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.

Pocket TTS September2026

Generate English preset speech locally with the September 2026 24-layer English checkpoint.

Desktop inferencePrivate · Runs on your computer
About this model

Generate English preset speech locally with the September 2026 24-layer English checkpoint.

Created by
Kyutai
License
CC-BY-4.0
Model released
Unknown
Source repository created
Dec 29, 2025
Model weights · FP32
1.3 GB

Best for

  • Local English narration and accessibility drafts with review
  • Measured CPU/Metal preset speech generation and incremental playback

Keep in mind

  • English text and two official preset voices. No personal voice reference input.
  • Numbers, names and unusual phrasing can be spoken incorrectly. Results were checked on a limited four-text fixture suite.
  • Only the published FP32 runtime configurations with repeated local passing evidence are advertised.
  • Streaming chunks reproduce the final waveform from that request. Independent calls may differ.

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 · cpu (verified platform only)

Sources & technical details
Original model
https://huggingface.co/kyutai/pocket-tts
Release verification
Unknown
Release evidence
Exact original checkpoint first-public day is unverified. The publisher configuration names September 2026.
Source updated
Oct 21, 2026
Native model files
https://huggingface.co/kyutai/pocket-tts-without-voice-cloning
Files updated
Oct 21, 2026
Artifact revision
a7056f505a4a8ca72d8b5fb1440071d6f1a35739

Desktop app

Sample result

Pocket TTS 24-layer English · September 2026

Run Pocket TTS September2026 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 / FP32TestedNot establishedRAM estimate
Windows Intel / AMDCPU / FP32UnverifiedNot establishedRAM estimate
macOS Intel / AMDCPU / FP32UnverifiedNot establishedRAM 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.

Pocket TTS 24-layer English · September 2026

49 configurations tested
Passed42
Failed7
TestedNov 15, 2026
Runtimepocket-tts 4.0.1

