthesisRealTimeFactor": 0.023523833153636327,
      "synthesisMs": 82.33341603772715,
      "totalMs": 3856.804623908829,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644726784,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3760.9289579559118,
      "warmupMs": 506.72837503952906,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 110.75716698542237,
      "inferenceMs": 123.42604203149676,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.0352645834375705,
      "audioSecondsPerSecond": 28.357062597104463,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 110.04933400545269,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14398956298828125,
      "clippedFraction": 0.00035714285714285714,
      "driverAllocatedMemoryBytes": 134627328,
      "synthesisRealTimeFactor": 0.031443095416761935,
      "synthesisMs": 110.05083395866677,
      "totalMs": 3884.3549999874085,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644726784,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3710.996832989622,
      "warmupMs": 502.1942919702269,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 88.09962501982227,
      "inferenceMs": 100.04595899954438,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.02858455971415554,
      "audioSecondsPerSecond": 34.98392173956711,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.23629198316485,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14398956298828125,
      "clippedFraction": 0.00035714285714285714,
      "driverAllocatedMemoryBytes": 134643712,
      "synthesisRealTimeFactor": 0.02435372628471149,
      "synthesisMs": 85.23804199649021,
      "totalMs": 3811.0427919891663,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642007040,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3710.996832989622,
      "warmupMs": 502.1942919702269,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 84.66891699936241,
      "inferenceMs": 97.04229200724512,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.02772636914492718,
      "audioSecondsPerSecond": 36.06674912149325,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 84.10733297932893,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14398956298828125,
      "clippedFraction": 0.00035714285714285714,
      "driverAllocatedMemoryBytes": 134643712,
      "synthesisRealTimeFactor": 0.02403113086308752,
      "synthesisMs": 84.10895802080631,
      "totalMs": 3808.039124996867,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642056192,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3710.996832989622,
      "warmupMs": 502.1942919702269,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 83.01220799330622,
      "inferenceMs": 95.57379101170227,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.027306797431914935,
      "audioSecondsPerSecond": 36.62091838097593,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 82.44729199213907,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14398956298828125,
      "clippedFraction": 0.00035714285714285714,
      "driverAllocatedMemoryBytes": 134643712,
      "synthesisRealTimeFactor": 0.02355676200906081,
      "synthesisMs": 82.44866703171283,
      "totalMs": 3806.570624001324,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642072576,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3710.996832989622,
      "warmupMs": 502.1942919702269,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 86.02387498831376,
      "inferenceMs": 98.73629198409617,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.02821036913831319,
      "audioSecondsPerSecond": 35.447958695509435,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.46829101396725,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14398956298828125,
      "clippedFraction": 0.00035714285714285714,
      "driverAllocatedMemoryBytes": 134627328,
      "synthesisRealTimeFactor": 0.02441991657750415,
      "synthesisMs": 85.46970802126452,
      "totalMs": 3809.733124973718,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642187264,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2649.8129999963567,
      "warmupMs": 83.88883399311453,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 82.74016599170864,
      "inferenceMs": 93.92370900604874,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.02683534543029964,
      "audioSecondsPerSecond": 37.2642864835608,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 81.14054100587964,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14396387338638306,
      "clippedFraction": 0.00038095238095238096,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023183535717959915,
      "synthesisMs": 81.1423750128597,
      "totalMs": 2743.7367090024054,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 349110272,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2649.8129999963567,
      "warmupMs": 83.88883399311453,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 80.4068329744041,
      "inferenceMs": 91.97870799107477,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.02627963085459279,
      "audioSecondsPerSecond": 38.0522848868417,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 79.73229099297896,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14400041103363037,
      "clippedFraction": 0.00034523809523809523,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022781309422238598,
      "synthesisMs": 79.73458297783509,
      "totalMs": 2741.7917079874314,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 351436800,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2649.8129999963567,
      "warmupMs": 83.88883399311453,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 78.68800003780052,
      "inferenceMs": 90.48545797122642,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.02585298799177898,
      "audioSecondsPerSecond": 38.680248500405106,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 78.03062500897795,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14405831694602966,
      "clippedFraction": 0.00038095238095238096,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022295000014959702,
      "synthesisMs": 78.03250005235896,
      "totalMs": 2740.298457967583,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 353468416,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2649.8129999963567,
      "warmupMs": 83.88883399311453,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 80.96349996048957,
      "inferenceMs": 95.84537503542379,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.027384392867263938,
      "audioSecondsPerSecond": 36.51715065757137,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 80.38229198427871,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1439506560564041,
      "clippedFraction": 0.0003095238095238095,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022966809703835418,
      "synthesisMs": 80.38383396342397,
      "totalMs": 2745.6583750317805,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 354287616,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2669.5299170096405,
      "warmupMs": 82.87933300016448,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 82.38470798823982,
      "inferenceMs": 94.1646249848418,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.026904178567097654,
      "audioSecondsPerSecond": 37.168947474313356,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 80.85112500702962,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14396387338638306,
      "clippedFraction": 0.00038095238095238096,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02310092856974474,
      "synthesisMs": 80.85324999410659,
      "totalMs": 2763.6945419944823,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 349192192,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2669.5299170096405,
      "warmupMs": 82.87933300016448,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 79.77379200747237,
      "inferenceMs": 91.64733299985528,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.026184952285672938,
      "audioSecondsPerSecond": 38.18987291212857,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 79.11341701401398,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14400041103363037,
      "clippedFraction": 0.00034523809523809523,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02260442858096212,
      "synthesisMs": 79.11550003336743,
      "totalMs": 2761.1772500094958,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 350633984,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2669.5299170096405,
      "warmupMs": 82.87933300016448,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 78.39891704497859,
      "inferenceMs": 90.34470800543204,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.025812773715837727,
      "audioSecondsPerSecond": 38.7405092923601,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 77.78562500607222,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.14405831694602966,
      "clippedFraction": 0.00038095238095238096,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022225154863138284,
      "synthesisMs": 77.788042020984,
      "totalMs": 2759.8746250150725,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 350994432,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Guten Morgen, dies ist ein lokaler Sprachtest.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2669.5299170096405,
      "warmupMs": 82.87933300016448,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "de"
      ],
      "sampleRate": 24000,
      "durationMs": 79.81833402300254,
      "inferenceMs": 91.52083401568234,
      "audioSeconds": 3.5,
      "realTimeFactor": 0.026148809718766382,
      "audioSecondsPerSecond": 38.24265849019979,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 79.27204202860594,
      "voice": "de:anna",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1439506560564041,
      "clippedFraction": 0.0003095238095238095,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02264961914625019,
      "synthesisMs": 79.27366701187566,
      "totalMs": 2761.050751025323,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 350994432,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Pico · English Desktop

44 configurations tested
Passed44
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch6496 ms / 3538 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6497 ms / 3538 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6497 ms / 3637 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 3461 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 3637 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 3637 ms0.032 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 3637 ms0.033 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64103 ms / 3538 ms0.034 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64103 ms / 3538 ms0.034 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642217 ms / 3461 ms0.097 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64929 ms / 3461 ms0.269 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64904 ms / 3461 ms0.294 RTF
FP32cpuPassedVery fastmacos · aarch64576 ms / 2859 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64604 ms / 3326 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch6495 ms / 3326 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch6486 ms / 2859 ms0.028 RTF
FP32cpuPassedVery fastmacos · aarch6497 ms / 2859 ms0.028 RTF
Q8cpuPassedVery fastmacos · aarch6485 ms / 3501 ms0.028 RTF
Q8cpuPassedVery fastmacos · aarch6487 ms / 3326 ms0.028 RTF
Q8cpuPassedVery fastmacos · aarch6486 ms / 3479 ms0.028 RTF
Q8cpuPassedVery fastmacos · aarch6486 ms / 3479 ms0.029 RTF
Q8cpuPassedVery fastmacos · aarch6486 ms / 3479 ms0.029 RTF
Q8cpuPassedVery fastmacos · aarch6487 ms / 3501 ms0.029 RTF
Q8cpuPassedVery fastmacos · aarch6488 ms / 3501 ms0.029 RTF
FP32cpuPassedVery fastmacos · aarch6489 ms / 2961 ms0.029 RTF
FP32cpuPassedVery fastmacos · aarch6489 ms / 2961 ms0.029 RTF
FP32cpuPassedVery fastmacos · aarch6489 ms / 2859 ms0.030 RTF
Q8cpuPassedVery fastmacos · aarch6490 ms / 3326 ms0.030 RTF
Q8cpuPassedVery fastmacos · aarch6490 ms / 3479 ms0.030 RTF
FP32cpuPassedVery fastmacos · aarch6491 ms / 2961 ms0.030 RTF
Q8cpuPassedVery fastmacos · aarch6491 ms / 3501 ms0.030 RTF
FP32cpuPassedVery fastmacos · aarch6492 ms / 2969 ms0.030 RTF
FP32cpuPassedVery fastmacos · aarch6492 ms / 2961 ms0.030 RTF
FP32cpuPassedVery fastmacos · aarch6493 ms / 2969 ms0.031 RTF
FP32cpuPassedVery fastmacos · aarch6494 ms / 2969 ms0.031 RTF
FP32cpuPassedVery fastmacos · aarch6498 ms / 2969 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64917 ms / 2898 ms0.040 RTF
FP32cpuPassedVery fastmacos · aarch64143 ms / 2898 ms0.041 RTF
FP32cpuPassedVery fastmacos · aarch64128 ms / 2898 ms0.042 RTF
FP32cpuPassedVery fastmacos · aarch64128 ms / 2898 ms0.042 RTF
Q8cpuPassedVery fastmacos · aarch641228 ms / 3405 ms0.054 RTF
Q8cpuPassedVery fastmacos · aarch64169 ms / 3405 ms0.055 RTF
Q8cpuPassedVery fastmacos · aarch64190 ms / 3405 ms0.055 RTF
Q8cpuPassedVery fastmacos · aarch64168 ms / 3405 ms0.056 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:04:07.839289+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
da8c9d029e16d9302f1827adb303d92316e25f73
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 115.24062498938292,
      "inferenceMs": 127.7132909744978,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.042219269743635636,
      "audioSecondsPerSecond": 23.68586681087125,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 111.76525003975257,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838505744934082,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03694765850870816,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 122.62124998960644,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 111.76666698884219,
      "loadMs": 2898.0010410305113,
      "totalMs": 3025.714332005009,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 115.893958020024,
      "inferenceMs": 128.30695800948888,
      "audioSeconds": 3.075,
      "realTimeFactor": 0.041725840003085816,
      "audioSecondsPerSecond": 23.965964494089167,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 115.36312499083579,
      "voice": "en:edith",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.10056133568286896,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03751692390523669,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 122.62124998960644,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 115.36454100860283,
      "loadMs": 2898.0010410305113,
      "totalMs": 3026.30799904,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello. This is a local speech synthesis test. hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 826.1584580177441,
      "inferenceMs": 916.720749984961,
      "audioSeconds": 22.925,
      "realTimeFactor": 0.0399878189742622,
      "audioSecondsPerSecond": 25.007615460189037,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 607.5545829953626,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.18483532965183258,
      "clippedFraction": 0.0000072700836059614686,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.035894038518758706,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 122.62124998960644,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 822.8708330425434,
      "loadMs": 2898.0010410305113,
      "totalMs": 3814.7217910154723,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 128.37883399333805,
      "inferenceMs": 143.08954199077561,
      "audioSeconds": 3.45,
      "realTimeFactor": 0.04147522956254366,
      "audioSecondsPerSecond": 24.110776734629614,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 127.82549997791648,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.17590585350990295,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03705129245012675,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 122.62124998960644,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 127.82695895293728,
      "loadMs": 2898.0010410305113,
      "totalMs": 3041.090583021287,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 76.81533397408202,
      "inferenceMs": 89.38120899256319,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.029547507104979567,
      "audioSecondsPerSecond": 33.84380267503084,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 73.52308399276808,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838505744934082,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024305647585448648,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 82.48104096855968,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 73.52458394598216,
      "loadMs": 2858.7060419959016,
      "totalMs": 2948.087250988465,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 74.25900001544505,
      "inferenceMs": 86.12374996300787,
      "audioSeconds": 3.075,
      "realTimeFactor": 0.02800772356520581,
      "audioSecondsPerSecond": 35.7044369447543,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 73.63699999405071,
      "voice": "en:edith",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.10056133568286896,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023947547329968313,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 82.48104096855968,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 73.63870803965256,
      "loadMs": 2858.7060419959016,
      "totalMs": 2944.8297919589095,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello. This is a local speech synthesis test. hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 486.5083339973353,
      "inferenceMs": 576.3545419904403,
      "audioSeconds": 22.925,
      "realTimeFactor": 0.025140874241676787,
      "audioSecondsPerSecond": 39.7758642116856,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 333.8560000411235,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.18483532965183258,
      "clippedFraction": 0.0000072700836059614686,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021082686324834402,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 82.48104096855968,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 483.32058399682865,
      "loadMs": 2858.7060419959016,
      "totalMs": 3435.060583986342,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 82.61833299184218,
      "inferenceMs": 96.9312500092201,
      "audioSeconds": 3.45,
      "realTimeFactor": 0.028096014495426112,
      "audioSecondsPerSecond": 35.59223676236339,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 82.0192500250414,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.17590585350990295,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023774057964159958,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 82.48104096855968,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 82.02049997635186,
      "loadMs": 2858.7060419959016,
      "totalMs": 2955.6372920051217,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 85.2735840016976,
      "inferenceMs": 98.1710419873707,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.03245323702061841,
      "audioSecondsPerSecond": 30.813567206398336,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 81.92579203750938,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13839849829673767,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134430720,
      "synthesisRealTimeFactor": 0.027083540162117767,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 510.46170794870704,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 81.92770899040624,
      "loadMs": 3461.4504170021974,
      "totalMs": 3559.621458989568,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 892.0173750375398,
      "inferenceMs": 904.0385420084931,
      "audioSeconds": 3.075,
      "realTimeFactor": 0.2939962738239002,
      "audioSecondsPerSecond": 3.4014036538401387,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 891.376040992327,
      "voice": "en:edith",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.10067830979824066,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 170901504,
      "synthesisRealTimeFactor": 0.28987921399586813,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 510.46170794870704,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 891.3785830372944,
      "loadMs": 3461.4504170021974,
      "totalMs": 4365.48895901069,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello. This is a local speech synthesis test. hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2128.6981670418754,
      "inferenceMs": 2217.172584030777,
      "audioSeconds": 22.925,
      "realTimeFactor": 0.09671418032849628,
      "audioSecondsPerSecond": 10.339745387940342,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1153.041125042364,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.18480215966701508,
      "clippedFraction": 0.0000072700836059614686,
      "driverAllocatedMemoryBytes": 1322631168,
      "synthesisRealTimeFactor": 0.09268015267372115,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 510.46170794870704,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2124.6925000450574,
      "loadMs": 3461.4504170021974,
      "totalMs": 5678.623001032975,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 914.8811250342987,
      "inferenceMs": 929.3262499850243,
      "audioSeconds": 3.45,
      "realTimeFactor": 0.2693699275318911,
      "audioSecondsPerSecond": 3.7123668895133384,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 914.2763750278391,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.17587395012378693,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1335525376,
      "synthesisRealTimeFactor": 0.2650081765318316,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 510.46170794870704,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 914.2782090348192,
      "loadMs": 3461.4504170021974,
      "totalMs": 4390.776666987222,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 156.2851250055246,
      "inferenceMs": 168.14479103777558,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.05558505488852085,
      "audioSecondsPerSecond": 17.99044728849436,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 152.94866601470858,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838236033916473,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05056220362701815,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 160.26654100278392,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 152.9506659717299,
      "loadMs": 3404.809458006639,
      "totalMs": 3572.9542490444146,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 157.03424997627735,
      "inferenceMs": 168.89716702280566,
      "audioSeconds": 3.075,
      "realTimeFactor": 0.05492590797489615,
      "audioSecondsPerSecond": 18.206344453277847,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 156.5235829912126,
      "voice": "en:edith",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1006636768579483,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05090276422963395,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 160.26654100278392,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 156.52600000612438,
      "loadMs": 3404.809458006639,
      "totalMs": 3573.7066250294447,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello. This is a local speech synthesis test. hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1139.425541041419,
      "inferenceMs": 1227.7579590445384,
      "audioSeconds": 22.925,
      "realTimeFactor": 0.053555418060830466,
      "audioSecondsPerSecond": 18.672247107923955,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 851.0018750093877,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1845390945672989,
      "clippedFraction": 0.0000018175209014903671,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04954473827706575,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 160.26654100278392,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1135.8131250017323,
      "loadMs": 3404.809458006639,
      "totalMs": 4632.5674170511775,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 176.39491596492007,
      "inferenceMs": 190.32895797863603,
      "audioSeconds": 3.45,
      "realTimeFactor": 0.05516781390685102,
      "audioSecondsPerSecond": 18.126511260505374,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 175.69066601572558,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1759212464094162,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05092538636771665,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 160.26654100278392,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 175.69258296862245,
      "loadMs": 3404.809458006639,
      "totalMs": 3595.138415985275,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 77.88741600234061,
      "inferenceMs": 89.51795799657702,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.02959271338729819,
      "audioSecondsPerSecond": 33.79210236359133,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 74.46291600354016,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838636875152588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024616583789244664,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 79.77683399803936,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 74.46516596246511,
      "loadMs": 3326.0908750235103,
      "totalMs": 3415.6088330200873,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 75.6859999964945,
      "inferenceMs": 87.36516698263586,
      "audioSeconds": 3.075,
      "realTimeFactor": 0.028411436417117353,
      "audioSecondsPerSecond": 35.19709406165466,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 75.15216700267047,
      "voice": "en:edith",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.10066390782594681,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024440325190865895,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 79.77683399803936,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 75.15399996191263,
      "loadMs": 3326.0908750235103,
      "totalMs": 3413.456042006146,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hello. This is a local speech synthesis test. hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test, hello. This is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 514.2571250326,
      "inferenceMs": 604.4227500096895,
      "audioSeconds": 22.925,
      "realTimeFactor": 0.026365223555493544,
      "audioSecondsPerSecond": 37.92875102671514,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 359.9497500108555,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.18454042077064514,
      "clippedFraction": 0.0000018175209014903671,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022231085061409483,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 79.77683399803936,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 509.64762503281236,
      "loadMs": 3326.0908750235103,
      "totalMs": 3930.5136250332,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " The meeting starts at 9.30. Please bring three notebooks.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 80.88508300716057,
      "inferenceMs": 94.77083396632224,
      "audioSeconds": 3.45,
      "realTimeFactor": 0.02746980694676007,
      "audioSecondsPerSecond": 36.403604944808144,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 80.33737499499694,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.17592327296733856,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02328678724302006,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 79.77683399803936,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 80.3394159884192,
      "loadMs": 3326.0908750235103,
      "totalMs": 3420.8617089898326,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2960.7432080083527,
      "warmupMs": 86.77716698730364,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 79.61462502134964,
      "inferenceMs": 91.84887498850003,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.030363264459008273,
      "audioSecondsPerSecond": 32.93453513043841,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 76.32787502370775,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838505744934082,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.025232920330787494,
      "synthesisMs": 76.32958400063217,
      "totalMs": 3052.5920829968527,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 682786816,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2960.7432080083527,
      "warmupMs": 86.77716698730364,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 76.6318750102073,
      "inferenceMs": 89.1695839818567,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.02947754842375428,
      "audioSecondsPerSecond": 33.924123730525594,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 76.10429101623595,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838505744934082,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02515889785044696,
      "synthesisMs": 76.10566599760205,
      "totalMs": 3049.9127919902094,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 684982272,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2960.7432080083527,
      "warmupMs": 86.77716698730364,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 76.07279199874029,
      "inferenceMs": 88.62304198555648,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.029296873383655036,
      "audioSecondsPerSecond": 34.13333521651181,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 75.5347500089556,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838505744934082,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024970743809018498,
      "synthesisMs": 75.53650002228096,
      "totalMs": 3049.366249993909,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 685080576,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2960.7432080083527,
      "warmupMs": 86.77716698730364,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 78.51004198892042,
      "inferenceMs": 90.73470899602398,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.029994945122652557,
      "audioSecondsPerSecond": 33.33895081024127,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 77.98262499272823,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838505744934082,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02577991734743734,
      "synthesisMs": 77.98424997599795,
      "totalMs": 3051.4779170043766,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 689356800,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2968.9286250504665,
      "warmupMs": 92.18637499725446,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 86.1319589894265,
      "inferenceMs": 98.42379100155085,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.03253679041373582,
      "audioSecondsPerSecond": 30.73443899303102,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 82.56583398906514,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838505744934082,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02729504132520192,
      "synthesisMs": 82.5675000087358,
      "totalMs": 3067.3524160520174,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 685998080,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2968.9286250504665,
      "warmupMs": 92.18637499725446,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 81.15174999693409,
      "inferenceMs": 94.14908301550895,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.03112366380677982,
      "audioSecondsPerSecond": 32.12989338942047,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 80.61245898716152,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838505744934082,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.026649283976409553,
      "synthesisMs": 80.6140840286389,
      "totalMs": 3063.0777080659755,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 688308224,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2968.9286250504665,
      "warmupMs": 92.18637499725446,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 80.30591701390222,
      "inferenceMs": 92.88166702026501,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.030704683312484302,
      "audioSecondsPerSecond": 32.568321575666836,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 79.78466700296849,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838505744934082,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.026375606294321126,
      "synthesisMs": 79.78620904032141,
      "totalMs": 3061.8102920707315,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 690438144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2968.9286250504665,
      "warmupMs": 92.18637499725446,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 79.23504197970033,
      "inferenceMs": 91.67024999624118,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.030304214874790472,
      "audioSecondsPerSecond": 32.99871005177837,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 78.68462498299778,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838505744934082,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.026011942146915542,
      "synthesisMs": 78.68612499441952,
      "totalMs": 3060.5988750467077,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 690438144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3636.7795830010436,
      "warmupMs": 526.4862909680232,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 85.85083397338167,
      "inferenceMs": 98.28833397477865,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.03249201123133179,
      "audioSecondsPerSecond": 30.776795960100742,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 82.40862499224022,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13839849829673767,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134430720,
      "synthesisRealTimeFactor": 0.02724311304902119,
      "synthesisMs": 82.4104169732891,
      "totalMs": 3735.0679169758223,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 834240512,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3636.7795830010436,
      "warmupMs": 526.4862909680232,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 85.44675004668534,
      "inferenceMs": 98.28516701236367,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.032490964301607826,
      "audioSecondsPerSecond": 30.777787655582593,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 84.80104204500094,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13839849829673767,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134430720,
      "synthesisRealTimeFactor": 0.02803399439223788,
      "synthesisMs": 84.80283303651959,
      "totalMs": 3735.0647500134073,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 834273280,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3636.7795830010436,
      "warmupMs": 526.4862909680232,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 86.29391703289002,
      "inferenceMs": 98.94179197726771,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.032708030405708334,
      "audioSecondsPerSecond": 30.573531563841154,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.75787500012666,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13839849829673767,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134430720,
      "synthesisRealTimeFactor": 0.028350275387017686,
      "synthesisMs": 85.7595830457285,
      "totalMs": 3735.7213749783114,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 834306048,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3636.7795830010436,
      "warmupMs": 526.4862909680232,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 85.60462499735877,
      "inferenceMs": 97.26274997228757,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.03215297519745044,
      "audioSecondsPerSecond": 31.101320915375037,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.08333301870152,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13839849829673767,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134414336,
      "synthesisRealTimeFactor": 0.028127286283006845,
      "synthesisMs": 85.08504100609571,
      "totalMs": 3734.042332973331,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 834322432,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3537.6557080307975,
      "warmupMs": 531.0099999769591,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 90.01649997662753,
      "inferenceMs": 102.72341698873788,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.03395815437644228,
      "audioSecondsPerSecond": 29.448007948680747,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 86.39262500219047,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13839849829673767,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134430720,
      "synthesisRealTimeFactor": 0.028560110079201537,
      "synthesisMs": 86.39433298958465,
      "totalMs": 3640.3791250195354,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 830898176,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3537.6557080307975,
      "warmupMs": 531.0099999769591,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 82.7718329965137,
      "inferenceMs": 95.70366702973843,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.03163757587759948,
      "audioSecondsPerSecond": 31.60798424850354,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 82.2647080058232,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13839849829673767,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134430720,
      "synthesisRealTimeFactor": 0.027195509746362846,
      "synthesisMs": 82.26641698274761,
      "totalMs": 3633.359375060536,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 831602688,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3537.6557080307975,
      "warmupMs": 531.0099999769591,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 83.12991703860462,
      "inferenceMs": 96.63724998245016,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.03194619834130584,
      "audioSecondsPerSecond": 31.30262916783493,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 82.59429200552404,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13839849829673767,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134414336,
      "synthesisRealTimeFactor": 0.027304518032695884,
      "synthesisMs": 82.59616704890504,
      "totalMs": 3634.2929580132477,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 832192512,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3537.6557080307975,
      "warmupMs": 531.0099999769591,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 90.4026249772869,
      "inferenceMs": 102.88841696456075,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.03401269982299529,
      "audioSecondsPerSecond": 29.40078280183805,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.88637500442564,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13839849829673767,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134414336,
      "synthesisRealTimeFactor": 0.029715054864756577,
      "synthesisMs": 89.88804096588865,
      "totalMs": 3640.5441249953583,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 832208896,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3500.898958009202,
      "warmupMs": 81.09912497457117,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 79.22499999403954,
      "inferenceMs": 91.10329201212153,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.03011679074780877,
      "audioSecondsPerSecond": 33.204069064787646,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 75.82916697720066,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838636875152588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02506818181704268,
      "synthesisMs": 75.8312499965541,
      "totalMs": 3592.0022500213236,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 670941184,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3500.898958009202,
      "warmupMs": 81.09912497457117,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 75.54216700373217,
      "inferenceMs": 87.42154197534546,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.028899683297634864,
      "audioSecondsPerSecond": 34.60245531762764,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 74.98858397593722,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13844309747219086,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02479038578419646,
      "synthesisMs": 74.99091699719429,
      "totalMs": 3588.3204999845475,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 671760384,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3500.898958009202,
      "warmupMs": 81.09912497457117,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 76.11925003584474,
      "inferenceMs": 87.89912500651553,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.02905756198562497,
      "audioSecondsPerSecond": 34.414449515575626,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 75.55379200493917,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1385612189769745,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024977259180116014,
      "synthesisMs": 75.55620901985094,
      "totalMs": 3588.7980830157176,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 672186368,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3500.898958009202,
      "warmupMs": 81.09912497457117,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 73.72012501582503,
      "inferenceMs": 85.08941699983552,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.02812873289250761,
      "audioSecondsPerSecond": 35.55083706832599,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 73.2058749999851,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1384536474943161,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024201060830007406,
      "synthesisMs": 73.2082090107724,
      "totalMs": 3585.9883750090376,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 672186368,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3479.3408749974333,
      "warmupMs": 81.20495796902105,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 79.11833300022408,
      "inferenceMs": 90.29158297926188,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.029848457183227067,
      "audioSecondsPerSecond": 33.502569123135,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 75.74533304432407,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13838636875152588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02504046810098177,
      "synthesisMs": 75.74741600546986,
      "totalMs": 3569.632457976695,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 671006720,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3479.3408749974333,
      "warmupMs": 81.20495796902105,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 74.89345903741196,
      "inferenceMs": 86.37895795982331,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.02855502742473498,
      "audioSecondsPerSecond": 35.020102944596665,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 74.34120902325958,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13844309747219086,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024576418848385004,
      "synthesisMs": 74.34366701636463,
      "totalMs": 3565.7198329572566,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 671268864,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3479.3408749974333,
      "warmupMs": 81.20495796902105,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 74.18512500589713,
      "inferenceMs": 86.18183300131932,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.02848986214919647,
      "audioSecondsPerSecond": 35.10020493476498,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 73.68491601664573,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1385612189769745,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024359200981037676,
      "synthesisMs": 73.68658296763897,
      "totalMs": 3565.5227079987526,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 671268864,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hello, this is a local speech synthesis test.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 3479.3408749974333,
      "warmupMs": 81.20495796902105,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "en"
      ],
      "sampleRate": 24000,
      "durationMs": 75.12574997963384,
      "inferenceMs": 86.49883297039196,
      "audioSeconds": 3.025,
      "realTimeFactor": 0.0285946555274023,
      "audioSecondsPerSecond": 34.97157009084088,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 74.60749999154359,
      "voice": "en:eleanor",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1384536474943161,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024664270083520037,
      "synthesisMs": 74.60941700264812,
      "totalMs": 3565.8397079678252,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 671268864,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Pico · Spanish Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch6496 ms / 3440 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6497 ms / 3440 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6497 ms / 3513 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6498 ms / 3513 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 3513 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64101 ms / 3513 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64101 ms / 3440 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64102 ms / 3440 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64108 ms / 4226 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642357 ms / 4226 ms0.086 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64883 ms / 4226 ms0.235 RTF
FP32cpuPassedVery fastmacos · aarch64664 ms / 3148 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64687 ms / 2712 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch6494 ms / 2892 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch6494 ms / 2892 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch6494 ms / 2892 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch6494 ms / 2724 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch6495 ms / 2892 ms0.027 RTF
Q8cpuPassedVery fastmacos · aarch64100 ms / 2712 ms0.027 RTF
Q8cpuPassedVery fastmacos · aarch6497 ms / 2724 ms0.027 RTF
Q8cpuPassedVery fastmacos · aarch6497 ms / 2505 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch6498 ms / 3009 ms0.027 RTF
Q8cpuPassedVery fastmacos · aarch6498 ms / 2724 ms0.027 RTF
Q8cpuPassedVery fastmacos · aarch6498 ms / 2505 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch6498 ms / 3009 ms0.027 RTF
Q8cpuPassedVery fastmacos · aarch6498 ms / 2505 ms0.027 RTF
Q8cpuPassedVery fastmacos · aarch6499 ms / 2505 ms0.028 RTF
Q8cpuPassedVery fastmacos · aarch6499 ms / 2724 ms0.028 RTF
FP32cpuPassedVery fastmacos · aarch6499 ms / 3009 ms0.028 RTF
FP32cpuPassedVery fastmacos · aarch6499 ms / 3009 ms0.028 RTF
FP32cpuPassedVery fastmacos · aarch64105 ms / 3148 ms0.028 RTF
Q8cpuPassedVery fastmacos · aarch64100 ms / 2712 ms0.028 RTF
FP32cpuPassedVery fastmacos · aarch64101 ms / 3148 ms0.028 RTF
FP32cpuPassedVery fastmacos · aarch641071 ms / 3157 ms0.039 RTF
FP32cpuPassedVery fastmacos · aarch64152 ms / 3157 ms0.042 RTF
FP32cpuPassedVery fastmacos · aarch64161 ms / 3157 ms0.043 RTF
Q8cpuPassedVery fastmacos · aarch641437 ms / 2773 ms0.053 RTF
Q8cpuPassedVery fastmacos · aarch64206 ms / 2773 ms0.055 RTF
Q8cpuPassedVery fastmacos · aarch64198 ms / 2773 ms0.055 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:04:49.847905+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
da8c9d029e16d9302f1827adb303d92316e25f73
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 138.5573330335319,
      "inferenceMs": 151.51662495918572,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.042382272715856144,
      "audioSecondsPerSecond": 23.594770547212253,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 135.64837502781302,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11756931245326996,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0379439860165536,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 142.40454201353714,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 135.64975000917912,
      "loadMs": 3157.019875012338,
      "totalMs": 3308.536499971524,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 148.45575002254918,
      "inferenceMs": 161.24137496808544,
      "audioSeconds": 3.75,
      "realTimeFactor": 0.04299769999148945,
      "audioSecondsPerSecond": 23.2570579402603,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 147.8550000465475,
      "voice": "es:elena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06243574246764183,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03942852214289208,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 142.40454201353714,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 147.8569580358453,
      "loadMs": 3157.019875012338,
      "totalMs": 3318.2612499804236,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. hola. Esta es una prueba de voz en este dispositivo. hola, Esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 982.1401659864932,
      "inferenceMs": 1070.908125024289,
      "audioSeconds": 27.4,
      "realTimeFactor": 0.03908423813957259,
      "audioSecondsPerSecond": 25.58576161645851,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 524.8502079630271,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13309553265571594,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03570400696990495,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 142.40454201353714,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 978.2897909753956,
      "loadMs": 3157.019875012338,
      "totalMs": 4227.928000036627,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 89.32791702682152,
      "inferenceMs": 101.44066601060331,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.02837501147149743,
      "audioSecondsPerSecond": 35.24227650108602,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 86.41862502554432,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11756931245326996,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024173484750456744,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 93.29708298901096,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 86.42020798288286,
      "loadMs": 3147.73987501394,
      "totalMs": 3249.1805410245433,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 91.61691699409857,
      "inferenceMs": 104.55466597341001,
      "audioSeconds": 3.75,
      "realTimeFactor": 0.027881244259576003,
      "audioSecondsPerSecond": 35.86640505315839,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 90.99879203131422,
      "voice": "es:elena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06243574246764183,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02426673333781461,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 93.29708298901096,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 91.00025001680478,
      "loadMs": 3147.73987501394,
      "totalMs": 3252.29454098735,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. hola. Esta es una prueba de voz en este dispositivo. hola, Esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 574.7703750384971,
      "inferenceMs": 663.8251669937745,
      "audioSeconds": 27.4,
      "realTimeFactor": 0.0242271958756852,
      "audioSecondsPerSecond": 41.275928305166175,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 294.21162500511855,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13309553265571594,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020824463211210703,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 93.29708298901096,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 570.5902919871733,
      "loadMs": 3147.73987501394,
      "totalMs": 3811.5650420077145,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 94.60512502118945,
      "inferenceMs": 108.23200002778322,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.030274685322456845,
      "audioSecondsPerSecond": 33.03089658402594,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 91.7849590186961,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11763092130422592,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134987776,
      "synthesisRealTimeFactor": 0.025674533994438555,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1043.0842919740826,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 91.78645903011784,
      "loadMs": 4226.120749954134,
      "totalMs": 4334.352749981917,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 869.9491249863058,
      "inferenceMs": 882.9590000095777,
      "audioSeconds": 3.75,
      "realTimeFactor": 0.2354557333358874,
      "audioSecondsPerSecond": 4.247082820334039,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 869.3586249719374,
      "voice": "es:elena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06242169067263603,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 180256768,
      "synthesisRealTimeFactor": 0.2318295223948856,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1043.0842919740826,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 869.3607089808211,
      "loadMs": 4226.120749954134,
      "totalMs": 5109.079749963712,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. hola. Esta es una prueba de voz en este dispositivo. hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2267.6635840325616,
      "inferenceMs": 2356.7123329848982,
      "audioSeconds": 27.4,
      "realTimeFactor": 0.08601139901404738,
      "audioSecondsPerSecond": 11.62636594059678,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1132.8167500323616,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13309134542942047,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1336983552,
      "synthesisRealTimeFactor": 0.08259939934368361,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1043.0842919740826,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2263.223542016931,
      "loadMs": 4226.120749954134,
      "totalMs": 6582.8330829390325,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 186.74066703533754,
      "inferenceMs": 198.23891698615626,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.05545144531081294,
      "audioSecondsPerSecond": 18.03379505069459,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 185.05887500941753,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11655822396278381,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05176537315250485,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 190.26450003730133,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 185.06120902020484,
      "loadMs": 2773.114667041227,
      "totalMs": 2971.3535840273835,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 193.78945900825784,
      "inferenceMs": 205.9450420201756,
      "audioSeconds": 3.75,
      "realTimeFactor": 0.05491867787204683,
      "audioSecondsPerSecond": 18.20874133805381,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 193.09491704916582,
      "voice": "es:elena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0620948001742363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.05149441120835642,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 190.26450003730133,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 193.10404203133658,
      "loadMs": 2773.114667041227,
      "totalMs": 2979.059709061403,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. hola. Esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1350.6065409746952,
      "inferenceMs": 1437.204749963712,
      "audioSeconds": 27.375,
      "realTimeFactor": 0.05250063013566071,
      "audioSecondsPerSecond": 19.047390429715175,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 732.2724999976344,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13321681320667267,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.049174662098951945,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 190.26450003730133,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1346.1563749588095,
      "loadMs": 2773.114667041227,
      "totalMs": 4210.319417004939,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 87.20549999270588,
      "inferenceMs": 100.37970897974446,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.028078240274054395,
      "audioSecondsPerSecond": 35.61476752957509,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.48991702264175,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11655250191688538,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023913811192150924,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 89.11058399826288,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 85.49187501193956,
      "loadMs": 2712.39516598871,
      "totalMs": 2812.7748749684542,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 87.9191659623757,
      "inferenceMs": 100.24633299326524,
      "audioSeconds": 3.75,
      "realTimeFactor": 0.02673235546487073,
      "audioSecondsPerSecond": 37.407852118161095,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 87.15049998136237,
      "voice": "es:elena",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06208724528551102,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023240666665757696,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 89.11058399826288,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 87.15249999659136,
      "loadMs": 2712.39516598871,
      "totalMs": 2812.641498981975,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. hola. Esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo. Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 599.4285409688018,
      "inferenceMs": 687.2982920031063,
      "audioSeconds": 27.375,
      "realTimeFactor": 0.0251067869224879,
      "audioSecondsPerSecond": 39.82986764046298,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 309.8198749939911,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13321681320667267,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021728188711639545,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 89.11058399826288,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 594.8091659811325,
      "loadMs": 2712.39516598871,
      "totalMs": 3399.693457991816,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3008.9793750084937,
      "warmupMs": 85.48300003167242,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 85.63354099169374,
      "inferenceMs": 97.80145902186632,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.027357051474647922,
      "audioSecondsPerSecond": 36.55364690623589,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 83.02162500331178,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11756931245326996,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0232233099274702,
      "synthesisMs": 83.02333299070597,
      "totalMs": 3106.78083403036,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 519897088,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3008.9793750084937,
      "warmupMs": 85.48300003167242,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 86.67824999429286,
      "inferenceMs": 99.470583954826,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.027823939567783496,
      "audioSecondsPerSecond": 35.940273574985405,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 86.03720896644518,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11756931245326996,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.024066841672663087,