Quality criteria

{ "file": "research/pocket-tts-family-expansion/quality-gates.json", "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d", "description": "Frozen WER<=.35, repeated-word<=3, duration1–40sec, RMS>=.005, clipping<=.02, RTF<=1; independent GraniteCTC after TTS disposal." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32cpuPassedFastmacos · aarch643405 ms / 3114 ms0.311 RTF
FP32cpuPassedFastmacos · aarch641176 ms / 3114 ms0.320 RTF
FP32cpuPassedFastmacos · aarch641407 ms / 2654 ms0.320 RTF
FP32cpuPassedFastmacos · aarch643509 ms / 1110 ms0.320 RTF
FP32cpuPassedFastmacos · aarch641183 ms / 3114 ms0.321 RTF
FP32cpuPassedFastmacos · aarch643541 ms / 2654 ms0.323 RTF
FP32cpuPassedFastmacos · aarch641189 ms / 3114 ms0.323 RTF
FP32cpuPassedFastmacos · aarch641190 ms / 2654 ms0.323 RTF
FP32cpuPassedFastmacos · aarch641433 ms / 3114 ms0.326 RTF
FP32cpuPassedFastmacos · aarch643590 ms / 999 ms0.328 RTF
FP32cpuPassedFastmacos · aarch641221 ms / 1110 ms0.332 RTF
FP32cpuPassedFastmacos · aarch641226 ms / 1110 ms0.333 RTF
FP32cpuPassedFastmacos · aarch64936 ms / 1110 ms0.334 RTF
FP32cpuPassedFastmacos · aarch641242 ms / 1110 ms0.337 RTF
FP32cpuPassedFastmacos · aarch641487 ms / 999 ms0.338 RTF
FP32cpuPassedFastmacos · aarch641251 ms / 2654 ms0.340 RTF
FP32cpuPassedFastmacos · aarch64953 ms / 999 ms0.340 RTF
FP32cpuPassedFastmacos · aarch641264 ms / 3114 ms0.343 RTF
FP32cpuPassedFastmacos · aarch64966 ms / 2654 ms0.345 RTF
FP32cpuPassedFastmacos · aarch641278 ms / 999 ms0.347 RTF
FP32cpuPassedFastmacos · aarch641287 ms / 1110 ms0.350 RTF
FP32cpuPassedFastmacos · aarch641289 ms / 2654 ms0.350 RTF
FP32cpuPassedFastmacos · aarch643907 ms / 1969 ms0.356 RTF
FP32cpuPassedFastmacos · aarch641001 ms / 3114 ms0.358 RTF
FP32cpuPassedFastmacos · aarch641577 ms / 1969 ms0.358 RTF
FP32cpuPassedFastmacos · aarch641319 ms / 1969 ms0.358 RTF
FP32cpuPassedFastmacos · aarch641023 ms / 3316 ms0.365 RTF
FP32cpuPassedFastmacos · aarch641345 ms / 999 ms0.365 RTF
FP32cpuPassedFastmacos · aarch641629 ms / 3316 ms0.370 RTF
FP32cpuPassedFastmacos · aarch641379 ms / 999 ms0.375 RTF
FP32cpuPassedFastmacos · aarch641664 ms / 1110 ms0.378 RTF
FP32cpuPassedFastmacos · aarch641408 ms / 3316 ms0.383 RTF
FP32cpuPassedFastmacos · aarch641429 ms / 1969 ms0.388 RTF
FP32cpuPassedFastmacos · aarch641473 ms / 3316 ms0.400 RTF
FP32cpuPassedFastmacos · aarch641173 ms / 1969 ms0.419 RTF
FP32cpuPassedFastmacos · aarch644594 ms / 3316 ms0.419 RTF
FP32cpuPassedFastmacos · aarch641619 ms / 1969 ms0.440 RTF
FP32cpuPassedFastmacos · aarch641632 ms / 999 ms0.444 RTF
FP32cpuPassedFastmacos · aarch641692 ms / 3316 ms0.460 RTF
FP32cpuPassedFastmacos · aarch641713 ms / 3316 ms0.465 RTF
FP32cpuPassedFastmacos · aarch641835 ms / 2654 ms0.499 RTF
FP32cpuPassedModeratemacos · aarch642091 ms / 1969 ms0.568 RTF
FP32mpsFailed—macos · aarch645891 ms / 6039 ms1.712 RTF
FP32mpsFailed—macos · aarch644680 ms / 6039 ms1.104 RTF
FP32mpsFailed—macos · aarch645264 ms / 6039 ms1.779 RTF
FP32mpsFailed—macos · aarch6412709 ms / 6039 ms1.251 RTF
FP32mpsFailed—macos · aarch642530 ms / 6039 ms0.736 RTF
FP32mpsFailed—macos · aarch642563 ms / 6039 ms0.745 RTF
FP32mpsFailed—macos · aarch642406 ms / 6039 ms0.700 RTF
Test setup & sample details
Quality gate
{ "file": "research/pocket-tts-family-expansion/quality-gates.json", "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d", "description": "Frozen WER<=.35, repeated-word<=3, duration1–40sec, RMS>=.005, clipping<=.02, RTF<=1; independent GraniteCTC after TTS disposal." }
Tested
2026-11-15T08:21:34.004767+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3Max;Mac15,11;96GiB", "cpu": "Apple M3Max", "gpu": "Apple M3Max", "driver": "OS-managedMetal/MPS macOS27.2 build26B118", "runtime": "PocketTTS4.0.1/Torch2.10.0; isolated app-owned CPython3.13.18 development snapshot", "timingScope": "Up to three fresh helpers per profile, four frozen fixtures and three warm repeats. Synthesis and full sidecar request timings are retained. No network transfer timing.", "concurrentUploadActivity": "The workspace's background mirror agent was active. Individual overlapping transfers were not instrumented." }
Runtime
pocket-tts 4.0.1
Artifact revision
a7056f505a4a8ca72d8b5fb1440071d6f1a35739
Fixture
{
  "file": "research/pocket-tts-family-expansion/quality-gates.json",
  "size": 2455,
  "sha1": "2ee2afd004929d50b3d7eee062620472a957398d",
  "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d",
  "url": "https://assets.localai.tools/samples/v1/files/4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d/quality-gates.json"
}
Rerun when
artifact revision change; runtime or helper/model generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3114.1506250132807,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1266.4951250189915,
      "inferenceMs": 1263.6675409739837,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.34338791874293034,
      "audioSecondsPerSecond": 2.9121583649791347,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4377.818165987264,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9 thirty0. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3114.1506250132807,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1433.9843750349246,
      "inferenceMs": 1433.147917035967,
      "audioSeconds": 4.4,
      "realTimeFactor": 0.3257154356899925,
      "audioSecondsPerSecond": 3.0701645989899418,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08798035979270935,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4547.298542049248,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3114.1506250132807,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1001.8244170350954,
      "inferenceMs": 1001.2527080252767,
      "audioSeconds": 2.8,
      "realTimeFactor": 0.35759025286617024,
      "audioSecondsPerSecond": 2.7964968060084523,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.017500048503279686,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4115.403333038557,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3114.1506250132807,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 3406.248124956619,
      "inferenceMs": 3404.5912500005215,
      "audioSeconds": 10.96,
      "realTimeFactor": 0.3106378877737702,
      "audioSecondsPerSecond": 3.2191823320929704,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08620789647102356,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 6518.741875013802,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3114.1506250132807,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1176.5119169722311,
      "inferenceMs": 1175.850083003752,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.3195244790771065,
      "audioSecondsPerSecond": 3.1296506699215483,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4290.000708017033,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3114.1506250132807,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1189.766499970574,
      "inferenceMs": 1189.05604199972,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.32311305489122827,
      "audioSecondsPerSecond": 3.094891973141218,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4303.206667013001,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3114.1506250132807,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1183.5876669501886,
      "inferenceMs": 1182.9120829934254,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.3214435008134308,
      "audioSecondsPerSecond": 3.110966616122099,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4297.062708006706,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1109.7997500328347,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1289.3964580143802,
      "inferenceMs": 1286.500749993138,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.349592595106831,
      "audioSecondsPerSecond": 2.8604724871086384,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2396.3005000259727,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9 thirty0. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1109.7997500328347,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1665.1339170057327,
      "inferenceMs": 1664.345459023025,
      "audioSeconds": 4.4,
      "realTimeFactor": 0.37826033159614203,
      "audioSecondsPerSecond": 2.643681920809164,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08798035979270935,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2774.1452090558596,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1109.7997500328347,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 936.078250000719,
      "inferenceMs": 935.5248750071041,
      "audioSeconds": 2.8,
      "realTimeFactor": 0.3341160267882515,
      "audioSecondsPerSecond": 2.9929722606026243,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.017500048503279686,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2045.3246250399388,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1109.7997500328347,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 3510.396249999758,
      "inferenceMs": 3508.726291998755,
      "audioSeconds": 10.96,
      "realTimeFactor": 0.3201392602188645,
      "audioSecondsPerSecond": 3.1236406285075624,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08620789647102356,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4618.52604203159,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1109.7997500328347,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1226.9051250186749,
      "inferenceMs": 1226.278165995609,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.3332277624988068,
      "audioSecondsPerSecond": 3.000950438526504,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2336.0779160284437,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1109.7997500328347,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1221.567332977429,
      "inferenceMs": 1220.8697500173002,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.3317580842438316,
      "audioSecondsPerSecond": 3.014244557986512,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2330.669500050135,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1109.7997500328347,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1242.2922080149874,
      "inferenceMs": 1241.5888750110753,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.33738828125300957,
      "audioSecondsPerSecond": 2.9639440833159636,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2351.38862504391,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3316.3682500016876,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1715.6989999930374,
      "inferenceMs": 1712.5539170228876,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.4653679122344803,
      "audioSecondsPerSecond": 2.1488374546462925,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 5028.922167024575,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9 thirty0. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3316.3682500016876,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1629.447707964573,
      "inferenceMs": 1628.678749955725,
      "audioSeconds": 4.4,
      "realTimeFactor": 0.3701542613535738,
      "audioSecondsPerSecond": 2.701576354526399,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08798035979270935,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4945.0469999574125,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3316.3682500016876,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1023.8201249740086,
      "inferenceMs": 1023.1950000161305,
      "audioSeconds": 2.8,
      "realTimeFactor": 0.3654267857200466,
      "audioSecondsPerSecond": 2.736526273052408,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.017500048503279686,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4339.563250017818,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3316.3682500016876,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 4595.731583016459,
      "inferenceMs": 4593.955082993489,
      "audioSeconds": 10.96,
      "realTimeFactor": 0.4191564856745884,
      "audioSecondsPerSecond": 2.3857438311865913,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08620789647102356,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 7910.323332995176,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3316.3682500016876,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1408.4725410211831,
      "inferenceMs": 1407.8067910159007,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.38255619321084255,
      "audioSecondsPerSecond": 2.6139950620244137,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4724.175041017588,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3316.3682500016876,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1474.1829590057023,
      "inferenceMs": 1473.484667018056,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.4004034421244717,
      "audioSecondsPerSecond": 2.497481027371224,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4789.8529170197435,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 3316.3682500016876,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1692.4622499500401,
      "inferenceMs": 1691.7342499946244,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.45971039402027836,
      "audioSecondsPerSecond": 2.1752825539896077,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 5008.102499996312,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1969.072000007145,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1432.7660840353929,
      "inferenceMs": 1429.3107920093462,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.38839967174167017,
      "audioSecondsPerSecond": 2.5746674695057763,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3398.3827920164913,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9 thirty0. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1969.072000007145,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1577.5609579868615,
      "inferenceMs": 1576.7959580407478,
      "audioSeconds": 4.4,
      "realTimeFactor": 0.3583627177365336,
      "audioSecondsPerSecond": 2.790468847641665,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08798035979270935,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3545.867958047893,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1969.072000007145,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1174.0419589914382,
      "inferenceMs": 1173.3937920071185,
      "audioSeconds": 2.8,
      "realTimeFactor": 0.41906921143111375,
      "audioSecondsPerSecond": 2.3862406798748546,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.017500048503279686,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3142.4657920142636,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1969.072000007145,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 3908.4988340036944,
      "inferenceMs": 3906.818541989196,
      "audioSeconds": 10.96,
      "realTimeFactor": 0.35646154580193395,
      "audioSecondsPerSecond": 2.805351690181036,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08620789647102356,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 5875.890541996341,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1969.072000007145,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1319.515000039246,
      "inferenceMs": 1318.8492500339635,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.3583829483787944,
      "audioSecondsPerSecond": 2.7903113262605497,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3287.9212500411086,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1969.072000007145,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1619.9707920313813,
      "inferenceMs": 1619.1441669943742,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.4399848279876017,
      "audioSecondsPerSecond": 2.2728056432622696,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3588.2161670015194,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1969.072000007145,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 2091.9980829930864,
      "inferenceMs": 2091.3325409637764,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.568296886131461,
      "audioSecondsPerSecond": 1.7596436376895361,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4060.4045409709215,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 2654.217542032711,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1838.1308749667369,
      "inferenceMs": 1835.1831249892712,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.4986910665731715,
      "audioSecondsPerSecond": 2.0052494761368918,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4489.400667021982,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9 thirty0. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 2654.217542032711,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1407.8397499979474,
      "inferenceMs": 1406.8403339479119,
      "audioSeconds": 4.4,
      "realTimeFactor": 0.31973643953361636,
      "audioSecondsPerSecond": 3.127575954303646,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08798035979270935,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4061.057875980623,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 2654.217542032711,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 966.3094579591416,
      "inferenceMs": 965.7587499823421,
      "audioSeconds": 2.8,
      "realTimeFactor": 0.3449138392794079,
      "audioSecondsPerSecond": 2.8992747930590275,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.017500048503279686,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3619.9762920150533,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 2654.217542032711,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 3542.635583959054,
      "inferenceMs": 3541.0075839608908,
      "audioSeconds": 10.96,
      "realTimeFactor": 0.323084633573074,
      "audioSecondsPerSecond": 3.0951642266013994,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08620789647102356,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 6195.225125993602,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 2654.217542032711,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1191.1206250078976,
      "inferenceMs": 1190.4712920077145,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.32349763369774853,
      "audioSecondsPerSecond": 3.0912127194547696,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3844.6888340404257,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 2654.217542032711,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1252.1338749793358,
      "inferenceMs": 1251.4585409662686,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.3400702556973556,
      "audioSecondsPerSecond": 2.940568847896967,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3905.67608299898,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 2654.217542032711,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1290.4785409918986,
      "inferenceMs": 1289.319666044321,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.3503586049033481,
      "audioSecondsPerSecond": 2.8542184664648547,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3943.537208077032,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 998.712082975544,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1282.3682500165887,
      "inferenceMs": 1278.4341250080615,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.3474005774478428,
      "audioSecondsPerSecond": 2.8785214099137058,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2277.1462079836056,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9 thirty0. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 998.712082975544,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1487.945957982447,
      "inferenceMs": 1487.1904579922557,
      "audioSeconds": 4.4,
      "realTimeFactor": 0.3379978313618763,
      "audioSecondsPerSecond": 2.9585988642907988,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08798035979270935,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2485.9025409677997,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 998.712082975544,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 953.7004579906352,
      "inferenceMs": 953.1681669759564,
      "audioSeconds": 2.8,
      "realTimeFactor": 0.340417202491413,
      "audioSecondsPerSecond": 2.937571875572959,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.017500048503279686,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1951.8802499515004,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 998.712082975544,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 3592.1478749951348,
      "inferenceMs": 3590.2164169820026,
      "audioSeconds": 10.96,
      "realTimeFactor": 0.32757449060054766,
      "audioSecondsPerSecond": 3.0527407618544524,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08620789647102356,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4588.928499957547,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 998.712082975544,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1632.9879579716362,
      "inferenceMs": 1632.230624963995,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.4435409306967378,
      "audioSecondsPerSecond": 2.2545833558791215,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2630.942707939539,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 998.712082975544,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1379.4875830062665,
      "inferenceMs": 1378.79950000206,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.37467377717447287,
      "audioSecondsPerSecond": 2.668988493246844,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2377.511582977604,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 998.712082975544,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 354
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1345.5817499780096,
      "inferenceMs": 1344.8124999995343,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.36543817934769957,
      "audioSecondsPerSecond": 2.736440953665492,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08008776605129242,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2343.5245829750784,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 2727.0722079556435,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 6038.822082977276,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 354
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 5894.819832989015,
      "inferenceMs": 5890.653375012334,
      "audioSeconds": 3.44,
      "realTimeFactor": 1.7123992369221903,
      "audioSecondsPerSecond": 0.5839759668413348,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.07823096960783005,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 2306605056,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 11929.47545798961,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "text": "The meeting starts at nine thirty. Please bring three notebooks.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 2727.0722079556435,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 6038.822082977276,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 354
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 4682.097083015833,
      "inferenceMs": 4679.743125045206,
      "audioSeconds": 4.24,
      "realTimeFactor": 1.1037130011899072,
      "audioSecondsPerSecond": 0.9060326361308649,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08929258584976196,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 2316550144,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 10718.565208022483,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 2727.0722079556435,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 6038.822082977276,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 354
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 5266.591500025243,
      "inferenceMs": 5264.408041024581,
      "audioSeconds": 2.96,
      "realTimeFactor": 1.778516230075872,
      "audioSecondsPerSecond": 0.5622664460910429,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "marius",
      "rms": 0.02245883084833622,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 2316550144,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 11303.230124001857,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 2727.0722079556435,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 6038.822082977276,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 354
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 12711.522457946558,
      "inferenceMs": 12708.948749990668,
      "audioSeconds": 10.16,
      "realTimeFactor": 1.2508807824793964,
      "audioSecondsPerSecond": 0.7994366961317285,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.08163488656282425,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 2456469504,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 18747.770832967944,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 2727.0722079556435,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 6038.822082977276,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 354
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 2531.051958038006,
      "inferenceMs": 2530.2433330216445,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.735535852622571,
      "audioSecondsPerSecond": 1.3595530339336628,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.07823096960783005,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 2456469504,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 8569.06541599892,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 2727.0722079556435,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 6038.822082977276,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 354
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 2564.208166033495,
      "inferenceMs": 2563.2219160324894,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.7451226500094446,
      "audioSecondsPerSecond": 1.3420609345150423,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.07823096960783005,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 2456469504,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 8602.043999009766,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 2727.0722079556435,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 6038.822082977276,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 354
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 2407.366916013416,
      "inferenceMs": 2406.3549999846146,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.6995218023211088,
      "audioSecondsPerSecond": 1.4295480093427588,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.07823096960783005,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 2456469504,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 8445.17708296189,
      "grade": "moderate"
    }
  }
]