      "synthesisMs": 86.03895897977054,
      "totalMs": 3108.4499589633197,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 542883840,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3008.9793750084937,
      "warmupMs": 85.48300003167242,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 86.32070804014802,
      "inferenceMs": 99.41262501524761,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.02780772727699234,
      "audioSecondsPerSecond": 35.961227253094634,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.6811249977909,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11756931245326996,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023967167834744795,
      "synthesisMs": 85.68262500921264,
      "totalMs": 3108.3920000237413,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 544620544,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3008.9793750084937,
      "warmupMs": 85.48300003167242,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 85.08658298524097,
      "inferenceMs": 97.6239160518162,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.027307389105403132,
      "audioSecondsPerSecond": 36.62012490978629,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 84.50358302798122,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11756931245326996,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023637739029693437,
      "synthesisMs": 84.50491703115404,
      "totalMs": 3106.60329106031,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 545406976,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2892.0785420341417,
      "warmupMs": 85.39058302994817,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 83.20941601414233,
      "inferenceMs": 95.03162500914186,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.02658227272982989,
      "audioSecondsPerSecond": 37.6190557580815,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 80.49691596534103,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11756931245326996,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022517074407501655,
      "synthesisMs": 80.49854100681841,
      "totalMs": 2987.1101670432836,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 518291456,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2892.0785420341417,
      "warmupMs": 85.39058302994817,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 81.12720801727846,
      "inferenceMs": 93.7135000131093,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.026213566437233368,
      "audioSecondsPerSecond": 38.14818568829362,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 80.52616700297222,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11756931245326996,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022525209786982595,
      "synthesisMs": 80.52762498846278,
      "totalMs": 2985.792042047251,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 523141120,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2892.0785420341417,
      "warmupMs": 85.39058302994817,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 81.69012499274686,
      "inferenceMs": 94.15604104287922,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.026337354137868314,
      "audioSecondsPerSecond": 37.968886121411195,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 81.11379202455282,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11756931245326996,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02268954544894762,
      "synthesisMs": 81.11512497998774,
      "totalMs": 2986.234583077021,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 523157504,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2892.0785420341417,
      "warmupMs": 85.39058302994817,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 81.39612502418458,
      "inferenceMs": 94.02916702674702,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.02630186490258658,
      "audioSecondsPerSecond": 38.020117725631614,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 80.8070829953067,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11756931245326996,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022603787963727973,
      "synthesisMs": 80.8085419703275,
      "totalMs": 2986.1077090608887,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 523206656,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3440.4532909975387,
      "warmupMs": 429.08437497681007,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 90.32404201570898,
      "inferenceMs": 102.18691697809845,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.0285837530008667,
      "audioSecondsPerSecond": 34.98490908348104,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 87.72991702426225,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11763092130422592,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135004160,
      "synthesisRealTimeFactor": 0.024540349662017364,
      "synthesisMs": 87.73175004171208,
      "totalMs": 3542.640207975637,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 645644288,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3440.4532909975387,
      "warmupMs": 429.08437497681007,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 84.30624997708946,
      "inferenceMs": 96.62195795681328,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.027027121106800917,
      "audioSecondsPerSecond": 36.99987120523787,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 83.73787498567253,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11763092130422592,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135004160,
      "synthesisRealTimeFactor": 0.023423752998216793,
      "synthesisMs": 83.73991696862504,
      "totalMs": 3537.075248954352,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 645709824,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3440.4532909975387,
      "warmupMs": 429.08437497681007,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 83.5214169928804,
      "inferenceMs": 96.02754103252664,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.02686085063846899,
      "audioSecondsPerSecond": 37.228902891401425,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 82.97937497263774,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11763092130422592,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135004160,
      "synthesisRealTimeFactor": 0.023211526987305678,
      "synthesisMs": 82.9812089796178,
      "totalMs": 3536.4808320300654,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 645742592,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3440.4532909975387,
      "warmupMs": 429.08437497681007,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 88.5562080075033,
      "inferenceMs": 100.97629198571667,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.028245116639361308,
      "audioSecondsPerSecond": 35.40435016672718,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.01787497941405,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11763092130422592,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135004160,
      "synthesisRealTimeFactor": 0.02462080419324078,
      "synthesisMs": 88.01937499083579,
      "totalMs": 3541.4295829832554,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 645758976,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3513.1702499929816,
      "warmupMs": 435.3457909892313,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 86.81091596372426,
      "inferenceMs": 98.4909160179086,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.027549906578435972,
      "audioSecondsPerSecond": 36.297763738434085,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 84.23237496754155,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11763092130422592,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135004160,
      "synthesisRealTimeFactor": 0.023562027975083232,
      "synthesisMs": 84.23425001092255,
      "totalMs": 3611.66116601089,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642957312,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3513.1702499929816,
      "warmupMs": 435.3457909892313,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 85.95120802056044,
      "inferenceMs": 98.9357500220649,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.02767433567050766,
      "audioSecondsPerSecond": 36.13456206884462,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.4123750468716,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11763092130422592,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135004160,
      "synthesisRealTimeFactor": 0.023892062937619387,
      "synthesisMs": 85.4141250019893,
      "totalMs": 3612.1060000150464,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643039232,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3513.1702499929816,
      "warmupMs": 435.3457909892313,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 85.10812500026077,
      "inferenceMs": 97.25141699891537,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.027203193566130175,
      "audioSecondsPerSecond": 36.76038982588677,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 84.57254199311137,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11763092130422592,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135004160,
      "synthesisRealTimeFactor": 0.02365716781884015,
      "synthesisMs": 84.57437495235354,
      "totalMs": 3610.421666991897,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643055616,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3513.1702499929816,
      "warmupMs": 435.3457909892313,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 88.3917500032112,
      "inferenceMs": 100.72170797502622,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.028173904328678664,
      "audioSecondsPerSecond": 35.4938381395043,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 87.86162501201034,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11763092130422592,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135004160,
      "synthesisRealTimeFactor": 0.024577202796415017,
      "synthesisMs": 87.86349999718368,
      "totalMs": 3613.891957968008,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643072000,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2505.157333042007,
      "warmupMs": 88.12158298678696,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 86.90258400747553,
      "inferenceMs": 97.92408300563693,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.02739135188968865,
      "audioSecondsPerSecond": 36.50787314285299,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.27962496737018,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11655250191688538,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023855093420463633,
      "synthesisMs": 85.28195897815749,
      "totalMs": 2603.081416047644,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 349044736,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2505.157333042007,
      "warmupMs": 88.12158298678696,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 86.22762496815994,
      "inferenceMs": 98.62820798298344,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.027588309925310053,
      "audioSecondsPerSecond": 36.2472366994319,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.54504200583324,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11589865386486053,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023929394123802235,
      "synthesisMs": 85.54758399259299,
      "totalMs": 2603.7855410249904,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 353271808,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2505.157333042007,
      "warmupMs": 88.12158298678696,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 85.15858399914578,
      "inferenceMs": 97.35208301572129,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.027231351892509453,
      "audioSecondsPerSecond": 36.72237808638031,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 84.55520903225988,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1163053810596466,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02365235441162453,
      "synthesisMs": 84.55716702155769,
      "totalMs": 2602.5094160577282,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 353271808,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2505.157333042007,
      "warmupMs": 88.12158298678696,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 85.69295797497034,
      "inferenceMs": 97.75745897786692,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.027344743770032705,
      "audioSecondsPerSecond": 36.570099482735216,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.07887495215982,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1161695271730423,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023798811187563867,
      "synthesisMs": 85.08074999554083,
      "totalMs": 2602.914792019874,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 353271808,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2723.552832962014,
      "warmupMs": 86.91108302446082,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 86.9020419777371,
      "inferenceMs": 98.98083400912583,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.027686946575979254,
      "audioSecondsPerSecond": 36.118103426673414,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.29620798071846,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11655250191688538,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023859801956198433,
      "synthesisMs": 85.2987919934094,
      "totalMs": 2822.5336669711396,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 351977472,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2723.552832962014,
      "warmupMs": 86.91108302446082,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 84.75245902081951,
      "inferenceMs": 96.56725003151223,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.02701181819063279,
      "audioSecondsPerSecond": 37.02083262010041,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 84.03487503528595,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.11589865386486053,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023506760004569183,
      "synthesisMs": 84.03666701633483,
      "totalMs": 2820.120082993526,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 352960512,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2723.552832962014,
      "warmupMs": 86.91108302446082,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 83.29433301696554,
      "inferenceMs": 94.49470898834988,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.02643208643030766,
      "audioSecondsPerSecond": 37.832806072144805,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 82.68770802533254,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1163053810596466,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023130058193696547,
      "synthesisMs": 82.68995804246515,
      "totalMs": 2818.0475419503637,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 353189888,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Hola, esta es una prueba de voz en este dispositivo.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2723.552832962014,
      "warmupMs": 86.91108302446082,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "es"
      ],
      "sampleRate": 24000,
      "durationMs": 85.41612501721829,
      "inferenceMs": 97.70370798651129,
      "audioSeconds": 3.575,
      "realTimeFactor": 0.027329708527695465,
      "audioSecondsPerSecond": 36.59021825961359,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 84.73283401690423,
      "voice": "es:sofia",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1161695271730423,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02370209791339361,
      "synthesisMs": 84.73500004038215,
      "totalMs": 2821.256540948525,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 355336192,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Pico · French Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch6499 ms / 3722 ms0.024 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64101 ms / 3649 ms0.024 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64101 ms / 3572 ms0.024 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64101 ms / 3722 ms0.024 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64102 ms / 3649 ms0.024 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64104 ms / 3722 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64104 ms / 3649 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64104 ms / 3649 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64105 ms / 3722 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch641685 ms / 3572 ms0.056 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64863 ms / 3572 ms0.204 RTF
FP32cpuPassedVery fastmacos · aarch64694 ms / 2978 ms0.023 RTF
Q8cpuPassedVery fastmacos · aarch64101 ms / 2460 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64733 ms / 2460 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64101 ms / 2618 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64101 ms / 2657 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64101 ms / 2618 ms0.024 RTF
FP32cpuPassedVery fastmacos · aarch64103 ms / 2978 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64102 ms / 2618 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64102 ms / 2657 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64103 ms / 2657 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64103 ms / 2618 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64103 ms / 2460 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64104 ms / 2657 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64104 ms / 3338 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64105 ms / 3219 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64105 ms / 3338 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64105 ms / 2978 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64105 ms / 3219 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64106 ms / 3219 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64107 ms / 3338 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64109 ms / 3338 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64153 ms / 3219 ms0.037 RTF
FP32cpuPassedVery fastmacos · aarch641151 ms / 3402 ms0.038 RTF
FP32cpuPassedVery fastmacos · aarch64165 ms / 3402 ms0.040 RTF
FP32cpuPassedVery fastmacos · aarch64172 ms / 3402 ms0.041 RTF
Q8cpuPassedVery fastmacos · aarch64205 ms / 2671 ms0.048 RTF
Q8cpuPassedVery fastmacos · aarch641499 ms / 2671 ms0.049 RTF
Q8cpuPassedVery fastmacos · aarch64209 ms / 2671 ms0.050 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:06:40.614896+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
da8c9d029e16d9302f1827adb303d92316e25f73
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 149.94045800995082,
      "inferenceMs": 165.1795840007253,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.039563972215742586,
      "audioSecondsPerSecond": 25.275520732523866,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 146.4726670528762,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0448574535548687,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.035083712577752904,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 185.00291695818305,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 146.47450001211837,
      "loadMs": 3402.2269579581916,
      "totalMs": 3567.406541958917,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 155.73037503054366,
      "inferenceMs": 171.7700839508325,
      "audioSeconds": 4.225,
      "realTimeFactor": 0.04065564117179467,
      "audioSecondsPerSecond": 24.596832596352257,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 154.42520804936066,
      "voice": "fr:lea",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05949798598885536,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.036556094680973414,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 185.00291695818305,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 154.44950002711266,
      "loadMs": 3402.2269579581916,
      "totalMs": 3573.997041909024,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Bonjour, ceci est un test de synthèse vocale locale. bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour. Ceci est un test de synthèse vocale locale, bonjour. Ceci est un test de synthèse vocale locale, bonjour. Ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1054.3365410412662,
      "inferenceMs": 1150.6397500052117,
      "audioSeconds": 30.325,
      "realTimeFactor": 0.03794360263825925,
      "audioSecondsPerSecond": 26.354903869662635,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 676.6691250377335,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.051634859293699265,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.034625614180958234,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 185.00291695818305,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1050.0217500375584,
      "loadMs": 3402.2269579581916,
      "totalMs": 4552.866707963403,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 90.59058304410428,
      "inferenceMs": 104.99612498097122,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.025148772450532028,
      "audioSecondsPerSecond": 39.763372226895505,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 87.75050000986084,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0448574535548687,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02101841318549332,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 94.14104197639972,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 87.7518750494346,
      "loadMs": 2978.0713329673745,
      "totalMs": 3083.0674579483457,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 89.28549999836832,
      "inferenceMs": 103.24679198674858,
      "audioSeconds": 4.225,
      "realTimeFactor": 0.024437110529407946,
      "audioSecondsPerSecond": 40.921368293382585,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.60091696260497,
      "voice": "fr:lea",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05949798598885536,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02097094673994025,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 94.14104197639972,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 88.60224997624755,
      "loadMs": 2978.0713329673745,
      "totalMs": 3081.318124954123,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Bonjour, ceci est un test de synthèse vocale locale. bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour. Ceci est un test de synthèse vocale locale, bonjour. Ceci est un test de synthèse vocale locale, bonjour. Ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 600.2262079855427,
      "inferenceMs": 694.3205840070732,
      "audioSeconds": 30.325,
      "realTimeFactor": 0.022895979686960367,
      "audioSecondsPerSecond": 43.67578997152571,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 361.86616600025445,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.051634859293699265,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.019654523197758057,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 94.14104197639972,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 596.023415972013,
      "loadMs": 2978.0713329673745,
      "totalMs": 3672.3919169744477,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 86.22250001644716,
      "inferenceMs": 100.71145801339298,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024122504913387542,
      "audioSecondsPerSecond": 41.45506462079809,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 83.40204099658877,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044866710901260376,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134987776,
      "synthesisRealTimeFactor": 0.019977065862831243,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 477.25720796734095,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 83.40424997732043,
      "loadMs": 3571.535083989147,
      "totalMs": 3672.24654200254,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 848.2119590044022,
      "inferenceMs": 863.1845419877209,
      "audioSeconds": 4.225,
      "realTimeFactor": 0.20430403360656116,
      "audioSecondsPerSecond": 4.894665965948336,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 847.3697919980623,
      "voice": "fr:lea",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05945048853754997,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 180322304,
      "synthesisRealTimeFactor": 0.2005615287609178,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 477.25720796734095,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 847.3724590148777,
      "loadMs": 3571.535083989147,
      "totalMs": 4434.719625976868,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Bonjour, ceci est un test de synthèse vocale locale. bonjour. ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour. Ceci est un test de synthèse vocale locale, bonjour. Ceci est un test de synthèse vocale locale, bonjour. Ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1588.4345829836093,
      "inferenceMs": 1685.4904999490827,
      "audioSeconds": 30.325,
      "realTimeFactor": 0.05558089035281394,
      "audioSecondsPerSecond": 17.99179526726261,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 546.6111669666134,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05163951218128204,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1351122944,
      "synthesisRealTimeFactor": 0.05220881147503619,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 477.25720796734095,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1583.2322079804726,
      "loadMs": 3571.535083989147,
      "totalMs": 5257.02558393823,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 195.54241601144895,
      "inferenceMs": 209.32958298362792,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.05013882227152765,
      "audioSecondsPerSecond": 19.944624837505817,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 193.87895800173283,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04487321153283119,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.046438612685187494,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 195.08191698696464,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 193.88120796065778,
      "loadMs": 2670.6540000159293,
      "totalMs": 2879.9835829995573,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 191.10545801231638,
      "inferenceMs": 204.55466699786484,
      "audioSeconds": 4.225,
      "realTimeFactor": 0.04841530579831121,
      "audioSecondsPerSecond": 20.654625298987195,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 190.4062909889035,
      "voice": "fr:lea",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05942503735423088,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.045067061072204415,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 195.08191698696464,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 190.40833303006366,
      "loadMs": 2670.6540000159293,
      "totalMs": 2875.208667013794,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Bonjour, ceci est un test de synthèse vocale locale. bonjour. ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour. Ceci est un test de synthèse vocale locale, bonjour. Ceci est un test de synthèse vocale locale, bonjour. Ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1404.6072089695372,
      "inferenceMs": 1498.5614170436747,
      "audioSeconds": 30.325,
      "realTimeFactor": 0.04941669965519125,
      "audioSecondsPerSecond": 20.236074180946428,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 919.8832089896314,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05163692310452461,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04616913025406391,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 195.08191698696464,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1400.0788749544881,
      "loadMs": 2670.6540000159293,
      "totalMs": 4169.215417059604,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 89.96679197298363,
      "inferenceMs": 103.28333399957046,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024738523113669572,
      "audioSecondsPerSecond": 40.422784957903886,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.21308298502117,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044870197772979736,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021129411249282117,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 90.0530000217259,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 88.21529196575284,
      "loadMs": 2459.997290978208,
      "totalMs": 2563.2806249777786,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 87.70395797910169,
      "inferenceMs": 101.25333297764882,
      "audioSeconds": 4.225,
      "realTimeFactor": 0.023965285911869545,
      "audioSecondsPerSecond": 41.72702147921045,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 86.89308300381526,
      "voice": "fr:lea",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.059424467384815216,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020566933014093947,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 90.0530000217259,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 86.89529198454693,
      "loadMs": 2459.997290978208,
      "totalMs": 2561.250623955857,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Bonjour, ceci est un test de synthèse vocale locale. bonjour. ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour ceci est un test de synthèse vocale locale bonjour. Ceci est un test de synthèse vocale locale, bonjour. Ceci est un test de synthèse vocale locale, bonjour. Ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 638.4313749731518,
      "inferenceMs": 733.1885409657843,
      "audioSeconds": 30.325,
      "realTimeFactor": 0.02417769302442817,
      "audioSecondsPerSecond": 41.3604391035009,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 396.26854099333286,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05163682624697685,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02089128880404544,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 90.0530000217259,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 633.528332982678,
      "loadMs": 2459.997290978208,
      "totalMs": 3193.1858319439925,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3338.1262079929,
      "warmupMs": 94.8458340135403,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 95.32191598555073,
      "inferenceMs": 109.27191702648997,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.02617291425784191,
      "audioSecondsPerSecond": 38.20743804639106,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 92.51633298117667,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0448574535548687,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022159979879454582,
      "synthesisMs": 92.51791599672288,
      "totalMs": 3447.39812501939,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 527302656,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3338.1262079929,
      "warmupMs": 94.8458340135403,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 90.22141702007502,
      "inferenceMs": 104.72550004487857,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.02508395210655774,
      "audioSecondsPerSecond": 39.86612618904532,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.50525004183874,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0448574535548687,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021438792342331238,
      "synthesisMs": 89.50695802923292,
      "totalMs": 3442.8517080377787,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 527581184,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3338.1262079929,
      "warmupMs": 94.8458340135403,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 89.1117500141263,
      "inferenceMs": 103.9365420001559,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024894980119797818,
      "audioSecondsPerSecond": 40.16874065325108,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.42645800905302,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0448574535548687,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02118030946400619,
      "synthesisMs": 88.42779201222584,
      "totalMs": 3442.062749993056,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 528482304,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3338.1262079929,
      "warmupMs": 94.8458340135403,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 91.83950000442564,
      "inferenceMs": 106.53737495886162,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.025517934121883024,
      "audioSecondsPerSecond": 39.18812530918972,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 91.1807079683058,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0448574535548687,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021840000000899425,
      "synthesisMs": 91.18200000375509,
      "totalMs": 3444.663582951762,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 530579456,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3219.0527499769814,
      "warmupMs": 129.12100000539795,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 139.29704198380932,
      "inferenceMs": 152.8872920316644,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.03661971066626692,
      "audioSecondsPerSecond": 27.307698007597114,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 136.46558398613706,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0448574535548687,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03268681652007435,
      "synthesisMs": 136.4674589713104,
      "totalMs": 3371.940042008646,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 541523968,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3219.0527499769814,
      "warmupMs": 129.12100000539795,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 90.07225005188957,
      "inferenceMs": 104.66841596644372,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.025070279273399693,
      "audioSecondsPerSecond": 39.88786838370124,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.36087501933798,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0448574535548687,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021404171499150422,
      "synthesisMs": 89.362416008953,
      "totalMs": 3323.721165943425,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 542146560,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3219.0527499769814,
      "warmupMs": 129.12100000539795,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 91.66354197077453,
      "inferenceMs": 105.99420801736414,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.02538783425565608,
      "audioSecondsPerSecond": 39.38894471777217,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 90.99512494867668,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0448574535548687,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02179556885946118,
      "synthesisMs": 90.99649998825043,
      "totalMs": 3325.0469579943456,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 542982144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3219.0527499769814,
      "warmupMs": 129.12100000539795,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 90.60908295214176,
      "inferenceMs": 105.0474580260925,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.02516106779068084,
      "audioSecondsPerSecond": 39.74394124761192,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.95787496678531,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0448574535548687,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021547115683934824,
      "synthesisMs": 89.95920798042789,
      "totalMs": 3324.100208003074,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 542982144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3648.6505000502802,
      "warmupMs": 482.0975000038743,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 90.17629199661314,
      "inferenceMs": 103.71495800791308,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.02484190610967978,
      "audioSecondsPerSecond": 40.2545599997395,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 87.41008403012529,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044866710901260376,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134987776,
      "synthesisRealTimeFactor": 0.020936936290633534,
      "synthesisMs": 87.41170901339501,
      "totalMs": 3752.3654580581933,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644694016,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3648.6505000502802,
      "warmupMs": 482.0975000038743,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 86.16125001572073,
      "inferenceMs": 100.59762501623482,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024095239524846664,
      "audioSecondsPerSecond": 41.50197382220726,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.50079201813787,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044866710901260376,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134987776,
      "synthesisRealTimeFactor": 0.020479570787043093,
      "synthesisMs": 85.50220803590491,
      "totalMs": 3749.248125066515,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644775936,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3648.6505000502802,
      "warmupMs": 482.0975000038743,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 88.690166012384,
      "inferenceMs": 103.7945420248434,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024860968149663088,
      "audioSecondsPerSecond": 40.22369498967207,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.05141597986221,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044866710901260376,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134987776,
      "synthesisRealTimeFactor": 0.021090468985703357,
      "synthesisMs": 88.05270801531151,
      "totalMs": 3752.4450420751236,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644808704,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3648.6505000502802,
      "warmupMs": 482.0975000038743,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 86.84037497732788,
      "inferenceMs": 101.81054100394249,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024385758324297603,
      "audioSecondsPerSecond": 41.00754164382967,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 86.18404198205099,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044866710901260376,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134971392,
      "synthesisRealTimeFactor": 0.020643193539633543,
      "synthesisMs": 86.18533302797005,
      "totalMs": 3750.4610410542227,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644808704,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3722.451000008732,
      "warmupMs": 471.3827919913456,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 91.1915420438163,
      "inferenceMs": 104.55054102931172,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.025042045755523764,
      "audioSecondsPerSecond": 39.93283974331132,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.42900005402043,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044866710901260376,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134987776,
      "synthesisRealTimeFactor": 0.02118103784748805,
      "synthesisMs": 88.4308330132626,
      "totalMs": 3827.001541038044,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642908160,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3722.451000008732,
      "warmupMs": 471.3827919913456,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 84.07133299624547,
      "inferenceMs": 98.57533394824713,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.023610858430717876,
      "audioSecondsPerSecond": 42.353394432241124,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 83.39595794677734,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044866710901260376,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134987776,
      "synthesisRealTimeFactor": 0.01997555879329493,
      "synthesisMs": 83.39795796200633,
      "totalMs": 3821.026333956979,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643022848,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3722.451000008732,
      "warmupMs": 471.3827919913456,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 88.94683298422024,
      "inferenceMs": 103.58804202405736,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024811507071630504,
      "audioSecondsPerSecond": 40.303879853529764,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.29625003272668,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044866710901260376,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134987776,
      "synthesisRealTimeFactor": 0.021149201672747284,
      "synthesisMs": 88.29791698371992,
      "totalMs": 3826.0390420327894,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643022848,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3722.451000008732,
      "warmupMs": 471.3827919913456,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 86.47112495964393,
      "inferenceMs": 101.05820797616616,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024205558796686504,
      "audioSecondsPerSecond": 41.31282439705089,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 85.81887499894947,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044866710901260376,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 134971392,
      "synthesisRealTimeFactor": 0.02055619759652429,
      "synthesisMs": 85.82212496548891,
      "totalMs": 3823.5092079848982,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643088384,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2657.164499978535,
      "warmupMs": 93.0190000217408,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 90.88400000473484,
      "inferenceMs": 103.80625003017485,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.02486377246231733,
      "audioSecondsPerSecond": 40.21915827598428,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.04120797524229,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044870197772979736,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021327834250126593,
      "synthesisMs": 89.04370799427852,
      "totalMs": 2760.9707500087097,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 350552064,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2657.164499978535,
      "warmupMs": 93.0190000217408,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 87.74325001286343,
      "inferenceMs": 101.39779205201194,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.0242868962998831,
      "audioSecondsPerSecond": 41.17446657870455,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 86.97850001044571,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04484185948967934,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020833712581455888,
      "synthesisMs": 86.98075002757832,
      "totalMs": 2758.562292030547,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 351780864,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2657.164499978535,
      "warmupMs": 93.0190000217408,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 88.72779103694484,
      "inferenceMs": 102.62187500484288,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.02458008982151925,
      "audioSecondsPerSecond": 40.683333838940044,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.04250002140179,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04486614465713501,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021088542765240647,
      "synthesisMs": 88.0446660448797,
      "totalMs": 2759.7863749833778,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 351780864,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2657.164499978535,
      "warmupMs": 93.0190000217408,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 88.31795898731798,
      "inferenceMs": 102.18441695906222,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024475309451272388,
      "audioSecondsPerSecond": 40.85750180159677,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 87.50341698760167,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04486287012696266,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020959311371256492,
      "synthesisMs": 87.50512497499585,
      "totalMs": 2759.348916937597,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 354287616,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2618.020833004266,
      "warmupMs": 90.82733397372067,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 89.6772500127554,
      "inferenceMs": 103.03058300632983,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024677983953611937,
      "audioSecondsPerSecond": 40.521948708603375,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.01633300026879,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.044870197772979736,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02108229533350932,
      "synthesisMs": 88.0185830174014,
      "totalMs": 2721.051416010596,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 350142464,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2618.020833004266,
      "warmupMs": 90.82733397372067,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 87.96124998480082,
      "inferenceMs": 101.45408299285918,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024300379159966273,
      "audioSecondsPerSecond": 41.15162127377226,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 87.14741701260209,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04484185948967934,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020874121678089667,
      "synthesisMs": 87.14945800602436,
      "totalMs": 2719.474915997125,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 351551488,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2618.020833004266,
      "warmupMs": 90.82733397372067,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 88.33850000519305,
      "inferenceMs": 102.15833300026134,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.024469061796469783,
      "audioSecondsPerSecond": 40.86793389619347,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 87.6792500494048,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04486614465713501,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02100148695537787,
      "synthesisMs": 87.68120803870261,
      "totalMs": 2720.1791660045274,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 352387072,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Bonjour, ceci est un test de synthèse vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2618.020833004266,
      "warmupMs": 90.82733397372067,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "fr"
      ],
      "sampleRate": 24000,
      "durationMs": 87.45887503027916,
      "inferenceMs": 101.236832968425,
      "audioSeconds": 4.175,
      "realTimeFactor": 0.02424834322597006,
      "audioSecondsPerSecond": 41.23993093800308,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 86.66320802876726,
      "voice": "fr:amelie",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04486287012696266,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02075815378147328,
      "synthesisMs": 86.66529203765094,
      "totalMs": 2719.257665972691,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 354516992,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Pico · Hindi Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64113 ms / 4159 ms0.024 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64116 ms / 3794 ms0.024 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64119 ms / 3794 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64120 ms / 3794 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64121 ms / 3794 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64121 ms / 3782 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64127 ms / 3782 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64130 ms / 3782 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64130 ms / 3782 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642414 ms / 4159 ms0.064 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64790 ms / 4159 ms0.186 RTF
FP32cpuPassedVery fastmacos · aarch64880 ms / 3270 ms0.023 RTF
FP32cpuPassedVery fastmacos · aarch64114 ms / 3104 ms0.024 RTF
FP32cpuPassedVery fastmacos · aarch64115 ms / 3104 ms0.024 RTF
FP32cpuPassedVery fastmacos · aarch64116 ms / 3104 ms0.024 RTF
FP32cpuPassedVery fastmacos · aarch64118 ms / 3104 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64932 ms / 2692 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64121 ms / 2548 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64122 ms / 2548 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64122 ms / 2548 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64123 ms / 3164 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64123 ms / 2692 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64123 ms / 3270 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64123 ms / 3164 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64123 ms / 3164 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64124 ms / 2548 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64124 ms / 3164 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64110 ms / 2692 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64126 ms / 2611 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64126 ms / 2611 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64112 ms / 3270 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64128 ms / 2611 ms0.027 RTF
Q8cpuPassedVery fastmacos · aarch64132 ms / 2611 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch641474 ms / 3117 ms0.039 RTF
FP32cpuPassedVery fastmacos · aarch64191 ms / 3117 ms0.040 RTF
FP32cpuPassedVery fastmacos · aarch64171 ms / 3117 ms0.040 RTF
Q8cpuPassedVery fastmacos · aarch64246 ms / 2651 ms0.051 RTF
Q8cpuPassedVery fastmacos · aarch641950 ms / 2651 ms0.052 RTF
Q8cpuPassedVery fastmacos · aarch64222 ms / 2651 ms0.052 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:06:18.667761+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
da8c9d029e16d9302f1827adb303d92316e25f73
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 174.89983298582956,
      "inferenceMs": 191.22254202375188,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.03963161492720246,
      "audioSecondsPerSecond": 25.23238081104833,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 172.04654100351036,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12454953044652939,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03565772020359345,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 179.48820802848786,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 172.0484999823384,
      "loadMs": 3116.6271249530837,
      "totalMs": 3307.8496669768356,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जांच है.",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 156.84158296789974,
      "inferenceMs": 171.10445897560567,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.04025987270014251,
      "audioSecondsPerSecond": 24.83862796705913,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 156.15933301160112,
      "voice": "hi:arjun",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.191255584359169,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.036744764704695516,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 179.48820802848786,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 156.16524999495596,
      "loadMs": 3116.6271249530837,
      "totalMs": 3287.7315839286894,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उकरन पर आवास बनाने की एक जांच है।  नमस्ते या इस उपकरन पर आवास बनाने की एक जांच है नमस्ते  यह इस उपकरन पर आवास बनाने की एक जांच है  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उकरन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.30303030303030304,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1354.4091659714468,
      "inferenceMs": 1473.8237919518724,
      "audioSeconds": 37.6,
      "realTimeFactor": 0.039197441275315756,
      "audioSecondsPerSecond": 25.511869332903146,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 597.6504579884931,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1394357979297638,
      "clippedFraction": 0.000008865248226950355,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.035882840185993393,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 179.48820802848786,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1349.1947909933515,
      "loadMs": 3116.6271249530837,
      "totalMs": 4590.450916904956,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 106.18779098149389,
      "inferenceMs": 123.04762500571087,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.025502098446779455,
      "audioSecondsPerSecond": 39.212459401602125,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 103.2705829711631,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12454953044652939,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02140357512546404,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 109.16233400348574,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 103.272249980364,
      "loadMs": 3270.332749991212,
      "totalMs": 3393.380374996923,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जांच है.",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 97.3000000230968,
      "inferenceMs": 112.21637501148507,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.02640385294387884,
      "audioSecondsPerSecond": 37.873260471700526,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 96.57875000266358,
      "voice": "hi:arjun",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.191255584359169,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02272485295201049,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 109.16233400348574,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 96.58062504604459,
      "loadMs": 3270.332749991212,
      "totalMs": 3382.549125002697,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उकरन पर आवास बनाने की एक जांच है।  नमस्ते या इस उपकरन पर आवास बनाने की एक जांच है नमस्ते  यह इस उपकरन पर आवास बनाने की एक जांच है  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उकरन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.30303030303030304,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 760.6329590198584,
      "inferenceMs": 879.9001660081558,
      "audioSeconds": 37.6,
      "realTimeFactor": 0.023401600159791376,
      "audioSecondsPerSecond": 42.73212058883904,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 318.8036670326255,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1394357979297638,
      "clippedFraction": 0.000008865248226950355,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020093093989999034,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 109.16233400348574,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 755.5003340239637,
      "loadMs": 3270.332749991212,
      "totalMs": 4150.232915999368,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 97.69312501884997,
      "inferenceMs": 113.48504200577736,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.023520215959746602,
      "audioSecondsPerSecond": 42.51661641676412,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 94.93791597196832,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.124549001455307,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 152518656,
      "synthesisRealTimeFactor": 0.019676614711977965,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1088.4349159896374,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 94.93966598529369,
      "loadMs": 4158.528375031892,
      "totalMs": 4272.01341703767,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जांच है.",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 775.9192499797791,
      "inferenceMs": 790.1233329903334,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.18591137246831374,
      "audioSecondsPerSecond": 5.378907092789267,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 775.2212089835666,
      "voice": "hi:arjun",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1911514550447464,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 164446208,
      "synthesisRealTimeFactor": 0.18240547058282092,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1088.4349159896374,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 775.2232499769889,
      "loadMs": 4158.528375031892,
      "totalMs": 4948.651708022226,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उपकरन पर आवास बनाने की एक जांच है नमस्ते  यह इस उपकरन पर आवास बनाने की एक जांच है  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उकरन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.30303030303030304,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2293.559790996369,
      "inferenceMs": 2413.6902909958735,
      "audioSeconds": 37.6,
      "realTimeFactor": 0.06419389071797536,
      "audioSecondsPerSecond": 15.577806374025922,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1068.1191249750555,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1394372135400772,
      "clippedFraction": 0.00000775709219858156,
      "driverAllocatedMemoryBytes": 1361805312,
      "synthesisRealTimeFactor": 0.06084431736684216,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1088.4349159896374,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2287.7463329932652,
      "loadMs": 4158.528375031892,
      "totalMs": 6572.218666027766,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 229.78041699388996,
      "inferenceMs": 245.97845802782103,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.050979991301102805,
      "audioSecondsPerSecond": 19.615538851187836,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 227.96475002542138,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12444872409105301,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.047247089949856304,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 230.516167008318,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 227.96720900805667,
      "loadMs": 2651.000042038504,
      "totalMs": 2896.978500066325,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जांच है.",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 207.6528340112418,
      "inferenceMs": 221.8154160073027,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.05219186258995358,
      "audioSecondsPerSecond": 19.160074969090875,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 206.98691700818017,
      "voice": "hi:arjun",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1912691295146942,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04870321575607009,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 230.516167008318,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 206.98866696329787,
      "loadMs": 2651.000042038504,
      "totalMs": 2872.8154580458067,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उपकरन पर आवास बनाने की एक जांच है नमस्ते  यह इस उपकरन पर आवास बनाने की एक जांच है  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उकरन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.30303030303030304,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1830.0905000069179,