Pocket TTS Drifting English · September 2026

63 configurations tested
Passed63
Failed0
TestedNov 15, 2026
Runtimepocket-tts 4.0.1

Quality criteria

{ "file": "research/pocket-tts-family-expansion/quality-gates.json", "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d", "description": "Frozen WER<=.35, repeated-word<=3, duration1–40sec, RMS>=.005, clipping<=.02, RTF<=1; independent GraniteCTC after TTS disposal." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedFastmacos · aarch641136 ms / 1820 ms0.284 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641175 ms / 1820 ms0.294 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641208 ms / 1820 ms0.302 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641341 ms / 1584 ms0.335 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64751 ms / 1654 ms0.335 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641343 ms / 1654 ms0.336 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641350 ms / 1654 ms0.338 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64757 ms / 1820 ms0.338 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64758 ms / 1584 ms0.338 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641365 ms / 1584 ms0.341 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641384 ms / 1584 ms0.346 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641403 ms / 1654 ms0.351 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641101 ms / 1654 ms0.353 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641123 ms / 1820 ms0.360 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641137 ms / 1584 ms0.365 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch645508 ms / 1820 ms0.662 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch645954 ms / 1584 ms0.716 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch645987 ms / 1654 ms0.720 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch643363 ms / 1584 ms0.841 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch643497 ms / 1820 ms0.874 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch643530 ms / 1654 ms0.882 RTF
FP32cpuPassedFastmacos · aarch64993 ms / 344 ms0.125 RTF
FP32cpuPassedFastmacos · aarch64416 ms / 344 ms0.130 RTF
FP32cpuPassedFastmacos · aarch64379 ms / 344 ms0.132 RTF
FP32cpuPassedFastmacos · aarch641060 ms / 332 ms0.134 RTF
FP32cpuPassedFastmacos · aarch64388 ms / 344 ms0.135 RTF
FP32cpuPassedFastmacos · aarch64281 ms / 344 ms0.135 RTF
FP32cpuPassedFastmacos · aarch64391 ms / 344 ms0.136 RTF
FP32cpuPassedFastmacos · aarch64392 ms / 344 ms0.136 RTF
FP32cpuPassedFastmacos · aarch641086 ms / 733 ms0.137 RTF
FP32cpuPassedFastmacos · aarch64417 ms / 332 ms0.145 RTF
FP32cpuPassedFastmacos · aarch64420 ms / 733 ms0.146 RTF
FP32cpuPassedFastmacos · aarch64421 ms / 332 ms0.146 RTF
FP32cpuPassedFastmacos · aarch64422 ms / 733 ms0.146 RTF
FP32cpuPassedFastmacos · aarch64469 ms / 733 ms0.146 RTF
FP32cpuPassedFastmacos · aarch64423 ms / 733 ms0.147 RTF
FP32cpuPassedFastmacos · aarch641173 ms / 323 ms0.148 RTF
FP32cpuPassedFastmacos · aarch64427 ms / 332 ms0.148 RTF
FP32cpuPassedFastmacos · aarch64480 ms / 332 ms0.150 RTF
FP32cpuPassedFastmacos · aarch64313 ms / 332 ms0.150 RTF
FP32cpuPassedFastmacos · aarch64443 ms / 487 ms0.154 RTF
FP32cpuPassedFastmacos · aarch64445 ms / 315 ms0.155 RTF
FP32cpuPassedFastmacos · aarch641229 ms / 315 ms0.155 RTF
FP32cpuPassedFastmacos · aarch64323 ms / 733 ms0.155 RTF
FP32cpuPassedFastmacos · aarch641232 ms / 487 ms0.156 RTF
FP32cpuPassedFastmacos · aarch64449 ms / 323 ms0.156 RTF
FP32cpuPassedFastmacos · aarch64449 ms / 487 ms0.156 RTF
FP32cpuPassedFastmacos · aarch64451 ms / 733 ms0.157 RTF
FP32cpuPassedFastmacos · aarch64505 ms / 487 ms0.158 RTF
FP32cpuPassedFastmacos · aarch64456 ms / 323 ms0.158 RTF
FP32cpuPassedFastmacos · aarch64332 ms / 323 ms0.159 RTF
FP32cpuPassedFastmacos · aarch64459 ms / 323 ms0.160 RTF
FP32cpuPassedFastmacos · aarch64511 ms / 315 ms0.160 RTF
FP32cpuPassedFastmacos · aarch64461 ms / 487 ms0.160 RTF
FP32cpuPassedFastmacos · aarch64335 ms / 315 ms0.161 RTF
FP32cpuPassedFastmacos · aarch64464 ms / 315 ms0.161 RTF
FP32cpuPassedFastmacos · aarch64466 ms / 315 ms0.162 RTF
FP32cpuPassedFastmacos · aarch64518 ms / 323 ms0.162 RTF
FP32cpuPassedFastmacos · aarch64474 ms / 315 ms0.164 RTF
FP32cpuPassedFastmacos · aarch64477 ms / 487 ms0.165 RTF
FP32cpuPassedFastmacos · aarch64350 ms / 487 ms0.168 RTF
FP32cpuPassedFastmacos · aarch64498 ms / 323 ms0.173 RTF
FP32cpuPassedFastmacos · aarch64498 ms / 332 ms0.173 RTF
Test setup & sample details
Quality gate
{ "file": "research/pocket-tts-family-expansion/quality-gates.json", "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d", "description": "Frozen WER<=.35, repeated-word<=3, duration1–40sec, RMS>=.005, clipping<=.02, RTF<=1; independent GraniteCTC after TTS disposal." }
Tested
2026-11-15T08:24:15.681813+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3Max;Mac15,11;96GiB", "cpu": "Apple M3Max", "gpu": "Apple M3Max", "driver": "OS-managedMetal/MPS macOS27.2 build26B118", "runtime": "PocketTTS4.0.1/Torch2.10.0; isolated app-owned CPython3.13.18 development snapshot", "timingScope": "Up to three fresh helpers per profile, four frozen fixtures and three warm repeats. Synthesis and full sidecar request timings are retained. No network transfer timing.", "concurrentUploadActivity": "The workspace's background mirror agent was active. Individual overlapping transfers were not instrumented." }
Runtime
pocket-tts 4.0.1
Artifact revision
a7056f505a4a8ca72d8b5fb1440071d6f1a35739
Fixture
{
  "file": "research/pocket-tts-family-expansion/quality-gates.json",
  "size": 2455,
  "sha1": "2ee2afd004929d50b3d7eee062620472a957398d",
  "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d",
  "url": "https://assets.localai.tools/samples/v1/files/4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d/quality-gates.json"
}
Rerun when
artifact revision change; runtime or helper/model generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.67545800097287,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 430.1058749551885,
      "inferenceMs": 427.48187499819323,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.14843120659659487,
      "audioSecondsPerSecond": 6.737127743748603,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 759.1573329991661,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.67545800097287,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 480.37804203340784,
      "inferenceMs": 479.709500039462,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.14990921876233188,
      "audioSecondsPerSecond": 6.6707038316663745,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11256179213523865,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 811.3849580404349,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.67545800097287,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 313.05712496396154,
      "inferenceMs": 312.6301249722019,
      "audioSeconds": 2.08,
      "realTimeFactor": 0.15030294469817398,
      "audioSecondsPerSecond": 6.6532295957881455,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.024928046390414238,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 644.3055829731748,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.67545800097287,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1061.142791993916,
      "inferenceMs": 1059.7598330350593,
      "audioSeconds": 7.92,
      "realTimeFactor": 0.1338080597266489,
      "audioSecondsPerSecond": 7.473391378986137,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12433581799268723,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1391.4352910360321,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.67545800097287,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 417.9809170309454,
      "inferenceMs": 417.4032080336474,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.14493166945612757,
      "audioSecondsPerSecond": 6.899803222805704,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 749.0786660346203,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.67545800097287,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 421.1780829937197,
      "inferenceMs": 420.5553749925457,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.1460261718724117,
      "audioSecondsPerSecond": 6.848087484438993,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 752.2308329935186,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.67545800097287,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 498.72695800149813,
      "inferenceMs": 498.1092499801889,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.1729546006875656,
      "audioSecondsPerSecond": 5.781864119396588,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 829.7847079811618,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 733.0072920303792,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 453.6728750099428,
      "inferenceMs": 450.9373330511153,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.1565754628649706,
      "audioSecondsPerSecond": 6.386696751216964,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1183.9446250814945,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 733.0072920303792,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 469.4123329827562,
      "inferenceMs": 468.7792919576168,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.14649352873675525,
      "audioSecondsPerSecond": 6.826240098270634,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11256179213523865,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1201.786583987996,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 733.0072920303792,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 323.5269589931704,
      "inferenceMs": 323.10924999183044,
      "audioSeconds": 2.08,
      "realTimeFactor": 0.15534098557299542,
      "audioSecondsPerSecond": 6.43745110996541,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.024928046390414238,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1056.1165420222096,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 733.0072920303792,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1087.4613749911077,
      "inferenceMs": 1086.2386249937117,
      "audioSeconds": 7.92,
      "realTimeFactor": 0.13715134153961006,
      "audioSecondsPerSecond": 7.291215592748646,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12433581799268723,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1819.2459170240909,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 733.0072920303792,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 422.4934160010889,
      "inferenceMs": 421.76762502640486,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.14644709202305725,
      "audioSecondsPerSecond": 6.828404621667433,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1154.774917056784,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 733.0072920303792,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 423.68237499613315,
      "inferenceMs": 423.06266695959494,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.14689675936097046,
      "audioSecondsPerSecond": 6.807502114751849,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1156.069958989974,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 733.0072920303792,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 420.9905419847928,
      "inferenceMs": 420.4147079726681,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.14597732915717643,
      "audioSecondsPerSecond": 6.850378793567883,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1153.4220000030473,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 343.56833301717415,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 390.9606669913046,
      "inferenceMs": 387.6225419808179,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.1345911604100062,
      "audioSecondsPerSecond": 7.429908449809715,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 731.1908749979921,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 343.56833301717415,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 417.09491598885506,
      "inferenceMs": 416.4043749915436,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.13012636718485737,
      "audioSecondsPerSecond": 7.6848376054285845,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11256179213523865,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 759.9727080087177,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 343.56833301717415,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 281.4340420300141,
      "inferenceMs": 280.9804590069689,
      "audioSeconds": 2.08,
      "realTimeFactor": 0.1350867591379658,
      "audioSecondsPerSecond": 7.402650018264835,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.024928046390414238,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 624.548792024143,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 343.56833301717415,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 993.9255830249749,
      "inferenceMs": 992.6394170033745,
      "audioSeconds": 7.92,
      "realTimeFactor": 0.1253332597226483,
      "audioSecondsPerSecond": 7.97872809031628,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12433581799268723,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1336.2077500205487,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 343.56833301717415,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 392.8949169931002,
      "inferenceMs": 392.37441698787734,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.13624111700967964,
      "audioSecondsPerSecond": 7.339928077138065,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 735.9427500050515,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 343.56833301717415,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 379.47804102441296,
      "inferenceMs": 378.87562497053295,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.1315540364481017,
      "audioSecondsPerSecond": 7.601439127217518,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 722.4439579877071,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 343.56833301717415,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 391.58929203404114,
      "inferenceMs": 391.02108299266547,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.1357712093724533,
      "audioSecondsPerSecond": 7.365331756431202,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 734.5894160098396,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 486.81149998446926,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 479.49320799671113,
      "inferenceMs": 476.5239170519635,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.16545969342082067,
      "audioSecondsPerSecond": 6.043767997663682,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 963.3354170364328,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 486.81149998446926,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 505.2993340068497,
      "inferenceMs": 504.59695898462087,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.15768654968269402,
      "audioSecondsPerSecond": 6.341694976599195,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11256179213523865,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 991.4084589690901,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 486.81149998446926,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 350.6676250253804,
      "inferenceMs": 350.1965419854969,
      "audioSeconds": 2.08,
      "realTimeFactor": 0.16836372210841197,
      "audioSecondsPerSecond": 5.939521813114139,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.024928046390414238,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 837.0080419699661,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 486.81149998446926,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1232.9260840197094,
      "inferenceMs": 1231.5869170124643,
      "audioSeconds": 7.92,
      "realTimeFactor": 0.1555033986126849,
      "audioSecondsPerSecond": 6.43072761702603,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12433581799268723,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1718.3984169969335,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 486.81149998446926,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 443.92425002297387,
      "inferenceMs": 443.3652500156313,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.15394626736653866,
      "audioSecondsPerSecond": 6.495772954462404,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 930.1767500001006,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 486.81149998446926,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 449.9752499978058,
      "inferenceMs": 449.42008302314207,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.156048639938591,
      "audioSecondsPerSecond": 6.408258350688123,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 936.2315830076113,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 486.81149998446926,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 461.68516599573195,
      "inferenceMs": 461.10804099589586,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.1601069586791305,
      "audioSecondsPerSecond": 6.245824717738188,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 947.9195409803651,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 322.9385829763487,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 502.4714579922147,
      "inferenceMs": 497.9536249884404,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.17290056423209738,
      "audioSecondsPerSecond": 5.7836711201105455,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 820.8922079647891,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 322.9385829763487,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 519.0882079768926,
      "inferenceMs": 518.4175829635933,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.1620054946761229,
      "audioSecondsPerSecond": 6.1726301444230245,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11256179213523865,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 841.356165939942,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 322.9385829763487,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 332.2456249734387,
      "inferenceMs": 331.6609169705771,
      "audioSeconds": 2.08,
      "realTimeFactor": 0.15945236392816206,
      "audioSecondsPerSecond": 6.271465504584988,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.024928046390414238,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 654.5994999469258,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 322.9385829763487,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1174.1244999575429,
      "inferenceMs": 1172.7409169543535,
      "audioSeconds": 7.92,
      "realTimeFactor": 0.14807334810029715,
      "audioSecondsPerSecond": 6.753409798788721,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12433581799268723,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1495.6794999307021,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 322.9385829763487,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 456.86966698849574,
      "inferenceMs": 456.23337500728667,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.15841436632197453,
      "audioSecondsPerSecond": 6.312558786287834,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 779.1719579836354,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 322.9385829763487,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 449.46866697864607,
      "inferenceMs": 448.8317089853808,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.15584434339770167,
      "audioSecondsPerSecond": 6.4166589444192015,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 771.7702919617295,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 322.9385829763487,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 460.0840409984812,
      "inferenceMs": 459.364790993277,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.1595016635393323,
      "audioSecondsPerSecond": 6.269527087116588,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 782.3033739696257,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 315.3604579856619,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 467.0640410040505,
      "inferenceMs": 463.8739579822868,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.16106734652162735,
      "audioSecondsPerSecond": 6.20858306538082,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 779.2344159679487,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 315.3604579856619,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 511.71262498246506,
      "inferenceMs": 510.9778749756515,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.1596805859298911,
      "audioSecondsPerSecond": 6.262502070471217,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11256179213523865,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 826.3383329613134,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 315.3604579856619,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 335.5070000397973,
      "inferenceMs": 334.97141598491,
      "audioSeconds": 2.08,
      "realTimeFactor": 0.16104394999274518,
      "audioSecondsPerSecond": 6.209485050789233,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.024928046390414238,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 650.3318739705719,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 315.3604579856619,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1230.4972090059891,
      "inferenceMs": 1228.99391700048,
      "audioSeconds": 7.92,
      "realTimeFactor": 0.15517599962127274,
      "audioSecondsPerSecond": 6.4442955253430325,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12433581799268723,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1544.354374986142,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 315.3604579856619,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 466.63291699951515,
      "inferenceMs": 466.0702080000192,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.16182993333334,
      "audioSecondsPerSecond": 6.17932652756016,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 781.4306659856811,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 315.3604579856619,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 474.0994169842452,
      "inferenceMs": 473.5445419792086,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.1644251881872252,
      "audioSecondsPerSecond": 6.081793252146593,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 788.9049999648705,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 315.3604579856619,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 200
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 445.96054195426404,
      "inferenceMs": 445.2714590006508,
      "audioSeconds": 2.88,
      "realTimeFactor": 0.15460814548633708,
      "audioSecondsPerSecond": 6.467964523178187,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1018715351819992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 760.6319169863127,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1366.971832991112,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1820.1790840248577,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 3501.9743340089917,
      "inferenceMs": 3497.3152090096846,
      "audioSeconds": 4,
      "realTimeFactor": 0.8743288022524212,
      "audioSecondsPerSecond": 1.1437344822952513,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1180303360,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 5317.494293034542,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1366.971832991112,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1820.1790840248577,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1123.5580829670653,
      "inferenceMs": 1122.7424169774167,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.35985333877481307,
      "audioSecondsPerSecond": 2.778909884245299,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12209800630807877,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1180712960,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2942.9215010022745,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1366.971832991112,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1820.1790840248577,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 758.2261249772273,
      "inferenceMs": 757.3835409712046,
      "audioSeconds": 2.24,
      "realTimeFactor": 0.3381176522192878,
      "audioSecondsPerSecond": 2.9575504071921257,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "marius",
      "rms": 0.028680138289928436,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1180712960,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2577.5626249960624,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1366.971832991112,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1820.1790840248577,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 5510.533666994888,
      "inferenceMs": 5508.318250009324,
      "audioSeconds": 8.32,
      "realTimeFactor": 0.6620574819722745,
      "audioSecondsPerSecond": 1.5104428652040787,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12237940728664398,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 7328.497334034182,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1366.971832991112,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1820.1790840248577,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1209.1275420389138,
      "inferenceMs": 1207.8969170106575,
      "audioSeconds": 4,
      "realTimeFactor": 0.3019742292526644,
      "audioSecondsPerSecond": 3.311540863850642,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3028.076001035515,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1366.971832991112,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1820.1790840248577,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1136.6822080453858,
      "inferenceMs": 1135.515125002712,
      "audioSeconds": 4,
      "realTimeFactor": 0.283878781250678,
      "audioSecondsPerSecond": 3.5226303128198726,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2955.6942090275697,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1366.971832991112,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1820.1790840248577,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1176.287374983076,
      "inferenceMs": 1175.0749169732444,
      "audioSeconds": 4,
      "realTimeFactor": 0.2937687292433111,
      "audioSecondsPerSecond": 3.4040382806427285,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2995.254000998102,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1165.0479169911705,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1653.5875420086086,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 3533.197459008079,
      "inferenceMs": 3529.9418339855038,
      "audioSeconds": 4,
      "realTimeFactor": 0.8824854584963759,
      "audioSecondsPerSecond": 1.1331631477575295,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1180303360,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 5183.529375994112,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1165.0479169911705,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1653.5875420086086,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1101.526707992889,
      "inferenceMs": 1100.6085419794545,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.352759148070338,
      "audioSecondsPerSecond": 2.834795370921483,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12209800630807877,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1180712960,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2754.196083988063,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1165.0479169911705,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1653.5875420086086,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 752.2842500475235,
      "inferenceMs": 751.3577500358224,
      "audioSeconds": 2.24,
      "realTimeFactor": 0.33542756698027787,
      "audioSecondsPerSecond": 2.9812695748372917,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "marius",
      "rms": 0.028680138289928436,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1180712960,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2404.945292044431,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1165.0479169911705,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1653.5875420086086,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 5989.0617079799995,
      "inferenceMs": 5986.887208011467,
      "audioSeconds": 8.32,
      "realTimeFactor": 0.7195777894244552,
      "audioSecondsPerSecond": 1.3897038161778685,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12237940728664398,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 7640.474750020076,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1165.0479169911705,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1653.5875420086086,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1344.4829999934882,
      "inferenceMs": 1343.3570000343025,
      "audioSeconds": 4,
      "realTimeFactor": 0.3358392500085756,
      "audioSecondsPerSecond": 2.9776150344977994,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2996.944542042911,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1165.0479169911705,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1653.5875420086086,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1351.5222500427626,
      "inferenceMs": 1350.3788330126554,
      "audioSeconds": 4,
      "realTimeFactor": 0.33759470825316384,
      "audioSecondsPerSecond": 2.9621317383034786,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3003.966375021264,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1165.0479169911705,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1653.5875420086086,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1404.324583010748,
      "inferenceMs": 1403.1149580259807,
      "audioSeconds": 4,
      "realTimeFactor": 0.3507787395064952,
      "audioSecondsPerSecond": 2.8507999128079526,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3056.7025000345893,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1117.4635000061244,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1583.8486660504714,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 3367.1557499910705,
      "inferenceMs": 3363.404875039123,
      "audioSeconds": 4,
      "realTimeFactor": 0.8408512187597807,
      "audioSecondsPerSecond": 1.1892710359330356,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1180303360,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 4947.253541089594,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1117.4635000061244,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1583.8486660504714,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1138.2633750326931,
      "inferenceMs": 1137.2586660436355,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.3645059827062934,
      "audioSecondsPerSecond": 2.7434391956352777,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12209800630807877,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1180712960,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2721.107332094107,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1117.4635000061244,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1583.8486660504714,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 759.0753329568543,
      "inferenceMs": 758.1520829699002,
      "audioSeconds": 2.24,
      "realTimeFactor": 0.3384607513258483,
      "audioSecondsPerSecond": 2.9545523257355892,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "marius",
      "rms": 0.028680138289928436,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1180712960,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2342.0007490203716,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1117.4635000061244,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1583.8486660504714,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 5955.950166971888,
      "inferenceMs": 5954.072958964389,
      "audioSeconds": 8.32,
      "realTimeFactor": 0.7156337691062967,
      "audioSecondsPerSecond": 1.3973627896973444,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12237940728664398,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 7537.92162501486,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1117.4635000061244,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1583.8486660504714,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1365.551750001032,
      "inferenceMs": 1364.599333028309,
      "audioSeconds": 4,
      "realTimeFactor": 0.34114983325707726,
      "audioSecondsPerSecond": 2.931263340956813,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2948.4479990787804,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1117.4635000061244,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1583.8486660504714,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1342.2284999978729,
      "inferenceMs": 1341.2888339953497,
      "audioSeconds": 4,
      "realTimeFactor": 0.3353222084988374,
      "audioSecondsPerSecond": 2.982206291902873,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2925.137500045821,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello. This is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1117.4635000061244,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1583.8486660504714,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 200
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1385.6362500227988,
      "inferenceMs": 1384.4335000030696,
      "audioSeconds": 4,
      "realTimeFactor": 0.3461083750007674,
      "audioSecondsPerSecond": 2.8892684263932726,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10059409588575363,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1268727808,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2968.282166053541,
      "grade": "fast"
    }
  }
]