      "inferenceMs": 1949.7679169871844,
      "audioSeconds": 37.625,
      "realTimeFactor": 0.05182107420563945,
      "audioSecondsPerSecond": 19.297168484615753,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 799.7190419700928,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13929012417793274,
      "clippedFraction": 0.000008859357696567,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0484812137410911,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 230.516167008318,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1824.1056670085527,
      "loadMs": 2651.000042038504,
      "totalMs": 4600.7679590256885,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 106.98800004320219,
      "inferenceMs": 122.9097910108976,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.025473531815730074,
      "audioSecondsPerSecond": 39.25643319637732,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 105.19583302084357,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12446284294128418,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021802728713388734,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 107.46537504019216,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 105.19816604210064,
      "loadMs": 2691.8152500293218,
      "totalMs": 2814.7250410402194,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवाज बनाने की एक जांच है.",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 96.11091698752716,
      "inferenceMs": 109.76516699884087,
      "audioSeconds": 4.25,
      "realTimeFactor": 0.025827098117374322,
      "audioSecondsPerSecond": 38.71902276652921,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 95.44541698414832,
      "voice": "hi:arjun",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19127002358436584,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02245819623417714,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 107.46537504019216,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 95.44733399525285,
      "loadMs": 2691.8152500293218,
      "totalMs": 2801.5804170281626,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उपकरन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उपकरन पर आवास बनाने की एक जांच है नमस्ते  यह इस उपकरन पर आवास बनाने की एक जांच है  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते यह इस उक्करन पर आवास बनाने की एक जांच है।  नमस्ते या इस उकरन पर आवाज बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.30303030303030304,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 813.0472089978866,
      "inferenceMs": 931.8848749971949,
      "audioSeconds": 37.625,
      "realTimeFactor": 0.024767704318862323,
      "audioSecondsPerSecond": 40.37515900246075,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 338.57229200657457,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.13929012417793274,
      "clippedFraction": 0.000008859357696567,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02144530232519258,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 107.46537504019216,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 806.8794999853708,
      "loadMs": 2691.8152500293218,
      "totalMs": 3623.7001250265166,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3164.0711660147645,
      "warmupMs": 113.71750000398606,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 107.93737496715039,
      "inferenceMs": 123.82079201051965,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.02566234031306107,
      "audioSecondsPerSecond": 38.96760731097629,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 104.95937499217689,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12454953044652939,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021753687462214514,
      "synthesisMs": 104.96154200518504,
      "totalMs": 3287.891958025284,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 545488896,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3164.0711660147645,
      "warmupMs": 113.71750000398606,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 106.74141702475026,
      "inferenceMs": 123.35733295185491,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.025566286622146096,
      "audioSecondsPerSecond": 39.114010367613474,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 106.00245802197605,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12454953044652939,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021969784048767132,
      "synthesisMs": 106.00420803530142,
      "totalMs": 3287.4284989666194,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 549502976,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3164.0711660147645,
      "warmupMs": 113.71750000398606,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 105.8358330046758,
      "inferenceMs": 123.07941599283367,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.025508687252400762,
      "audioSecondsPerSecond": 39.20233095906904,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 104.90358300739899,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12454953044652939,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021741951712035607,
      "synthesisMs": 104.90491701057181,
      "totalMs": 3287.150582007598,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 550535168,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3164.0711660147645,
      "warmupMs": 113.71750000398606,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 105.74745800113305,
      "inferenceMs": 122.84579203696921,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.02546026777968274,
      "audioSecondsPerSecond": 39.2768846209072,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 105.01366702374071,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12454953044652939,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02176474093416117,
      "synthesisMs": 105.01487500732765,
      "totalMs": 3286.9169580517337,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 551174144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3104.2933329590596,
      "warmupMs": 104.5946249505505,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 102.29812498437241,
      "inferenceMs": 117.88304196670651,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.024431718542322595,
      "audioSecondsPerSecond": 40.930399483266775,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 99.24420801689848,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12454953044652939,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02056911046052149,
      "synthesisMs": 99.24595797201619,
      "totalMs": 3222.176374925766,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 546078720,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3104.2933329590596,
      "warmupMs": 104.5946249505505,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 97.5826249923557,
      "inferenceMs": 114.27087499760091,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.023683082901057185,
      "audioSecondsPerSecond": 42.224232553582,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 96.82295803213492,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12454953044652939,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02006730570221426,
      "synthesisMs": 96.8247500131838,
      "totalMs": 3218.5642079566605,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 548339712,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3104.2933329590596,
      "warmupMs": 104.5946249505505,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 98.93533401191235,
      "inferenceMs": 115.40737497853115,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.02391862693855568,
      "audioSecondsPerSecond": 41.80841996360786,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 98.15800003707409,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12454953044652939,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02034398963568742,
      "synthesisMs": 98.15974999219179,
      "totalMs": 3219.700707937591,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 549502976,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3104.2933329590596,
      "warmupMs": 104.5946249505505,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 99.31212494848296,
      "inferenceMs": 116.46674998337403,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.024138186525051612,
      "audioSecondsPerSecond": 41.42813292797114,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 98.5455839545466,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12454953044652939,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020424300512791142,
      "synthesisMs": 98.54724997421727,
      "totalMs": 3220.7600829424337,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 549830656,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3781.955292040948,
      "warmupMs": 564.0459999558516,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 110.85399996954948,
      "inferenceMs": 127.03820900060236,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.026329162487171473,
      "audioSecondsPerSecond": 37.98069917671086,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 107.62070800410584,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.124549001455307,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 152535040,
      "synthesisRealTimeFactor": 0.02230513824024503,
      "synthesisMs": 107.62229200918227,
      "totalMs": 3908.99350104155,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642416640,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3781.955292040948,
      "warmupMs": 564.0459999558516,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 104.02537498157471,
      "inferenceMs": 120.71904196636751,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.025019490562977722,
      "audioSecondsPerSecond": 39.968839392746766,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 103.26954198535532,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.124549001455307,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 152535040,
      "synthesisRealTimeFactor": 0.021403419683865933,
      "synthesisMs": 103.27149997465312,
      "totalMs": 3902.6743340073153,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642465792,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3781.955292040948,
      "warmupMs": 564.0459999558516,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 112.56933398544788,
      "inferenceMs": 129.6808339538984,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.026876856778010032,
      "audioSecondsPerSecond": 37.206731734276865,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 111.74733401276171,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.124549001455307,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 152518656,
      "synthesisRealTimeFactor": 0.023160475026333577,
      "synthesisMs": 111.74929200205952,
      "totalMs": 3911.636125994846,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642580480,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3781.955292040948,
      "warmupMs": 564.0459999558516,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 113.23537502903491,
      "inferenceMs": 130.06387499626726,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.026956243522542438,
      "audioSecondsPerSecond": 37.09715707100434,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 112.51091602025554,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.124549001455307,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 152518656,
      "synthesisRealTimeFactor": 0.023318721871784468,
      "synthesisMs": 112.51283303136006,
      "totalMs": 3912.019167037215,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 642596864,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3794.4538749870844,
      "warmupMs": 530.7583750109188,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 103.52604201762006,
      "inferenceMs": 120.51983299897984,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.024978203730358514,
      "audioSecondsPerSecond": 40.034904462909786,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 100.30183399794623,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.124549001455307,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 152535040,
      "synthesisRealTimeFactor": 0.02078836788592252,
      "synthesisMs": 100.30387504957616,
      "totalMs": 3914.9737079860643,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643629056,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3794.4538749870844,
      "warmupMs": 530.7583750109188,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 98.70262502226979,
      "inferenceMs": 115.67487497814,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.02397406735298238,
      "audioSecondsPerSecond": 41.711737323354086,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.9593750089407,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.124549001455307,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 152535040,
      "synthesisRealTimeFactor": 0.020302849739712754,
      "synthesisMs": 97.96124999411404,
      "totalMs": 3910.1287499652244,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643694592,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3794.4538749870844,
      "warmupMs": 530.7583750109188,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 102.30545798549429,
      "inferenceMs": 120.08366599911824,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.024887806424687717,
      "audioSecondsPerSecond": 40.180318945587736,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 101.54520801734179,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.124549001455307,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 152518656,
      "synthesisRealTimeFactor": 0.021046070886264347,
      "synthesisMs": 101.54729202622548,
      "totalMs": 3914.5375409862027,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 645431296,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3794.4538749870844,
      "warmupMs": 530.7583750109188,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 102.2721670451574,
      "inferenceMs": 119.3941660458222,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.024744904879963153,
      "audioSecondsPerSecond": 40.412359831285364,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 101.51062504155561,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.124549001455307,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 152518656,
      "synthesisRealTimeFactor": 0.02103882549822794,
      "synthesisMs": 101.5123330289498,
      "totalMs": 3913.8480410329066,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 645431296,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2548.352708981838,
      "warmupMs": 110.47845799475908,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 106.01995803881437,
      "inferenceMs": 121.8422909732908,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.025252288284619857,
      "audioSecondsPerSecond": 39.60037160707766,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 104.14679103996605,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12446284294128418,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021585319384237644,
      "synthesisMs": 104.14916602894664,
      "totalMs": 2670.194999955129,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 351338496,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2548.352708981838,
      "warmupMs": 110.47845799475908,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 104.98587496113032,
      "inferenceMs": 121.12012499710545,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.025102616579710975,
      "audioSecondsPerSecond": 39.83648464790892,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 104.11029099486768,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12431887537240982,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02157781513195504,
      "synthesisMs": 104.11295801168308,
      "totalMs": 2669.4728339789435,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 353206272,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2548.352708981838,
      "warmupMs": 110.47845799475908,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 105.18999997293577,
      "inferenceMs": 121.79054197622463,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.025241563103880753,
      "audioSecondsPerSecond": 39.61719786863181,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 104.39666698221117,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12451538443565369,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0216371417672909,
      "synthesisMs": 104.39920902717859,
      "totalMs": 2670.1432509580627,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 353484800,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2548.352708981838,
      "warmupMs": 110.47845799475908,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 107.0749590289779,
      "inferenceMs": 123.51087504066527,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.02559810881671819,
      "audioSecondsPerSecond": 39.06538592987375,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 106.29700002027676,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1244666799902916,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022030820517554648,
      "synthesisMs": 106.29870899720117,
      "totalMs": 2671.8635840225033,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 354205696,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2610.5568750062957,
      "warmupMs": 112.16975003480911,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 115.21837499458343,
      "inferenceMs": 131.72191695775837,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.027299879162229715,
      "audioSecondsPerSecond": 36.63019876599063,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 113.17574995337054,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12446284294128418,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023456640828671077,
      "synthesisMs": 113.17829199833795,
      "totalMs": 2742.278791964054,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 349716480,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2610.5568750062957,
      "warmupMs": 112.16975003480911,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 111.37358302948996,
      "inferenceMs": 128.20970901520923,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.026571960417659943,
      "audioSecondsPerSecond": 37.633655337503505,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 110.41854199720547,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12431887537240982,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022885190055179628,
      "synthesisMs": 110.4210420162417,
      "totalMs": 2738.766584021505,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 352010240,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2610.5568750062957,
      "warmupMs": 112.16975003480911,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 108.99379104375839,
      "inferenceMs": 125.71258301613852,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.02605442135049503,
      "audioSecondsPerSecond": 38.38120166046214,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 108.05904102744535,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.12451538443565369,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02239618300812541,
      "synthesisMs": 108.0615830142051,
      "totalMs": 2736.2694580224343,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 355352576,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " नमस्ते यह इस उकरन पर आवास बनाने की एक जांच है।",
      "errorMetric": "wer",
      "errorRate": 0.2727272727272727,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2610.5568750062957,
      "warmupMs": 112.16975003480911,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "hi"
      ],
      "sampleRate": 24000,
      "durationMs": 109.921416034922,
      "inferenceMs": 126.296583970543,
      "audioSeconds": 4.825,
      "realTimeFactor": 0.02617545781772912,
      "audioSecondsPerSecond": 38.20372529731578,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 109.0668750111945,
      "voice": "hi:priya",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.1244666799902916,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022604991295833352,
      "synthesisMs": 109.06908300239593,
      "totalMs": 2736.8534589768387,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 357253120,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Pico · Italian Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64107 ms / 3886 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64111 ms / 3706 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64112 ms / 3886 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64113 ms / 3886 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64113 ms / 3808 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64114 ms / 3706 ms0.028 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64117 ms / 3706 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64119 ms / 3706 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64141 ms / 3886 ms0.035 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64269 ms / 3808 ms0.067 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642375 ms / 3808 ms0.087 RTF
FP32cpuPassedVery fastmacos · aarch64674 ms / 3228 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64684 ms / 2611 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64102 ms / 2471 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64103 ms / 3196 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64104 ms / 3196 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64104 ms / 2471 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64104 ms / 2578 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64104 ms / 2471 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64104 ms / 2578 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64104 ms / 3196 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64105 ms / 2925 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64105 ms / 2925 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64105 ms / 2471 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64105 ms / 2578 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64105 ms / 2925 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64104 ms / 2611 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64105 ms / 2578 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64106 ms / 3196 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64107 ms / 2925 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64109 ms / 2611 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch64109 ms / 3228 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch64115 ms / 3228 ms0.028 RTF
FP32cpuPassedVery fastmacos · aarch641089 ms / 3187 ms0.040 RTF
FP32cpuPassedVery fastmacos · aarch64165 ms / 3187 ms0.041 RTF
FP32cpuPassedVery fastmacos · aarch64168 ms / 3187 ms0.041 RTF
Q8cpuPassedVery fastmacos · aarch64214 ms / 2698 ms0.054 RTF
Q8cpuPassedVery fastmacos · aarch641464 ms / 2698 ms0.054 RTF
Q8cpuPassedVery fastmacos · aarch64219 ms / 2698 ms0.054 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:05:11.427939+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
da8c9d029e16d9302f1827adb303d92316e25f73
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 153.7559579592198,
      "inferenceMs": 168.0738750146702,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.04149972222584449,
      "audioSecondsPerSecond": 24.09654682886617,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 151.0012499638833,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05977248400449753,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.037284660480778525,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 157.81404200242832,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 151.00287494715303,
      "loadMs": 3186.854833038524,
      "totalMs": 3354.928708053194,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 150.9921660181135,
      "inferenceMs": 165.006916038692,
      "audioSeconds": 4,
      "realTimeFactor": 0.041251729009673,
      "audioSecondsPerSecond": 24.24140815444397,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 150.33162501640618,
      "voice": "it:alessandro",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19446180760860443,
      "clippedFraction": 0.0017395833333333334,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.037583364508464,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 157.81404200242832,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 150.333458033856,
      "loadMs": 3186.854833038524,
      "totalMs": 3351.861749077216,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, Buongiorno, questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, buongiorno. questo è un test di sintesi vocale locale. Buongiorno, questo è un test di sintesi vocale locale, Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1002.294041973073,
      "inferenceMs": 1089.2652079928666,
      "audioSeconds": 27.15,
      "realTimeFactor": 0.040120265487766725,
      "audioSecondsPerSecond": 24.92505938937306,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 545.7983749802224,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06369679421186447,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0367716482487159,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 157.81404200242832,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 998.3502499526367,
      "loadMs": 3186.854833038524,
      "totalMs": 4276.12004103139,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 100.07445799419656,
      "inferenceMs": 114.50762499589473,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.028273487653307342,
      "audioSecondsPerSecond": 35.36882369314007,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.26425004191697,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05977248400449753,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.0240163990141203,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 103.81320799933746,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 97.26641600718722,
      "loadMs": 3227.6819170219824,
      "totalMs": 3342.189542017877,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 94.74837500602007,
      "inferenceMs": 108.53504203259945,
      "audioSeconds": 4,
      "realTimeFactor": 0.027133760508149862,
      "audioSecondsPerSecond": 36.85445663529172,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 94.06945802038535,
      "voice": "it:alessandro",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19446180760860443,
      "clippedFraction": 0.0017395833333333334,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02351780200842768,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 103.81320799933746,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 94.07120803371072,
      "loadMs": 3227.6819170219824,
      "totalMs": 3336.216959054582,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, Buongiorno, questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, buongiorno. questo è un test di sintesi vocale locale. Buongiorno, questo è un test di sintesi vocale locale, Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 586.8099579820409,
      "inferenceMs": 673.8956660265103,
      "audioSeconds": 27.15,
      "realTimeFactor": 0.024821203168563916,
      "audioSecondsPerSecond": 40.28813563987507,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 298.3342499937862,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06369679421186447,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021471233884538105,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 103.81320799933746,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 582.9439999652095,
      "loadMs": 3227.6819170219824,
      "totalMs": 3901.5775830484927,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 99.5168749941513,
      "inferenceMs": 113.43716701958328,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.028009177041872415,
      "audioSecondsPerSecond": 35.70258413894298,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 96.52904199901968,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06015165522694588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 143589376,
      "synthesisRealTimeFactor": 0.023834794317084698,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 488.31683304160833,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 96.53091698419303,
      "loadMs": 3807.9379170085303,
      "totalMs": 3921.3750840281136,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 254.5938339899294,
      "inferenceMs": 268.51216703653336,
      "audioSeconds": 4,
      "realTimeFactor": 0.06712804175913334,
      "audioSecondsPerSecond": 14.896904092453159,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 253.78370902035385,
      "voice": "it:alessandro",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19445709884166718,
      "clippedFraction": 0.0017916666666666667,
      "driverAllocatedMemoryBytes": 147013632,
      "synthesisRealTimeFactor": 0.06344634374545421,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 488.31683304160833,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 253.78537498181686,
      "loadMs": 3807.9379170085303,
      "totalMs": 4076.4500840450637,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, Buongiorno, questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, buongiorno. questo è un test di sintesi vocale locale. Buongiorno, questo è un test di sintesi vocale locale, Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2289.1766669927165,
      "inferenceMs": 2375.013582990505,
      "audioSeconds": 27.15,
      "realTimeFactor": 0.08747748003648269,
      "audioSecondsPerSecond": 11.431513568783048,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1154.2505839606747,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06351612508296967,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1295532032,
      "synthesisRealTimeFactor": 0.0841400015462039,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 488.31683304160833,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2284.401041979436,
      "loadMs": 3807.9379170085303,
      "totalMs": 6182.951499999035,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 204.6238339971751,
      "inferenceMs": 218.5018330346793,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.053951069885106,
      "audioSecondsPerSecond": 18.535313611566856,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 202.8858750127256,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05975615605711937,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.050095843703106595,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 209.4834999879822,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 202.88816699758172,
      "loadMs": 2697.6844169548713,
      "totalMs": 2916.1862499895506,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 200.56008297251537,
      "inferenceMs": 214.0855420148,
      "audioSeconds": 4,
      "realTimeFactor": 0.0535213855037,
      "audioSecondsPerSecond": 18.684120199595146,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 199.8383329482749,
      "voice": "it:alessandro",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19400152564048767,
      "clippedFraction": 0.0016979166666666666,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04996006248984486,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 209.4834999879822,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 199.84024995937943,
      "loadMs": 2697.6844169548713,
      "totalMs": 2911.7699589696713,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, Buongiorno, questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, buongiorno. questo è un test di sintesi vocale locale. Buongiorno, questo è un test di sintesi vocale locale, Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1376.9445000216365,
      "inferenceMs": 1463.582917000167,
      "audioSeconds": 27.15,
      "realTimeFactor": 0.05390728976059547,
      "audioSecondsPerSecond": 18.550366832408784,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 765.106333012227,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0636860728263855,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.050549706886651107,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 209.4834999879822,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1372.4245419725776,
      "loadMs": 2697.6844169548713,
      "totalMs": 4161.267333955038,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 94.57220800686628,
      "inferenceMs": 109.27366599207744,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.026981152096809246,
      "audioSecondsPerSecond": 37.062909560420835,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 92.81820798059925,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.059770356863737106,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022918590614687144,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 95.00841703265905,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 92.82029198948294,
      "loadMs": 2610.8929170295596,
      "totalMs": 2720.166583021637,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 90.69404099136591,
      "inferenceMs": 103.84704201715067,
      "audioSeconds": 4,
      "realTimeFactor": 0.025961760504287668,
      "audioSecondsPerSecond": 38.518189081778445,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.96504097012803,
      "voice": "it:alessandro",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.19400592148303986,
      "clippedFraction": 0.0016979166666666666,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02249178124475293,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 95.00841703265905,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 89.96712497901171,
      "loadMs": 2610.8929170295596,
      "totalMs": 2714.7399590467103,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, buongiorno, questo è un test di sintesi vocale locale, Buongiorno, questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, buongiorno. Questo è un test di sintesi vocale locale, buongiorno. questo è un test di sintesi vocale locale. Buongiorno, questo è un test di sintesi vocale locale, Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 597.7410409832373,
      "inferenceMs": 683.5705830017105,
      "audioSeconds": 27.15,
      "realTimeFactor": 0.025177553701720458,
      "audioSecondsPerSecond": 39.71791746914899,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 316.9089999864809,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0636860728263855,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021835397457288883,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 95.00841703265905,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 592.8310409653932,
      "loadMs": 2610.8929170295596,
      "totalMs": 3294.46350003127,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3195.6398339825682,
      "warmupMs": 94.29670899407938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 92.44354104157537,
      "inferenceMs": 106.10504198120907,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.0261987757978294,
      "audioSecondsPerSecond": 38.169722422024435,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.74950003903359,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05977248400449753,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02216070988158017,
      "synthesisMs": 89.75087502039969,
      "totalMs": 3301.7448759637773,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 526696448,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3195.6398339825682,
      "warmupMs": 94.29670899407938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 90.28829098679125,
      "inferenceMs": 104.46200001752004,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.02579308642407902,
      "audioSecondsPerSecond": 38.77007906531319,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.41420796327293,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05977248400449753,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02207799382036391,
      "synthesisMs": 89.41587497247383,
      "totalMs": 3300.1018340000883,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 530579456,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3195.6398339825682,
      "warmupMs": 94.29670899407938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 89.4464579760097,
      "inferenceMs": 103.60858397325501,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.025582366413149385,
      "audioSecondsPerSecond": 39.08942526466192,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.81083299638703,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05977248400449753,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021928981475440073,
      "synthesisMs": 88.8123749755323,
      "totalMs": 3299.2484179558232,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 530612224,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3195.6398339825682,
      "warmupMs": 94.29670899407938,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 89.06437503173947,
      "inferenceMs": 102.75550000369549,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.025371728395974194,
      "audioSecondsPerSecond": 39.41394864366721,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.42133299913257,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05977248400449753,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021832767416964525,
      "synthesisMs": 88.42270803870633,
      "totalMs": 3298.3953339862637,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 531185664,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2925.094500009436,
      "warmupMs": 94.67891597887501,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 92.96708297915757,
      "inferenceMs": 106.75562499091029,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.02635941357800254,
      "audioSecondsPerSecond": 37.937111045388356,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 90.19162500044331,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05977248400449753,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02227026740675447,
      "synthesisMs": 90.19458299735561,
      "totalMs": 3031.8501250003465,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 530399232,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2925.094500009436,
      "warmupMs": 94.67891597887501,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 90.66508302930743,
      "inferenceMs": 104.74779194919392,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.025863652333134302,
      "audioSecondsPerSecond": 38.6642995010757,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.97404103865847,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05977248400449753,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022216193337729315,
      "synthesisMs": 89.97558301780373,
      "totalMs": 3029.84229195863,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 530726912,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2925.094500009436,
      "warmupMs": 94.67891597887501,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 90.63220798270777,
      "inferenceMs": 104.66441698372364,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.025843065921907073,
      "audioSecondsPerSecond": 38.695099220108546,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.97162495506927,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05977248400449753,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022215483448594624,
      "synthesisMs": 89.97270796680823,
      "totalMs": 3029.75891699316,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 530972672,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 2925.094500009436,
      "warmupMs": 94.67891597887501,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 91.08787501463667,
      "inferenceMs": 104.97179202502593,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.025918960993833565,
      "audioSecondsPerSecond": 38.581793469186984,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 90.42729198699817,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.05977248400449753,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022328065932486898,
      "synthesisMs": 90.42866702657193,
      "totalMs": 3030.066292034462,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 531005440,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3706.2012079986744,
      "warmupMs": 514.0571250231005,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 100.00837501138449,
      "inferenceMs": 113.55345899937674,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.02803789111095722,
      "audioSecondsPerSecond": 35.666020530666785,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.14687499217689,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06015165522694588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 143589376,
      "synthesisRealTimeFactor": 0.023987448636304448,
      "synthesisMs": 97.14916697703302,
      "totalMs": 3819.754666998051,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643596288,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3706.2012079986744,
      "warmupMs": 514.0571250231005,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 102.76529198745266,
      "inferenceMs": 116.51308299042284,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.028768662466771074,
      "audioSecondsPerSecond": 34.760044932747185,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 102.09175001364201,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06015165522694588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 143589376,
      "synthesisRealTimeFactor": 0.0252083024688433,
      "synthesisMs": 102.09362499881536,
      "totalMs": 3822.714290989097,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643629056,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3706.2012079986744,
      "warmupMs": 514.0571250231005,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 104.64404197409749,
      "inferenceMs": 118.94041701452807,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.029368004201118042,
      "audioSecondsPerSecond": 34.0506625221039,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 103.9621249656193,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06015165522694588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 143589376,
      "synthesisRealTimeFactor": 0.025670051353374197,
      "synthesisMs": 103.9637079811655,
      "totalMs": 3825.1416250132024,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643710976,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3706.2012079986744,
      "warmupMs": 514.0571250231005,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 98.00458396784961,
      "inferenceMs": 111.44054098986089,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.02751618296045948,
      "audioSecondsPerSecond": 36.342249992922035,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.37425000639632,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06015165522694588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 143572992,
      "synthesisRealTimeFactor": 0.02404347753416701,
      "synthesisMs": 97.37608401337638,
      "totalMs": 3817.6417489885353,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643727360,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3886.0092500108294,
      "warmupMs": 547.8259159717709,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 98.5696249990724,
      "inferenceMs": 112.01370798517019,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.027657705675350662,
      "audioSecondsPerSecond": 36.15628901898499,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 95.70462501142174,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06015165522694588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 143589376,
      "synthesisRealTimeFactor": 0.023631347648017567,
      "synthesisMs": 95.70695797447115,
      "totalMs": 3998.0229579959996,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644087808,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3886.0092500108294,
      "warmupMs": 547.8259159717709,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 92.16729202307761,
      "inferenceMs": 107.00879199430346,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.02642192394921073,
      "audioSecondsPerSecond": 37.847357441579184,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 91.51012502843514,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06015165522694588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 143589376,
      "synthesisRealTimeFactor": 0.022595504192451453,
      "synthesisMs": 91.51179197942838,
      "totalMs": 3993.018042005133,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644169728,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3886.0092500108294,
      "warmupMs": 547.8259159717709,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 98.56516699073836,
      "inferenceMs": 113.16429200815037,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.027941800495839598,
      "audioSecondsPerSecond": 35.788674396587126,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.91091701481491,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06015165522694588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 143572992,
      "synthesisRealTimeFactor": 0.024176039014379553,
      "synthesisMs": 97.91295800823718,
      "totalMs": 3999.17354201898,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644349952,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3886.0092500108294,
      "warmupMs": 547.8259159717709,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 126.25191599363461,
      "inferenceMs": 141.2684170063585,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.03488109061885395,
      "audioSecondsPerSecond": 28.668828361102886,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 125.58941601309925,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06015165522694588,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 143572992,
      "synthesisRealTimeFactor": 0.03101008221288614,
      "synthesisMs": 125.59083296218887,
      "totalMs": 4027.277667017188,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 644415488,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2577.6851670234464,
      "warmupMs": 96.82958398479968,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 92.83483301987872,
      "inferenceMs": 105.39741697721183,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.026024053574620204,
      "audioSecondsPerSecond": 38.42598913857308,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 91.07362502254546,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.059770356863737106,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022487911355486254,
      "synthesisMs": 91.07604098971933,
      "totalMs": 2683.0825840006582,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 349192192,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2577.6851670234464,
      "warmupMs": 96.82958398479968,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 91.24237502692267,
      "inferenceMs": 104.95308297686279,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.025914341475768588,
      "audioSecondsPerSecond": 38.588671100713015,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 90.46279202448204,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06017688661813736,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022336964935271278,
      "synthesisMs": 90.46470798784867,
      "totalMs": 2682.638250000309,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 351617024,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2577.6851670234464,
      "warmupMs": 96.82958398479968,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 90.85899998899549,
      "inferenceMs": 104.07854098593816,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.025698405181713128,
      "audioSecondsPerSecond": 38.91292058511069,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 90.17112501896918,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.060352545231580734,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022264845672144012,
      "synthesisMs": 90.17262497218326,
      "totalMs": 2681.7637080093846,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 352550912,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2577.6851670234464,
      "warmupMs": 96.82958398479968,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 90.62041703145951,
      "inferenceMs": 103.7222080049105,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.025610421729607532,
      "audioSecondsPerSecond": 39.046604173797206,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.95345904259011,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.060215551406145096,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02221125531596718,
      "synthesisMs": 89.95558402966708,
      "totalMs": 2681.407375028357,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 352731136,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2470.605583046563,
      "warmupMs": 94.28374998969957,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 91.72437497181818,
      "inferenceMs": 103.96954196039587,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.025671491842073055,
      "audioSecondsPerSecond": 38.95371590213149,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 90.10545897763222,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.059770356863737106,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022248724435260634,
      "synthesisMs": 90.10733396280557,
      "totalMs": 2574.575125006959,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 349503488,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2470.605583046563,
      "warmupMs": 94.28374998969957,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 90.4698750237003,
      "inferenceMs": 103.62933302531019,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.02558748963587906,
      "audioSecondsPerSecond": 39.08159863395856,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.71666701836511,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.06017688661813736,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022152736553071456,
      "synthesisMs": 89.7185830399394,
      "totalMs": 2574.234916071873,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 352141312,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2470.605583046563,
      "warmupMs": 94.28374998969957,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 89.10333400126547,
      "inferenceMs": 102.19045804115012,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.025232211862012376,
      "audioSecondsPerSecond": 39.63188029130022,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.45916698919609,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.060352545231580734,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021842283952728282,
      "synthesisMs": 88.46125000854954,
      "totalMs": 2572.796041087713,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 352141312,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Buongiorno, questo è un test di sintesi vocale locale.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2470.605583046563,
      "warmupMs": 94.28374998969957,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "it"
      ],
      "sampleRate": 24000,
      "durationMs": 91.74808399984613,
      "inferenceMs": 104.88025000086054,
      "audioSeconds": 4.05,
      "realTimeFactor": 0.025896358024903837,
      "audioSecondsPerSecond": 38.615468593627206,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 91.10770898405463,
      "voice": "it:francesca",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.060215551406145096,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02249611110460979,
      "synthesisMs": 91.10924997366965,
      "totalMs": 2575.4858330474235,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 352141312,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Pico · Portuguese Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64102 ms / 3871 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64104 ms / 3871 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64106 ms / 4391 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64107 ms / 3871 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64107 ms / 3871 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64107 ms / 4122 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64111 ms / 4122 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64112 ms / 4122 ms0.027 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64122 ms / 4122 ms0.030 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64267 ms / 4391 ms0.069 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642346 ms / 4391 ms0.076 RTF
FP32cpuPassedVery fastmacos · aarch64733 ms / 3144 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64757 ms / 2570 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64106 ms / 2630 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64107 ms / 2688 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64107 ms / 2630 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64107 ms / 2688 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64100 ms / 2570 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64107 ms / 3355 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64108 ms / 2630 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64108 ms / 2688 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64108 ms / 2630 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64108 ms / 2688 ms0.026 RTF
Q8cpuPassedVery fastmacos · aarch64108 ms / 2570 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64109 ms / 3355 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64110 ms / 3157 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch64110 ms / 3355 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch64111 ms / 3157 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch64111 ms / 3157 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch64112 ms / 3157 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch64104 ms / 3144 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch64112 ms / 3355 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch64115 ms / 3144 ms0.028 RTF
FP32cpuPassedVery fastmacos · aarch641198 ms / 3278 ms0.039 RTF
FP32cpuPassedVery fastmacos · aarch64166 ms / 3278 ms0.040 RTF
FP32cpuPassedVery fastmacos · aarch64157 ms / 3278 ms0.041 RTF
Q8cpuPassedVery fastmacos · aarch641585 ms / 2801 ms0.052 RTF
Q8cpuPassedVery fastmacos · aarch64217 ms / 2801 ms0.053 RTF
Q8cpuPassedVery fastmacos · aarch64206 ms / 2801 ms0.053 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:05:34.101930+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
da8c9d029e16d9302f1827adb303d92316e25f73
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 152.0817499840632,
      "inferenceMs": 165.6251250533387,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.04015154546747605,
      "audioSecondsPerSecond": 24.905641572627133,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 149.3200000259094,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07429730147123337,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03619915152420149,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 157.9619160038419,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 149.32150003733113,
      "loadMs": 3278.0217499821447,
      "totalMs": 3443.6468750354834,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 144.0041660098359,
      "inferenceMs": 156.90629096934572,
      "audioSeconds": 3.85,
      "realTimeFactor": 0.04075488077125863,
      "audioSecondsPerSecond": 24.536938424936462,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 143.39624997228384,
      "voice": "pt:larissa",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.049440693110227585,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03724613636671626,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 157.9619160038419,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 143.3976250118576,
      "loadMs": 3278.0217499821447,
      "totalMs": 3434.9280409514904,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local, Olá. Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1099.3648750009015,
      "inferenceMs": 1197.9147500242107,
      "audioSeconds": 30.75,
      "realTimeFactor": 0.038956577236559695,
      "audioSecondsPerSecond": 25.669606288242566,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 682.3901670286432,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07832801342010498,
      "clippedFraction": 0.0000027100271002710026,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.035606331968059145,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 157.9619160038419,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1094.8947080178186,
      "loadMs": 3278.0217499821447,
      "totalMs": 4475.936500006355,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 100.49637500196695,
      "inferenceMs": 115.2692919713445,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.027944070780932,
      "audioSecondsPerSecond": 35.785766785359094,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.77912503341213,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07429730147123337,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023704505215088528,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 103.62999996868894,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 97.78108401224017,
      "loadMs": 3144.402874982916,
      "totalMs": 3259.6721669542603,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 91.67783299926668,
      "inferenceMs": 104.4954169774428,
      "audioSeconds": 3.85,
      "realTimeFactor": 0.02714166674738774,
      "audioSecondsPerSecond": 36.84372110626719,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 91.04166697943583,
      "voice": "pt:larissa",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.049440693110227585,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023647662336853417,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 103.62999996868894,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 91.04349999688566,
      "loadMs": 3144.402874982916,
      "totalMs": 3248.8982919603586,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local, Olá. Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 635.8826250070706,
      "inferenceMs": 732.6458750176243,
      "audioSeconds": 30.75,
      "realTimeFactor": 0.02382588211439429,
      "audioSecondsPerSecond": 41.971163762111246,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 373.8005410414189,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07832801342010498,
      "clippedFraction": 0.0000027100271002710026,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020537174797079307,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 103.62999996868894,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 631.5181250101887,
      "loadMs": 3144.402874982916,