Pocket TTS Realigned English · September 2026

63 configurations tested
Passed63
Failed0
TestedNov 15, 2026
Runtimepocket-tts 4.0.1

Quality criteria

{ "file": "research/pocket-tts-family-expansion/quality-gates.json", "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d", "description": "Frozen WER<=.35, repeated-word<=3, duration1–40sec, RMS>=.005, clipping<=.02, RTF<=1; independent GraniteCTC after TTS disposal." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedFastmacos · aarch64617 ms / 1281 ms0.234 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64637 ms / 1399 ms0.241 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64792 ms / 1399 ms0.241 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64798 ms / 1399 ms0.243 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64804 ms / 1399 ms0.245 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64840 ms / 1376 ms0.256 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64865 ms / 1281 ms0.264 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64874 ms / 1376 ms0.266 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64908 ms / 1376 ms0.277 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64737 ms / 1376 ms0.279 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64929 ms / 1281 ms0.283 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64933 ms / 1281 ms0.284 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641033 ms / 1399 ms0.300 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641063 ms / 1376 ms0.309 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641082 ms / 1281 ms0.315 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch644038 ms / 1399 ms0.510 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch644295 ms / 1281 ms0.542 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch644522 ms / 1376 ms0.571 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch641903 ms / 1376 ms0.580 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch642023 ms / 1281 ms0.617 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch642076 ms / 1399 ms0.633 RTF
FP32cpuPassedFastmacos · aarch64316 ms / 350 ms0.101 RTF
FP32cpuPassedFastmacos · aarch64863 ms / 501 ms0.107 RTF
FP32cpuPassedFastmacos · aarch64336 ms / 350 ms0.108 RTF
FP32cpuPassedFastmacos · aarch64881 ms / 350 ms0.109 RTF
FP32cpuPassedFastmacos · aarch64341 ms / 501 ms0.109 RTF
FP32cpuPassedFastmacos · aarch64341 ms / 358 ms0.109 RTF
FP32cpuPassedFastmacos · aarch64344 ms / 339 ms0.110 RTF
FP32cpuPassedFastmacos · aarch64382 ms / 350 ms0.111 RTF
FP32cpuPassedFastmacos · aarch64349 ms / 501 ms0.112 RTF
FP32cpuPassedFastmacos · aarch64350 ms / 350 ms0.112 RTF
FP32cpuPassedFastmacos · aarch64907 ms / 339 ms0.112 RTF
FP32cpuPassedFastmacos · aarch64359 ms / 350 ms0.115 RTF
FP32cpuPassedFastmacos · aarch64360 ms / 339 ms0.115 RTF
FP32cpuPassedFastmacos · aarch64306 ms / 350 ms0.116 RTF
FP32cpuPassedFastmacos · aarch64307 ms / 318 ms0.116 RTF
FP32cpuPassedFastmacos · aarch64406 ms / 501 ms0.118 RTF
FP32cpuPassedFastmacos · aarch64371 ms / 318 ms0.119 RTF
FP32cpuPassedFastmacos · aarch64966 ms / 318 ms0.120 RTF
FP32cpuPassedFastmacos · aarch64318 ms / 501 ms0.120 RTF
FP32cpuPassedFastmacos · aarch64380 ms / 501 ms0.122 RTF
FP32cpuPassedFastmacos · aarch64381 ms / 339 ms0.122 RTF
FP32cpuPassedFastmacos · aarch64382 ms / 339 ms0.122 RTF
FP32cpuPassedFastmacos · aarch64993 ms / 334 ms0.123 RTF
FP32cpuPassedFastmacos · aarch64387 ms / 318 ms0.124 RTF
FP32cpuPassedFastmacos · aarch64387 ms / 334 ms0.124 RTF
FP32cpuPassedFastmacos · aarch64329 ms / 339 ms0.125 RTF
FP32cpuPassedFastmacos · aarch64391 ms / 318 ms0.125 RTF
FP32cpuPassedFastmacos · aarch64395 ms / 334 ms0.127 RTF
FP32cpuPassedFastmacos · aarch64439 ms / 318 ms0.128 RTF
FP32cpuPassedFastmacos · aarch64401 ms / 318 ms0.129 RTF
FP32cpuPassedFastmacos · aarch64401 ms / 358 ms0.129 RTF
FP32cpuPassedFastmacos · aarch641050 ms / 358 ms0.130 RTF
FP32cpuPassedFastmacos · aarch64407 ms / 501 ms0.130 RTF
FP32cpuPassedFastmacos · aarch64417 ms / 334 ms0.134 RTF
FP32cpuPassedFastmacos · aarch64359 ms / 358 ms0.136 RTF
FP32cpuPassedFastmacos · aarch64468 ms / 334 ms0.136 RTF
FP32cpuPassedFastmacos · aarch64479 ms / 358 ms0.139 RTF
FP32cpuPassedFastmacos · aarch64480 ms / 339 ms0.140 RTF
FP32cpuPassedFastmacos · aarch64440 ms / 358 ms0.141 RTF
FP32cpuPassedFastmacos · aarch64449 ms / 334 ms0.144 RTF
FP32cpuPassedFastmacos · aarch64462 ms / 358 ms0.148 RTF
FP32cpuPassedFastmacos · aarch64429 ms / 334 ms0.163 RTF
Test setup & sample details
Quality gate
{ "file": "research/pocket-tts-family-expansion/quality-gates.json", "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d", "description": "Frozen WER<=.35, repeated-word<=3, duration1–40sec, RMS>=.005, clipping<=.02, RTF<=1; independent GraniteCTC after TTS disposal." }
Tested
2026-11-15T08:18:40.335584+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3Max;Mac15,11;96GiB", "cpu": "Apple M3Max", "gpu": "Apple M3Max", "driver": "OS-managedMetal/MPS macOS27.2 build26B118", "runtime": "PocketTTS4.0.1/Torch2.10.0; isolated app-owned CPython3.13.18 development snapshot", "timingScope": "Up to three fresh helpers per profile, four frozen fixtures and three warm repeats. Synthesis and full sidecar request timings are retained. No network transfer timing.", "concurrentUploadActivity": "The workspace's background mirror agent was active. Individual overlapping transfers were not instrumented." }
Runtime
pocket-tts 4.0.1
Artifact revision
a7056f505a4a8ca72d8b5fb1440071d6f1a35739
Fixture
{
  "file": "research/pocket-tts-family-expansion/quality-gates.json",
  "size": 2455,
  "sha1": "2ee2afd004929d50b3d7eee062620472a957398d",
  "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d",
  "url": "https://assets.localai.tools/samples/v1/files/4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d/quality-gates.json"
}
Rerun when
artifact revision change; runtime or helper/model generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 357.9021249897778,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 403.95520796300843,
      "inferenceMs": 401.34758298518136,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.12863704582858376,
      "audioSecondsPerSecond": 7.7738103635601,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 759.2497079749592,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 357.9021249897778,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 479.83912500785664,
      "inferenceMs": 479.17775000678375,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.13929585756011156,
      "audioSecondsPerSecond": 7.178964382113526,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12733793258666992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 837.0798749965616,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "A quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 357.9021249897778,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 359.20241702115163,
      "inferenceMs": 358.6112500051968,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.13583759469893816,
      "audioSecondsPerSecond": 7.361732237797177,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.029228810220956802,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 716.5133749949746,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 357.9021249897778,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1051.7018749960698,
      "inferenceMs": 1050.289208011236,
      "audioSeconds": 8.08,
      "realTimeFactor": 0.1299862881202025,
      "audioSecondsPerSecond": 7.693119131729248,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1216207891702652,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1408.1913330010138,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 357.9021249897778,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 462.38783304579556,
      "inferenceMs": 461.85825002612546,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.14803149039298893,
      "audioSecondsPerSecond": 6.755319407683016,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 819.7603750159033,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 357.9021249897778,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 440.0852500111796,
      "inferenceMs": 439.5246250205673,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.14087327725018184,
      "audioSecondsPerSecond": 7.098578378524301,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 797.4267500103451,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 357.9021249897778,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 341.8748329859227,
      "inferenceMs": 341.3000419968739,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.10939103910156216,
      "audioSecondsPerSecond": 9.141516601479285,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 699.2021669866517,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 350.14945798320696,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 362.2352500096895,
      "inferenceMs": 358.9564170106314,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.11505013365725365,
      "audioSecondsPerSecond": 8.69186300103835,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 709.1058749938384,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 350.14945798320696,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 382.5091249891557,
      "inferenceMs": 381.7938750144094,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.11098659157395623,
      "audioSecondsPerSecond": 9.010097398419944,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12733793258666992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 731.9433329976164,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "A quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 350.14945798320696,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 306.55383301200345,
      "inferenceMs": 306.02566699963063,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.11591881325743585,
      "audioSecondsPerSecond": 8.626727378403807,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.029228810220956802,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 656.1751249828376,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 350.14945798320696,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 883.0354170058854,
      "inferenceMs": 881.4198329928331,
      "audioSeconds": 8.08,
      "realTimeFactor": 0.10908661299416252,
      "audioSecondsPerSecond": 9.167027672346125,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1216207891702652,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1231.56929097604,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 350.14945798320696,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 350.155292020645,
      "inferenceMs": 349.6170419966802,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.1120567442297052,
      "audioSecondsPerSecond": 8.924050104026755,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 699.7664999798872,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 350.14945798320696,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 336.716708028689,
      "inferenceMs": 336.1649580183439,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.1077451788520333,
      "audioSecondsPerSecond": 9.281157733965083,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 686.3144160015509,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 350.14945798320696,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 316.88779103569686,
      "inferenceMs": 316.3170000188984,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.10138365385221103,
      "audioSecondsPerSecond": 9.863522984264504,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 666.4664580021054,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 339.03116697911173,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 384.26104199606925,
      "inferenceMs": 381.52220798656344,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.12228275897005239,
      "audioSecondsPerSecond": 8.177767728031919,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 720.5533749656752,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 339.03116697911173,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 480.5777500150725,
      "inferenceMs": 480.0140419974923,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.13953896569694543,
      "audioSecondsPerSecond": 7.166457017976093,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12733793258666992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 819.045208976604,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "A quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 339.03116697911173,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 329.5065000420436,
      "inferenceMs": 328.98412499343976,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.12461519886115142,
      "audioSecondsPerSecond": 8.024703319811083,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.029228810220956802,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 668.0152919725515,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 339.03116697911173,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 908.1831669900566,
      "inferenceMs": 906.8057080148719,
      "audioSeconds": 8.08,
      "realTimeFactor": 0.11222842920976138,
      "audioSecondsPerSecond": 8.910398256852927,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1216207891702652,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1245.8368749939837,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 339.03116697911173,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 381.6778750042431,
      "inferenceMs": 380.9967080014758,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.1221143294876525,
      "audioSecondsPerSecond": 8.18904713472725,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 720.0278749805875,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 339.03116697911173,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 360.6757920351811,
      "inferenceMs": 360.08666700217873,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.11541239326992908,
      "audioSecondsPerSecond": 8.664580741005672,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 699.1178339812905,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 339.03116697911173,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 344.11337500205263,
      "inferenceMs": 343.5581660014577,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.11011479679533902,
      "audioSecondsPerSecond": 9.081431643184292,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 682.5893329805695,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 500.67566602956504,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 409.960915974807,
      "inferenceMs": 406.7871660226956,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.13038050193035117,
      "audioSecondsPerSecond": 7.6698584926003495,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 907.4628320522606,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 500.67566602956504,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 406.8161670002155,
      "inferenceMs": 406.2164999777451,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.11808619185399567,
      "audioSecondsPerSecond": 8.46839062467542,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12733793258666992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 906.8921660073102,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "A quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 500.67566602956504,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 318.3831659844145,
      "inferenceMs": 317.84070801222697,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.120394207580389,
      "audioSecondsPerSecond": 8.306047442791508,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.029228810220956802,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 818.516374041792,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 500.67566602956504,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 866.8319159769453,
      "inferenceMs": 863.4206249844283,
      "audioSeconds": 8.08,
      "realTimeFactor": 0.10685898824064706,
      "audioSecondsPerSecond": 9.358127158643821,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1216207891702652,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1364.0962910139933,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 500.67566602956504,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 349.99825002159923,
      "inferenceMs": 349.3794579990208,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.11198059551250668,
      "audioSecondsPerSecond": 8.9301186104901,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 850.0551240285859,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 500.67566602956504,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 341.5369170252234,
      "inferenceMs": 340.99150000838563,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.10929214743858513,
      "audioSecondsPerSecond": 9.149788190976237,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 841.6671660379507,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 500.67566602956504,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 381.3937499653548,