      "totalMs": 3877.04875000054,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 91.04441601084545,
      "inferenceMs": 105.82904098555446,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.025655525087407142,
      "audioSecondsPerSecond": 38.97795880587312,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.25966599397361,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07432806491851807,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135053312,
      "synthesisRealTimeFactor": 0.02139669696264195,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1164.346875040792,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 88.26137497089803,
      "loadMs": 4391.424292000011,
      "totalMs": 4497.253332985565,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 253.90483305091038,
      "inferenceMs": 266.6694999788888,
      "audioSeconds": 3.85,
      "realTimeFactor": 0.06926480518932177,
      "audioSecondsPerSecond": 14.437346604335291,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 253.13345802715048,
      "voice": "pt:larissa",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04943174496293068,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 138428416,
      "synthesisRealTimeFactor": 0.06574954545766405,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1164.346875040792,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 253.1357500120066,
      "loadMs": 4391.424292000011,
      "totalMs": 4658.093791978899,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local, Olá. Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2248.4842920093797,
      "inferenceMs": 2346.268291992601,
      "audioSeconds": 30.75,
      "realTimeFactor": 0.07630140786967808,
      "audioSecondsPerSecond": 13.10591806782895,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1163.6861250153743,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07832906395196915,
      "clippedFraction": 0.000005420054200542005,
      "driverAllocatedMemoryBytes": 1309360128,
      "synthesisRealTimeFactor": 0.0729626341461681,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 1164.346875040792,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2243.600999994669,
      "loadMs": 4391.424292000011,
      "totalMs": 6737.692583992612,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 203.66445800755173,
      "inferenceMs": 217.0456659514457,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.052617131139744415,
      "audioSecondsPerSecond": 19.005217090687196,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 201.85254100942984,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07417352497577667,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04893438375526757,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 208.3386669983156,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 201.85433299047872,
      "loadMs": 2800.5488330381922,
      "totalMs": 3017.594498989638,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 192.96754099195823,
      "inferenceMs": 205.81670897081494,
      "audioSeconds": 3.85,
      "realTimeFactor": 0.05345888544696492,
      "audioSecondsPerSecond": 18.705964249704987,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 192.33549997443333,
      "voice": "pt:larissa",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04942925646901131,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04995780284879366,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 208.3386669983156,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 192.3375409678556,
      "loadMs": 2800.5488330381922,
      "totalMs": 3006.365542009007,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local, Olá. Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1484.8781659966335,
      "inferenceMs": 1584.9240830284543,
      "audioSeconds": 30.75,
      "realTimeFactor": 0.05154224660255136,
      "audioSecondsPerSecond": 19.4015601941282,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 939.6210830309428,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07828433066606522,
      "clippedFraction": 0.000008130081300813007,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04813664497588042,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 208.3386669983156,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1480.2018330083229,
      "loadMs": 2800.5488330381922,
      "totalMs": 4385.472916066647,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 93.8846250064671,
      "inferenceMs": 108.32900000968948,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.026261575759924723,
      "audioSecondsPerSecond": 38.078446211365744,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 92.19341701827943,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07417299598455429,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022350484858069456,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 94.74816697183996,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 92.1957500395365,
      "loadMs": 2570.1016660314053,
      "totalMs": 2678.430666041095,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 87.7106249681674,
      "inferenceMs": 100.06804100703448,
      "audioSeconds": 3.85,
      "realTimeFactor": 0.0259916989628661,
      "audioSecondsPerSecond": 38.473822024050186,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 87.00333395972848,
      "voice": "pt:larissa",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.049428410828113556,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022598636357791045,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 94.74816697183996,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 87.00474997749552,
      "loadMs": 2570.1016660314053,
      "totalMs": 2670.16970703844,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": "Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local, Olá. Este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local. Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 660.4655829723924,
      "inferenceMs": 756.6902920370921,
      "audioSeconds": 30.75,
      "realTimeFactor": 0.02460781437518999,
      "audioSecondsPerSecond": 40.63749769700054,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 396.37454197509214,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07828494161367416,
      "clippedFraction": 0.000008130081300813007,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021315635511333624,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 94.74816697183996,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 655.4557919735089,
      "loadMs": 2570.1016660314053,
      "totalMs": 3326.7919580684975,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3355.32020800747,
      "warmupMs": 102.437415975146,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 97.11341699585319,
      "inferenceMs": 110.23454199312255,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.02672352533166607,
      "audioSecondsPerSecond": 37.420212624980614,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 94.44612497463822,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07429730147123337,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022896404124119064,
      "synthesisMs": 94.44766701199114,
      "totalMs": 3465.5547500005923,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 533757952,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3355.32020800747,
      "warmupMs": 102.437415975146,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 94.58620799705386,
      "inferenceMs": 108.75966702587903,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.02636597988506158,
      "audioSecondsPerSecond": 37.9276630096566,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 93.92116602975875,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07429730147123337,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02276918181980198,
      "synthesisMs": 93.92287500668317,
      "totalMs": 3464.079875033349,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 533938176,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3355.32020800747,
      "warmupMs": 102.437415975146,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 97.83724998123944,
      "inferenceMs": 111.99587502051145,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.027150515156487625,
      "audioSecondsPerSecond": 36.83171366128019,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.1508749644272,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07429730147123337,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02355218182007472,
      "synthesisMs": 97.15275000780821,
      "totalMs": 3467.316083027981,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 534528000,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3355.32020800747,
      "warmupMs": 102.437415975146,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 92.59416698478162,
      "inferenceMs": 107.45566699188203,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.026049858664698673,
      "audioSecondsPerSecond": 38.38792420609731,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 91.69537498382851,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07429730147123337,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02222955562738758,
      "synthesisMs": 91.69691696297377,
      "totalMs": 3462.775874999352,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 537526272,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3157.191291975323,
      "warmupMs": 100.50258401315659,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 97.84804098308086,
      "inferenceMs": 111.82458396069705,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.027108990051078075,
      "audioSecondsPerSecond": 36.888131874917704,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 95.04658298101276,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07429730147123337,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023042040242051535,
      "synthesisMs": 95.04841599846259,
      "totalMs": 3269.01587593602,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 532103168,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3157.191291975323,
      "warmupMs": 100.50258401315659,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 95.3840829897672,
      "inferenceMs": 109.93608296848834,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.026651171628724445,
      "audioSecondsPerSecond": 37.52180256579066,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 94.70362501451746,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07429730147123337,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022958818173995525,
      "synthesisMs": 94.70512496773154,
      "totalMs": 3267.1273749438114,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 532529152,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3157.191291975323,
      "warmupMs": 100.50258401315659,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 96.04750003200024,
      "inferenceMs": 110.64875003648922,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.026823939402785265,
      "audioSecondsPerSecond": 37.280131936779014,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 95.38220800459385,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07429730147123337,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02312332315539772,
      "synthesisMs": 95.38370801601559,
      "totalMs": 3267.8400420118123,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 532611072,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3157.191291975323,
      "warmupMs": 100.50258401315659,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 96.93850000621751,
      "inferenceMs": 110.73816596763209,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.026845615992153234,
      "audioSecondsPerSecond": 37.25002995991198,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 96.240459010005,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07429730147123337,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023331393939301823,
      "synthesisMs": 96.24199999962002,
      "totalMs": 3267.929457942955,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 533741568,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3871.235999977216,
      "warmupMs": 536.784500000067,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 92.79591700760648,
      "inferenceMs": 106.60370800178498,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.025843323151947872,
      "audioSecondsPerSecond": 38.69471407064875,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 89.96520802611485,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07432806491851807,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135069696,
      "synthesisRealTimeFactor": 0.02181018181991848,
      "synthesisMs": 89.96700000716373,
      "totalMs": 3977.839707979001,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643612672,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3871.235999977216,
      "warmupMs": 536.784500000067,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 89.51729198452085,
      "inferenceMs": 104.18191697681323,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.025256222297409266,
      "audioSecondsPerSecond": 39.59420329075018,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 88.79216702189296,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07432806491851807,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135069696,
      "synthesisRealTimeFactor": 0.02152595975238717,
      "synthesisMs": 88.79458397859707,
      "totalMs": 3975.4179169540294,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643678208,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3871.235999977216,
      "warmupMs": 536.784500000067,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 92.52533299149945,
      "inferenceMs": 106.79666697978973,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.025890101086009632,
      "audioSecondsPerSecond": 38.62480091050611,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 91.86670801136643,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07432806491851807,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135053312,
      "synthesisRealTimeFactor": 0.022271121202996284,
      "synthesisMs": 91.86837496235967,
      "totalMs": 3978.032666957006,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643776512,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3871.235999977216,
      "warmupMs": 536.784500000067,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 87.81129203271121,
      "inferenceMs": 102.0805409643799,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.024746797809546642,
      "audioSecondsPerSecond": 40.40926861309818,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 87.11395802674815,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07432806491851807,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 143441920,
      "synthesisRealTimeFactor": 0.021118949576647894,
      "synthesisMs": 87.11566700367257,
      "totalMs": 3973.316540941596,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643792896,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 4122.393666999415,
      "warmupMs": 590.8265000325628,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 96.65633400436491,
      "inferenceMs": 110.56458298116922,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.026803535268162235,
      "audioSecondsPerSecond": 37.30851135849306,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 93.79274997627363,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07432806491851807,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135069696,
      "synthesisRealTimeFactor": 0.02273809089974472,
      "synthesisMs": 93.79462496144697,
      "totalMs": 4232.958249980584,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643334144,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 4122.393666999415,
      "warmupMs": 590.8265000325628,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 92.5304580014199,
      "inferenceMs": 107.44316701311618,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.026046828366816044,
      "audioSecondsPerSecond": 38.392390271746535,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 91.81658300803974,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07432806491851807,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135069696,
      "synthesisRealTimeFactor": 0.02225912122450995,
      "synthesisMs": 91.81887505110353,
      "totalMs": 4229.836834012531,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643366912,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 4122.393666999415,
      "warmupMs": 590.8265000325628,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 97.88283403031528,
      "inferenceMs": 112.33012500451878,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.027231545455640917,
      "audioSecondsPerSecond": 36.7221170619552,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.20720903715119,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07432806491851807,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 135053312,
      "synthesisRealTimeFactor": 0.023565888969284115,
      "synthesisMs": 97.20929199829698,
      "totalMs": 4234.723792003933,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643416064,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 4122.393666999415,
      "warmupMs": 590.8265000325628,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 105.49450002145022,
      "inferenceMs": 121.69175001326948,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.029501030306247147,
      "audioSecondsPerSecond": 33.89712120624613,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 104.82487501576543,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07432806491851807,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 143441920,
      "synthesisRealTimeFactor": 0.025412454552045376,
      "synthesisMs": 104.82637502718717,
      "totalMs": 4244.085417012684,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 643448832,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2630.391957995016,
      "warmupMs": 97.08562499145046,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 94.4196250056848,
      "inferenceMs": 107.92995797237381,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.026164838296333044,
      "audioSecondsPerSecond": 38.21923104107806,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 92.79666701331735,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07417299598455429,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022496747643207058,
      "synthesisMs": 92.79908402822912,
      "totalMs": 2738.32191596739,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 349421568,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2630.391957995016,
      "warmupMs": 97.08562499145046,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 93.35654199821874,
      "inferenceMs": 107.52741602482274,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.026067252369653997,
      "audioSecondsPerSecond": 38.36230937650117,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 92.63037500204518,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07415793836116791,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022456444602346783,
      "synthesisMs": 92.63283398468047,
      "totalMs": 2737.919374019839,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 351141888,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2630.391957995016,
      "warmupMs": 97.08562499145046,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 93.15170801710337,
      "inferenceMs": 106.89370799809694,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.025913626181356834,
      "audioSecondsPerSecond": 38.58973626467742,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 92.45816699694842,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0741841122508049,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02241461623855161,
      "synthesisMs": 92.46029198402539,
      "totalMs": 2737.285665993113,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 352354304,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2630.391957995016,
      "warmupMs": 97.08562499145046,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 92.26299996953458,
      "inferenceMs": 106.19895800482482,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.02574520194056359,
      "audioSecondsPerSecond": 38.842189014816825,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 91.36895800475031,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07415556907653809,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02215050496725422,
      "synthesisMs": 91.37083298992366,
      "totalMs": 2736.590915999841,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 352354304,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2688.0446250434034,
      "warmupMs": 94.55275000073016,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 95.23525001714006,
      "inferenceMs": 108.29366702819243,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.026253010188652712,
      "audioSecondsPerSecond": 38.0908700683866,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 93.58466701814905,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07417299598455429,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022687888972348337,
      "synthesisMs": 93.58754201093689,
      "totalMs": 2796.338292071596,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 349061120,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2688.0446250434034,
      "warmupMs": 94.55275000073016,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 93.90433301450685,
      "inferenceMs": 107.84195800079033,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.026143504969888563,
      "audioSecondsPerSecond": 38.25041826456609,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 93.14566705143079,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07415793836116791,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022581181827593933,
      "synthesisMs": 93.14737503882498,
      "totalMs": 2795.8865830441937,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 351600640,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2688.0446250434034,
      "warmupMs": 94.55275000073016,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 93.51508400868624,
      "inferenceMs": 107.19291697023436,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.025986161689753786,
      "audioSecondsPerSecond": 38.48202023595871,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 92.76979201240465,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0741841122508049,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02249026278767622,
      "synthesisMs": 92.7723339991644,
      "totalMs": 2795.2375420136377,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 353828864,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Olá, este é um teste de síntese de voz local.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2688.0446250434034,
      "warmupMs": 94.55275000073016,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "pt"
      ],
      "sampleRate": 24000,
      "durationMs": 92.7924580173567,
      "inferenceMs": 106.72354203416035,
      "audioSeconds": 4.125,
      "realTimeFactor": 0.025872373826463114,
      "audioSecondsPerSecond": 38.651265891078275,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 92.14541700202972,
      "voice": "pt:juliana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.07415556907653809,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.022338626186617395,
      "synthesisMs": 92.14683301979676,
      "totalMs": 2794.7681670775637,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 354107392,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]

KittenTTS 2 Pico · Russian Desktop

39 configurations tested
Passed39
Failed0
TestedNov 13, 2026
Runtimekittenml 0.9.8

Quality criteria

{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch64113 ms / 3681 ms0.024 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64115 ms / 3618 ms0.024 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64115 ms / 3681 ms0.024 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64118 ms / 3681 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64120 ms / 3708 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64120 ms / 3708 ms0.025 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64121 ms / 3708 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64123 ms / 3681 ms0.026 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64135 ms / 3708 ms0.029 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch642302 ms / 3618 ms0.066 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64862 ms / 3618 ms0.174 RTF
FP32cpuPassedVery fastmacos · aarch64793 ms / 3127 ms0.023 RTF
FP32cpuPassedVery fastmacos · aarch64117 ms / 3127 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64117 ms / 2639 ms0.024 RTF
FP32cpuPassedVery fastmacos · aarch64113 ms / 3127 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64113 ms / 2695 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64114 ms / 2695 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64114 ms / 2695 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64115 ms / 2709 ms0.024 RTF
FP32cpuPassedVery fastmacos · aarch64115 ms / 3236 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64115 ms / 2695 ms0.024 RTF
Q8cpuPassedVery fastmacos · aarch64856 ms / 2639 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64116 ms / 2639 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64116 ms / 2709 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64118 ms / 3236 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64118 ms / 2709 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64119 ms / 3236 ms0.025 RTF
Q8cpuPassedVery fastmacos · aarch64119 ms / 2709 ms0.025 RTF
FP32cpuPassedVery fastmacos · aarch64121 ms / 3236 ms0.026 RTF
FP32cpuPassedVery fastmacos · aarch64127 ms / 3186 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch64129 ms / 3186 ms0.027 RTF
FP32cpuPassedVery fastmacos · aarch64136 ms / 3186 ms0.029 RTF
FP32cpuPassedVery fastmacos · aarch64155 ms / 3186 ms0.033 RTF
FP32cpuPassedVery fastmacos · aarch64188 ms / 3308 ms0.038 RTF
FP32cpuPassedVery fastmacos · aarch64181 ms / 3308 ms0.038 RTF
FP32cpuPassedVery fastmacos · aarch641343 ms / 3308 ms0.039 RTF
Q8cpuPassedVery fastmacos · aarch64241 ms / 2719 ms0.049 RTF
Q8cpuPassedVery fastmacos · aarch64236 ms / 2719 ms0.050 RTF
Q8cpuPassedVery fastmacos · aarch641807 ms / 2719 ms0.052 RTF
Test setup & sample details
Quality gate
{ "file": "research/kitten-family-expansion/quality-gates.json", "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85", "description": "Frozen independent ASR error <=0.35, signal/duration/repetition checks and real-time factor <=1." }
Tested
2026-11-13T15:05:56.473135+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3 Max; Mac15,11; 96 GiB", "cpu": "Apple M3 Max", "gpu": "Apple M3 Max", "driver": "OS-managed Metal/MPS macOS 27.2", "runtime": "KittenML 0.9.8 / Torch 2.10.0 / Transformers 5.22.1 / ONNX Runtime 1.24.3; isolated app-owned CPython 3.13.18 development probe", "timingScope": "Three independent helper/model starts per selected precision/device/thread including initial full-suite screening; two additional starts each have three warm calls. CPU1/4 screened independently. Full request timing includes frontend, output encoding/serialization and real init warmup. Verified local snapshots; no Internet transfer or fully cold OS-cache claim. No task builds/downloads/ASR/uploads during timing.", "concurrentUploadActivity": "none; uploads excluded", "firstChunkTimingScope": "Screening/repetition helper firstChunkMs measures synthesis stage; public API records adapter request-to-first-emitted-audio separately." }
Runtime
kittenml 0.9.8
Artifact revision
da8c9d029e16d9302f1827adb303d92316e25f73
Fixture
{
  "file": "research/kitten-family-expansion/quality-gates.json",
  "size": 1428,
  "sha1": "56f497ba801ebb8d5bd40c28051bad76f37f230f",
  "sha256": "bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85",
  "url": "https://assets.localai.tools/samples/v1/files/bb79bb8b0469a77e82dc0051af9e87c8c158a0d0f216b07e5ee2511ac4b27f85/quality-gates.json"
}
Rerun when
artifact revision change; runtime/helper generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 164.46241695666686,
      "inferenceMs": 181.06237496249378,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.03832007935714154,
      "audioSecondsPerSecond": 26.095979360586437,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 161.54687496600673,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08801624923944473,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.03419015004679009,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 171.17029096698388,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 161.54845897108316,
      "loadMs": 3307.636374956928,
      "totalMs": 3488.698749919422,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 171.2803749833256,
      "inferenceMs": 188.14404198201373,
      "audioSeconds": 4.95,
      "realTimeFactor": 0.03800889737010379,
      "audioSecondsPerSecond": 26.30962930238956,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 170.53266597213224,
      "voice": "ru:tatiana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.045562684535980225,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.034451371910182216,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 171.17029096698388,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 170.53429095540196,
      "loadMs": 3307.636374956928,