      "inferenceMs": 379.91004198556766,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.12176603909793836,
      "audioSecondsPerSecond": 8.212470467202142,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 880.5857080151327,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 334.40116699784994,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 419.891000026837,
      "inferenceMs": 416.91975004505366,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.13362812501444027,
      "audioSecondsPerSecond": 7.483454548897823,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 751.3209170429036,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 334.40116699784994,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 468.64308399381116,
      "inferenceMs": 467.8253339952789,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.13599573662653458,
      "audioSecondsPerSecond": 7.353171686154805,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12733793258666992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 802.2265009931289,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "A quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 334.40116699784994,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 429.7750829719007,
      "inferenceMs": 429.2591249686666,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.1625981533972222,
      "audioSecondsPerSecond": 6.150131345938668,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.029228810220956802,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 763.6602919665165,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 334.40116699784994,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 993.8623330090195,
      "inferenceMs": 992.5472500035539,
      "audioSeconds": 8.08,
      "realTimeFactor": 0.12284000618855866,
      "audioSecondsPerSecond": 8.140670381154216,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1216207891702652,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1326.9484170014039,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 334.40116699784994,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 387.7538330270909,
      "inferenceMs": 387.07774999784306,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.12406338140956508,
      "audioSecondsPerSecond": 8.060396134929961,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 721.478916995693,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 334.40116699784994,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 395.59374999953434,
      "inferenceMs": 394.98979097697884,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.1265992919798009,
      "audioSecondsPerSecond": 7.898938330236092,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 729.3909579748288,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 334.40116699784994,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 449.21491597779095,
      "inferenceMs": 448.59141600318253,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.14377930000102004,
      "audioSecondsPerSecond": 6.95510410742649,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 782.9925830010325,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.02899995818734,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 394.19579203240573,
      "inferenceMs": 391.24629204161465,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.1253994525774406,
      "audioSecondsPerSecond": 7.974516470735378,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 709.275291999802,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.02899995818734,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 439.8352089920081,
      "inferenceMs": 439.21291700098664,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.12767817354679845,
      "audioSecondsPerSecond": 7.832192239446984,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12733793258666992,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 757.241916959174,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "A quick brown fox jumps over the lazy dog.",
      "wer": 0.1111111111111111,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.1111111111111111
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.02899995818734,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 307.5607919599861,
      "inferenceMs": 307.0552919525653,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.11630882270930504,
      "audioSecondsPerSecond": 8.597800035336418,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.029228810220956802,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 625.0842919107527,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.02899995818734,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 967.5221249926835,
      "inferenceMs": 966.1625840235502,
      "audioSeconds": 8.08,
      "realTimeFactor": 0.1195745772306374,
      "audioSecondsPerSecond": 8.362981690256648,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1216207891702652,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1284.1915839817375,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.02899995818734,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 401.78504202049226,
      "inferenceMs": 401.2448330176994,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.12860411314669853,
      "audioSecondsPerSecond": 7.775801065237326,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 719.2738329758868,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.02899995818734,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 387.59916700655594,
      "inferenceMs": 387.0159999933094,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.12404358974144532,
      "audioSecondsPerSecond": 8.061682204492676,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 705.0449999514967,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.02899995818734,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 371.8651669914834,
      "inferenceMs": 371.255292033311,
      "audioSeconds": 3.12,
      "realTimeFactor": 0.11899208077990736,
      "audioSecondsPerSecond": 8.403920609218028,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11061455309391022,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 689.2842919914983,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 909.2950830236077,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1398.961666971445,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 2079.9617919838056,
      "inferenceMs": 2076.4189999899827,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.6330545731676777,
      "audioSecondsPerSecond": 1.5796426443871991,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1172520960,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3475.380666961428,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 909.2950830236077,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1398.961666971445,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1033.3775419858284,
      "inferenceMs": 1032.5967079843394,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.3001734616233545,
      "audioSecondsPerSecond": 3.331407095723737,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12640686333179474,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1174880256,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2431.5583749557845,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 909.2950830236077,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1398.961666971445,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 637.4021249939688,
      "inferenceMs": 636.6940419538878,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.24117198558859387,
      "audioSecondsPerSecond": 4.146418571624078,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "marius",
      "rms": 0.039499178528785706,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1174880256,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2035.6557089253329,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 909.2950830236077,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1398.961666971445,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 4039.4873329787515,
      "inferenceMs": 4037.7452920074575,
      "audioSeconds": 7.92,
      "realTimeFactor": 0.5098163247484163,
      "audioSecondsPerSecond": 1.9614907398139496,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12433964759111404,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 5436.706958978903,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 909.2950830236077,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1398.961666971445,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 804.9529999843799,
      "inferenceMs": 804.0498750051484,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.24513715701376473,
      "audioSecondsPerSecond": 4.079348933396697,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2203.0115419765934,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 909.2950830236077,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1398.961666971445,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 798.8365000346676,
      "inferenceMs": 798.0160410515964,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.24329757349134037,
      "audioSecondsPerSecond": 4.110193067895898,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2196.9777080230415,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 909.2950830236077,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1398.961666971445,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 792.9831249639392,
      "inferenceMs": 791.5337909944355,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.24132127774220596,
      "audioSecondsPerSecond": 4.143853411335989,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2190.4954579658806,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 842.8326249704696,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1281.423416978214,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 2026.494542020373,
      "inferenceMs": 2023.2742080115713,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.6168518926864547,
      "audioSecondsPerSecond": 1.621134687039534,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1172520960,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3304.6976249897853,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 842.8326249704696,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1281.423416978214,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1083.2039160304703,
      "inferenceMs": 1082.0786660187878,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.3145577517496476,
      "audioSecondsPerSecond": 3.179066465339843,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12640686333179474,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1174880256,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2363.5020829970017,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 842.8326249704696,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1281.423416978214,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 618.1819590274245,
      "inferenceMs": 617.4024590291083,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.23386456781405618,
      "audioSecondsPerSecond": 4.2759790820261925,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "marius",
      "rms": 0.039499178528785706,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1174880256,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1898.8258760073222,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 842.8326249704696,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1281.423416978214,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 4297.175874991808,
      "inferenceMs": 4295.449500030372,
      "audioSeconds": 7.92,
      "realTimeFactor": 0.5423547348523198,
      "audioSecondsPerSecond": 1.8438116895435506,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12433964759111404,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 5576.872917008586,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 842.8326249704696,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1281.423416978214,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 933.5831670323387,
      "inferenceMs": 932.6493340195157,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.28434430915229136,
      "audioSecondsPerSecond": 3.516863069921375,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2214.0727509977296,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 842.8326249704696,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1281.423416978214,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 929.6469579567201,
      "inferenceMs": 928.636040946003,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.2831207441908546,
      "audioSecondsPerSecond": 3.532061922406822,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2210.059457924217,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 842.8326249704696,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1281.423416978214,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 866.1842909641564,
      "inferenceMs": 865.3820409672335,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.2638359880997663,
      "audioSecondsPerSecond": 3.7902334977207977,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2146.8054579454474,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 763.4599590091966,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1376.4615829568356,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1906.498875003308,
      "inferenceMs": 1903.2682500546798,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.5802647103825244,
      "audioSecondsPerSecond": 1.7233513982623139,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1172520960,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3279.7298330115154,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 763.4599590091966,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1376.4615829568356,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1063.8487079995684,
      "inferenceMs": 1062.9066660185345,
      "audioSeconds": 3.44,
      "realTimeFactor": 0.3089844959356205,
      "audioSecondsPerSecond": 3.236408341369848,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12640686333179474,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1174880256,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2439.36824897537,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 763.4599590091966,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1376.4615829568356,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 737.8167909919284,
      "inferenceMs": 736.8790410109796,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.2791208488677953,
      "audioSecondsPerSecond": 3.58267755367012,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "marius",
      "rms": 0.039499178528785706,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1174880256,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2113.340623967815,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 763.4599590091966,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1376.4615829568356,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 4524.0027499967255,
      "inferenceMs": 4521.956124983262,
      "audioSeconds": 7.92,
      "realTimeFactor": 0.5709540561847554,
      "audioSecondsPerSecond": 1.7514544106792533,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12433964759111404,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 5898.417707940098,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 763.4599590091966,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1376.4615829568356,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 909.0622499934398,
      "inferenceMs": 908.2342499750666,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.2769006859680081,
      "audioSecondsPerSecond": 3.6114031155399,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2284.6958329319023,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 763.4599590091966,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1376.4615829568356,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 874.5208749896847,
      "inferenceMs": 873.759541020263,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.2663901039695924,
      "audioSecondsPerSecond": 3.7538932006053307,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2250.2211239770986,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 763.4599590091966,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1376.4615829568356,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 840.7814159872942,
      "inferenceMs": 840.0114999967627,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.2561010670721838,
      "audioSecondsPerSecond": 3.904708447458922,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10703913122415543,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1259765760,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2216.4730829535984,
      "grade": "fast"
    }
  }
]