      "totalMs": 3495.780416938942,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте!  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1230.298499984201,
      "inferenceMs": 1342.627750011161,
      "audioSeconds": 34.775,
      "realTimeFactor": 0.0386089935301556,
      "audioSecondsPerSecond": 25.900701069012555,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 550.0231249607168,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09789133816957474,
      "clippedFraction": 0.0000035945363048166787,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.035214761551938235,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 171.17029096698388,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1224.5933329686522,
      "loadMs": 3307.636374956928,
      "totalMs": 4650.264124968089,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 97.37229195889086,
      "inferenceMs": 112.95112496009097,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.023904999991553644,
      "audioSecondsPerSecond": 41.8322526815867,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 94.37424998031929,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08801624923944473,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.01997382708987783,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 103.24179101735353,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 94.37633299967274,
      "loadMs": 3126.846208004281,
      "totalMs": 3239.797332964372,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 100.30458297114819,
      "inferenceMs": 116.69216601876542,
      "audioSeconds": 4.95,
      "realTimeFactor": 0.023574174953285943,
      "audioSecondsPerSecond": 42.419300017140685,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 99.53533299267292,
      "voice": "ru:tatiana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.045562684535980225,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02010845110017919,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 103.24179101735353,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 99.536832945887,
      "loadMs": 3126.846208004281,
      "totalMs": 3243.5383740230463,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте!  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 683.9505830430426,
      "inferenceMs": 793.4888750314713,
      "audioSeconds": 34.775,
      "realTimeFactor": 0.022817796550150145,
      "audioSecondsPerSecond": 43.82544115520304,
      "placement": {
        "cpu/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 295.0086670462042,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09789133816957474,
      "clippedFraction": 0.0000035945363048166787,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.019531832006308005,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 103.24179101735353,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 679.2194580193609,
      "loadMs": 3126.846208004281,
      "totalMs": 3920.335083035752,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 99.73224997520447,
      "inferenceMs": 114.99874998116866,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.02433835978437432,
      "audioSecondsPerSecond": 41.08740313067517,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 96.91962494980544,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08832743018865585,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": 135544832,
      "synthesisRealTimeFactor": 0.020512460309657313,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 479.7472089994699,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 96.9213749631308,
      "loadMs": 3618.4519170201384,
      "totalMs": 3733.450667001307,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 845.5167500069365,
      "inferenceMs": 861.5487499628216,
      "audioSeconds": 4.95,
      "realTimeFactor": 0.17405025251774173,
      "audioSecondsPerSecond": 5.74546710237071,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 844.7382079903036,
      "voice": "ru:tatiana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04559265077114105,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 189562880,
      "synthesisRealTimeFactor": 0.17065457919744229,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 479.7472089994699,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 844.7401670273393,
      "loadMs": 3618.4519170201384,
      "totalMs": 4480.00066698296,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте!  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 2191.9839999754913,
      "inferenceMs": 2302.4009999935515,
      "audioSeconds": 34.775,
      "realTimeFactor": 0.06620851186178438,
      "audioSecondsPerSecond": 15.1037981655226,
      "placement": {
        "mps:0/torch.float32": 553
      },
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 1078.980375023093,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.0978420153260231,
      "clippedFraction": 0.0000035945363048166787,
      "driverAllocatedMemoryBytes": 1349746688,
      "synthesisRealTimeFactor": 0.06287761440662644,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 479.7472089994699,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 2186.5690409904346,
      "loadMs": 3618.4519170201384,
      "totalMs": 5920.85291701369,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 219.736416998785,
      "inferenceMs": 235.57237500790507,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.049856587303260334,
      "audioSecondsPerSecond": 20.057530089601737,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 217.90575003251433,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08793524652719498,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04611816572606879,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 223.44149998389184,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 217.90833305567503,
      "loadMs": 2719.477500009816,
      "totalMs": 2955.049875017721,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 225.38466600235552,
      "inferenceMs": 241.47399998037145,
      "audioSeconds": 4.95,
      "realTimeFactor": 0.0487826262586609,
      "audioSecondsPerSecond": 20.499101354192863,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 224.63316598441452,
      "voice": "ru:tatiana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04552067816257477,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04538083333766671,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 223.44149998389184,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 224.63512502145022,
      "loadMs": 2719.477500009816,
      "totalMs": 2960.9514999901876,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте!  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 1697.2915420192294,
      "inferenceMs": 1807.3408330092207,
      "audioSeconds": 34.8,
      "realTimeFactor": 0.05193508140831094,
      "audioSecondsPerSecond": 19.254807596007243,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 769.647833018098,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09790635854005814,
      "clippedFraction": 0.0000035919540229885056,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.04860557471241417,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 223.44149998389184,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 1691.473999992013,
      "loadMs": 2719.477500009816,
      "totalMs": 4526.818333019037,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 1
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 99.24658399540931,
      "inferenceMs": 116.3105420419015,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.024615987733735765,
      "audioSecondsPerSecond": 40.624004643515406,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.61179197812453,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08793097734451294,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020658986029466465,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 100.98508396185935,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 97.61370898922905,
      "loadMs": 2639.219375036191,
      "totalMs": 2755.5299170780927,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 101.10391600755975,
      "inferenceMs": 116.70916603179649,
      "audioSeconds": 4.95,
      "realTimeFactor": 0.023577609299352827,
      "audioSecondsPerSecond": 42.41312116523403,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 100.20000004442409,
      "voice": "ru:tatiana",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.04552049934864044,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020242920815662452,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 100.98508396185935,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 100.20245803752914,
      "loadMs": 2639.219375036191,
      "totalMs": 2755.9285410679877,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве.  Здравствуйте, это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте!  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве. Здравствуйте.  Это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "durationMs": 745.7161250058562,
      "inferenceMs": 856.2662919866852,
      "audioSeconds": 34.8,
      "realTimeFactor": 0.024605353218008198,
      "audioSecondsPerSecond": 40.64156247381642,
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "activationPlacement": {},
      "streaming": true,
      "streamChunkCount": 2,
      "firstChunkMs": 324.83612501528114,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.09790635854005814,
      "clippedFraction": 0.0000035919540229885056,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021263350086653306,
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "warmupMs": 100.98508396185935,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "synthesisMs": 739.964583015535,
      "loadMs": 2639.219375036191,
      "totalMs": 3495.4856670228764,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3235.819583002012,
      "warmupMs": 105.76316702645272,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 103.17266697529703,
      "inferenceMs": 118.76033298904076,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.025134462008262595,
      "audioSecondsPerSecond": 39.786011718542625,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 100.3927499987185,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08801624923944473,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021247433858218964,
      "synthesisMs": 100.3941249800846,
      "totalMs": 3354.579915991053,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 548257792,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3235.819583002012,
      "warmupMs": 105.76316702645272,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 100.91912496136501,
      "inferenceMs": 117.54970799665898,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.024878245078657987,
      "audioSecondsPerSecond": 40.195761270068786,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 100.19595798803493,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08801624923944473,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02120587300932991,
      "synthesisMs": 100.19774996908382,
      "totalMs": 3353.369290998671,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 549617664,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3235.819583002012,
      "warmupMs": 105.76316702645272,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 98.9875829545781,
      "inferenceMs": 114.94225001661107,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.024326402119917687,
      "audioSecondsPerSecond": 41.1075996799885,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 98.12583297025412,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08801624923944473,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020767707296366258,
      "synthesisMs": 98.12741697533056,
      "totalMs": 3350.7618330186233,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 549781504,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3235.819583002012,
      "warmupMs": 105.76316702645272,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 104.03108299942687,
      "inferenceMs": 120.77770900214091,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.025561419894632998,
      "audioSecondsPerSecond": 39.12145741989728,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 103.31149998819456,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08801624923944473,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021865229201182803,
      "synthesisMs": 103.31320797558874,
      "totalMs": 3356.597292004153,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 549978112,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3185.7837910065427,
      "warmupMs": 119.67079201713204,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 119.39554096898064,
      "inferenceMs": 135.87749999715015,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.028757142856539714,
      "audioSecondsPerSecond": 34.77396920092804,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 116.49216600926593,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08801624923944473,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02465514095362138,
      "synthesisMs": 116.49554100586101,
      "totalMs": 3321.661291003693,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 549339136,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3185.7837910065427,
      "warmupMs": 119.67079201713204,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 136.9702079682611,
      "inferenceMs": 154.87816702807322,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.032778448048269467,
      "audioSecondsPerSecond": 30.507850723359518,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 136.1605409765616,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08801624923944473,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.028817513218474766,
      "synthesisMs": 136.16274995729327,
      "totalMs": 3340.661958034616,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 552747008,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3185.7837910065427,
      "warmupMs": 119.67079201713204,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 112.25575004937127,
      "inferenceMs": 129.45112498709932,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.027397063489333188,
      "audioSecondsPerSecond": 36.50026216821892,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 111.49804102024063,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08801624923944473,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.023597795343785373,
      "synthesisMs": 111.4995829993859,
      "totalMs": 3315.234915993642,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 553254912,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "PyTorchCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "cpu/torch.float32": 553
      },
      "loadMs": 3185.7837910065427,
      "warmupMs": 119.67079201713204,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 109.46666705422103,
      "inferenceMs": 127.21637496724725,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.02692410052216873,
      "audioSecondsPerSecond": 37.141445047593,
      "activationPlacement": {
        "neuralOutput": [
          "cpu"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 108.66991704097018,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08801624923944473,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02299937397418987,
      "synthesisMs": 108.67204202804714,
      "totalMs": 3313.00016597379,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 558497792,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3708.4455409785733,
      "warmupMs": 533.4985420340672,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 104.66308402828872,
      "inferenceMs": 120.96066703088582,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.025600141170557845,
      "audioSecondsPerSecond": 39.062284592011466,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 101.4640840003267,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08832743018865585,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": 135544832,
      "synthesisRealTimeFactor": 0.021474365078444992,
      "synthesisMs": 101.46637499565259,
      "totalMs": 3829.406208009459,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 651902976,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3708.4455409785733,
      "warmupMs": 533.4985420340672,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 103.92550000688061,
      "inferenceMs": 120.32658298267052,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.025465943488395877,
      "audioSecondsPerSecond": 39.26813080597906,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 103.02004200639203,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08832743018865585,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": 143933440,
      "synthesisRealTimeFactor": 0.02180360677706265,
      "synthesisMs": 103.02204202162102,
      "totalMs": 3828.772123961244,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 653475840,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3708.4455409785733,
      "warmupMs": 533.4985420340672,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 102.99608303466812,
      "inferenceMs": 119.81450003804639,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.025357566145618283,
      "audioSecondsPerSecond": 39.43596141117815,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 102.25108300801367,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08832743018865585,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": 143917056,
      "synthesisRealTimeFactor": 0.021640811221447412,
      "synthesisMs": 102.25283302133903,
      "totalMs": 3828.2600410166197,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 653492224,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3708.4455409785733,
      "warmupMs": 533.4985420340672,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 118.02162497770041,
      "inferenceMs": 134.76083398563787,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.028520811425531825,
      "audioSecondsPerSecond": 35.06211604852168,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 117.2906249994412,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08832743018865585,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": 143917056,
      "synthesisRealTimeFactor": 0.02482375661009755,
      "synthesisMs": 117.29224998271093,
      "totalMs": 3843.206374964211,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 653492224,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3680.6511670001782,
      "warmupMs": 545.3055829857476,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 99.9261669931002,
      "inferenceMs": 115.27554196072742,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.024396940097508448,
      "audioSecondsPerSecond": 40.98874678559077,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.03262499533594,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08832743018865585,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": 135544832,
      "synthesisRealTimeFactor": 0.02053637566320874,
      "synthesisMs": 97.0343750086613,
      "totalMs": 3795.9267089609057,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 651100160,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3680.6511670001782,
      "warmupMs": 545.3055829857476,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 107.06241702428088,
      "inferenceMs": 123.49891697522253,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.02613733692597302,
      "audioSecondsPerSecond": 38.25944482531756,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 106.30633303662762,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08832743018865585,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": 143933440,
      "synthesisRealTimeFactor": 0.022499144555281396,
      "synthesisMs": 106.30845802370459,
      "totalMs": 3804.150083975401,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 651149312,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3680.6511670001782,
      "warmupMs": 545.3055829857476,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 101.31287499098107,
      "inferenceMs": 117.82816698541865,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.024937178197972202,
      "audioSecondsPerSecond": 40.10076810059112,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 100.4398749792017,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08832743018865585,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": 143917056,
      "synthesisRealTimeFactor": 0.02125805100948407,
      "synthesisMs": 100.44429101981223,
      "totalMs": 3798.479333985597,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 653099008,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "torch",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "mps:0/torch.float32": 553
      },
      "loadMs": 3680.6511670001782,
      "warmupMs": 545.3055829857476,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 97.59141696849838,
      "inferenceMs": 113.27991704456508,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.023974585617897372,
      "audioSecondsPerSecond": 41.71083562977145,
      "activationPlacement": {
        "neuralOutput": [
          "mps:0"
        ]
      },
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 96.82012494886294,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08832743018865585,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": 143917056,
      "synthesisRealTimeFactor": 0.020491516823688195,
      "synthesisMs": 96.82241699192673,
      "totalMs": 3793.9310840447433,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "mps",
      "actualDevice": "mps",
      "requestedPrecision": "fp32",
      "actualPrecision": "fp32",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 653115392,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2709.1410830034874,
      "warmupMs": 108.03416598355398,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 103.76908298349008,
      "inferenceMs": 119.3302080500871,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.02525507048679092,
      "audioSecondsPerSecond": 39.59600906768512,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 101.98929201578721,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08793097734451294,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021585555557437516,
      "synthesisMs": 101.99175000889227,
      "totalMs": 2828.4712910535745,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 358170624,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2709.1410830034874,
      "warmupMs": 108.03416598355398,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 101.62795800715685,
      "inferenceMs": 117.90037498576567,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.02495246031444776,
      "audioSecondsPerSecond": 40.07620841384481,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 100.7658329908736,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08762533962726593,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.02132656084738238,
      "synthesisMs": 100.76800000388175,
      "totalMs": 2827.041457989253,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 362201088,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2709.1410830034874,
      "warmupMs": 108.03416598355398,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 100.68924998631701,
      "inferenceMs": 116.48229201091453,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.024652336933526885,
      "audioSecondsPerSecond": 40.564105654422235,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 99.8272500000894,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08810209482908249,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.021127971633283234,
      "synthesisMs": 99.82966596726328,
      "totalMs": 2825.623375014402,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 363134976,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2709.1410830034874,
      "warmupMs": 108.03416598355398,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 98.69095898466185,
      "inferenceMs": 114.86645898548886,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.024310361690050554,
      "audioSecondsPerSecond": 41.13472324063643,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.75495901703835,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08812806010246277,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020689329947960914,
      "synthesisMs": 97.75708400411531,
      "totalMs": 2824.0075419889763,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 363151360,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2694.6596250054426,
      "warmupMs": 104.70175003865734,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 100.47316603595391,
      "inferenceMs": 115.05637498339638,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.02435055555204156,
      "audioSecondsPerSecond": 41.066824855918306,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 98.65529101807624,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08793097734451294,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020879911746652352,
      "synthesisMs": 98.65758300293237,
      "totalMs": 2809.715999988839,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 359546880,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2694.6596250054426,
      "warmupMs": 104.70175003865734,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 98.17595797358081,
      "inferenceMs": 113.78833401249722,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.024082187092592005,
      "audioSecondsPerSecond": 41.52446769702296,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.34454198041931,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08762533962726593,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020602442750401757,
      "synthesisMs": 97.3465419956483,
      "totalMs": 2808.44795901794,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 362151936,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2694.6596250054426,
      "warmupMs": 104.70175003865734,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 98.07675000047311,
      "inferenceMs": 113.41637501027435,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.024003465610640075,
      "audioSecondsPerSecond": 41.66065085021421,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.24454197566956,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08810209482908249,
      "clippedFraction": 0.00001763668430335097,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020581305177300852,
      "synthesisMs": 97.24666696274653,
      "totalMs": 2808.076000015717,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 363020288,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  },
  {
    "precision": "q8",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "speedPassed": true,
      "asrText": " Здравствуйте, это проверка синтеза речи на этом устройстве.",
      "errorMetric": "wer",
      "errorRate": 0,
      "maximumRepeatedWords": 1,
      "maximumErrorRate": 0.35,
      "signalPassed": true
    },
    "runtime": {
      "engine": "kittenml",
      "version": "0.9.8",
      "torchVersion": "2.10.0",
      "transformersVersion": "5.22.1",
      "onnxruntimeVersion": "1.24.3",
      "backend": "onnxruntime",
      "provider": "ONNXRuntimeCPU",
      "acceleration": "cpu",
      "accelerationDetails": "TTS neural network on selected device; EnglishBART G2P, spaCy/espeak preprocessing, output transfer and WAV encoding onCPU. ONNXQ8weights useFP32arithmetic.",
      "placement": {
        "CPUExecutionProvider/int8-weights-fp32-arithmetic": 1
      },
      "loadMs": 2694.6596250054426,
      "warmupMs": 104.70175003865734,
      "warmupScope": "Real short synthesis included in measured initialization; no task certification inferred",
      "languages": [
        "ru"
      ],
      "sampleRate": 24000,
      "durationMs": 98.57016702881083,
      "inferenceMs": 114.15520898299292,
      "audioSeconds": 4.725,
      "realTimeFactor": 0.02415983258899321,
      "audioSecondsPerSecond": 41.391015286073724,
      "activationPlacement": {},
      "streaming": false,
      "streamChunkCount": 1,
      "firstChunkMs": 97.83820901066065,
      "voice": "ru:olga",
      "speed": 1,
      "seed": 42,
      "seedScope": "Torch/NumPy reset per request; ORT noise state is session-scoped and successiveQ8calls need not be bit-identical",
      "frontendVersion": "v1",
      "rms": 0.08812806010246277,
      "clippedFraction": 0.000008818342151675484,
      "driverAllocatedMemoryBytes": null,
      "synthesisRealTimeFactor": 0.020706948994230184,
      "synthesisMs": 97.84033399773762,
      "totalMs": 2808.8148339884356,
      "grade": "fast",
      "artifactRevision": "da8c9d029e16d9302f1827adb303d92316e25f73",
      "requestedDevice": "cpu",
      "actualDevice": "cpu",
      "requestedPrecision": "q8",
      "actualPrecision": "q8",
      "fallback": false,
      "cpuThreads": 4,
      "residentMemoryBytes": 363511808,
      "memoryMeasurementScope": "Post-request process RSS and Metal driver allocation, not peak memory or validated minima"
    }
  }
]
Release validation still required
  • Public asset publication, complete backup and forced source recovery excluded and pending

Something not quite right?

Tell us what happened with KittenTTS 2. Loading, results, or the page itself—we’re here for all of it.

Attachments Optional

Up to 3 files · 50 MB each

Browser, OS, CPU/GPU, memory, screen, storage, and network details are included automatically. Model inputs, results, and browsing history are not collected.