Pocket TTS Standard English · September 2026

63 configurations tested
Passed63
Failed0
TestedNov 15, 2026
Runtimepocket-tts 4.0.1

Quality criteria

{ "file": "research/pocket-tts-family-expansion/quality-gates.json", "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d", "description": "Frozen WER<=.35, repeated-word<=3, duration1–40sec, RMS>=.005, clipping<=.02, RTF<=1; independent GraniteCTC after TTS disposal." }

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedFastmacos · aarch64665 ms / 1333 ms0.238 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64791 ms / 1463 ms0.247 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64791 ms / 1297 ms0.247 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64695 ms / 1297 ms0.248 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64796 ms / 1333 ms0.249 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64807 ms / 1297 ms0.252 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64810 ms / 1333 ms0.253 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64725 ms / 1463 ms0.259 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64856 ms / 1463 ms0.268 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64873 ms / 1297 ms0.273 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64881 ms / 1333 ms0.275 RTF
FP32mpsAcceleratedPassedFastmacos · aarch64941 ms / 1463 ms0.294 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641084 ms / 1463 ms0.301 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641121 ms / 1333 ms0.311 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641266 ms / 1297 ms0.352 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch644803 ms / 1463 ms0.572 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch641841 ms / 1333 ms0.575 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch641842 ms / 1297 ms0.576 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch641949 ms / 1463 ms0.609 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch645344 ms / 1297 ms0.636 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch645472 ms / 1333 ms0.651 RTF
FP32cpuPassedFastmacos · aarch64334 ms / 340 ms0.104 RTF
FP32cpuPassedFastmacos · aarch64339 ms / 340 ms0.106 RTF
FP32cpuPassedFastmacos · aarch64866 ms / 340 ms0.106 RTF
FP32cpuPassedFastmacos · aarch64287 ms / 340 ms0.109 RTF
FP32cpuPassedFastmacos · aarch64401 ms / 340 ms0.109 RTF
FP32cpuPassedFastmacos · aarch64352 ms / 340 ms0.110 RTF
FP32cpuPassedFastmacos · aarch64353 ms / 340 ms0.110 RTF
FP32cpuPassedFastmacos · aarch64408 ms / 340 ms0.111 RTF
FP32cpuPassedFastmacos · aarch64356 ms / 340 ms0.111 RTF
FP32cpuPassedFastmacos · aarch64357 ms / 340 ms0.112 RTF
FP32cpuPassedFastmacos · aarch64359 ms / 340 ms0.112 RTF
FP32cpuPassedFastmacos · aarch64359 ms / 340 ms0.112 RTF
FP32cpuPassedFastmacos · aarch64303 ms / 340 ms0.115 RTF
FP32cpuPassedFastmacos · aarch64946 ms / 340 ms0.116 RTF
FP32cpuPassedFastmacos · aarch64375 ms / 331 ms0.117 RTF
FP32cpuPassedFastmacos · aarch64388 ms / 368 ms0.121 RTF
FP32cpuPassedFastmacos · aarch64394 ms / 331 ms0.123 RTF
FP32cpuPassedFastmacos · aarch64454 ms / 319 ms0.123 RTF
FP32cpuPassedFastmacos · aarch64396 ms / 319 ms0.124 RTF
FP32cpuPassedFastmacos · aarch64398 ms / 332 ms0.124 RTF
FP32cpuPassedFastmacos · aarch64399 ms / 319 ms0.125 RTF
FP32cpuPassedFastmacos · aarch64401 ms / 319 ms0.125 RTF
FP32cpuPassedFastmacos · aarch64402 ms / 332 ms0.126 RTF
FP32cpuPassedFastmacos · aarch64404 ms / 368 ms0.126 RTF
FP32cpuPassedFastmacos · aarch64472 ms / 332 ms0.128 RTF
FP32cpuPassedFastmacos · aarch641046 ms / 331 ms0.128 RTF
FP32cpuPassedFastmacos · aarch64414 ms / 332 ms0.129 RTF
FP32cpuPassedFastmacos · aarch64350 ms / 319 ms0.132 RTF
FP32cpuPassedFastmacos · aarch64425 ms / 331 ms0.133 RTF
FP32cpuPassedFastmacos · aarch641088 ms / 319 ms0.133 RTF
FP32cpuPassedFastmacos · aarch64427 ms / 332 ms0.134 RTF
FP32cpuPassedFastmacos · aarch64442 ms / 331 ms0.138 RTF
FP32cpuPassedFastmacos · aarch641136 ms / 368 ms0.139 RTF
FP32cpuPassedFastmacos · aarch641172 ms / 332 ms0.144 RTF
FP32cpuPassedFastmacos · aarch64381 ms / 331 ms0.144 RTF
FP32cpuPassedFastmacos · aarch64466 ms / 319 ms0.146 RTF
FP32cpuPassedFastmacos · aarch64469 ms / 368 ms0.147 RTF
FP32cpuPassedFastmacos · aarch64391 ms / 332 ms0.148 RTF
FP32cpuPassedFastmacos · aarch64550 ms / 331 ms0.149 RTF
FP32cpuPassedFastmacos · aarch64489 ms / 368 ms0.153 RTF
FP32cpuPassedFastmacos · aarch64578 ms / 368 ms0.157 RTF
FP32cpuPassedFastmacos · aarch64437 ms / 368 ms0.166 RTF
Test setup & sample details
Quality gate
{ "file": "research/pocket-tts-family-expansion/quality-gates.json", "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d", "description": "Frozen WER<=.35, repeated-word<=3, duration1–40sec, RMS>=.005, clipping<=.02, RTF<=1; independent GraniteCTC after TTS disposal." }
Tested
2026-11-15T08:16:46.042512+00:00
Environment
{ "os": "macos", "architecture": "aarch64", "osVersion": "27.2", "hardware": "Apple M3Max;Mac15,11;96GiB", "cpu": "Apple M3Max", "gpu": "Apple M3Max", "driver": "OS-managedMetal/MPS macOS27.2 build26B118", "runtime": "PocketTTS4.0.1/Torch2.10.0; isolated app-owned CPython3.13.18 development snapshot", "timingScope": "Up to three fresh helpers per profile, four frozen fixtures and three warm repeats. Synthesis and full sidecar request timings are retained. No network transfer timing.", "concurrentUploadActivity": "The workspace's background mirror agent was active. Individual overlapping transfers were not instrumented." }
Runtime
pocket-tts 4.0.1
Artifact revision
a7056f505a4a8ca72d8b5fb1440071d6f1a35739
Fixture
{
  "file": "research/pocket-tts-family-expansion/quality-gates.json",
  "size": 2455,
  "sha1": "2ee2afd004929d50b3d7eee062620472a957398d",
  "sha256": "4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d",
  "url": "https://assets.localai.tools/samples/v1/files/4db8184b255888852797ebc85e6c5fe9c77fd8484b2e32a7359e2f6b3ffd460d/quality-gates.json"
}
Rerun when
artifact revision change; runtime or helper/model generation behavior change; fixture/quality gate change; thread/provider/host behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 367.56391602102667,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 492.0209579868242,
      "inferenceMs": 488.58262499561533,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.1526820703111298,
      "audioSecondsPerSecond": 6.549557508371727,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 856.146541016642,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 367.56391602102667,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 578.556791995652,
      "inferenceMs": 577.8288339497522,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.15701870487765004,
      "audioSecondsPerSecond": 6.368667992639516,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12491369247436523,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 945.3927499707788,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 367.56391602102667,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 437.60570901213214,
      "inferenceMs": 436.96224997984245,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.16551600378024336,
      "audioSecondsPerSecond": 6.041711841519001,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.044472020119428635,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 804.5261660008691,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 367.56391602102667,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1137.0214589987881,
      "inferenceMs": 1135.7167499954812,
      "audioSeconds": 8.16,
      "realTimeFactor": 0.1391809742641521,
      "audioSecondsPerSecond": 7.184890070550132,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1261340230703354,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1503.280666016508,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 367.56391602102667,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 388.4726249962114,
      "inferenceMs": 387.83899997361004,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.12119968749175314,
      "audioSecondsPerSecond": 8.250846356910316,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 755.4029159946367,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 367.56391602102667,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 404.9753750441596,
      "inferenceMs": 404.4046670314856,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.12637645844733925,
      "audioSecondsPerSecond": 7.91286614837919,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 771.9685830525123,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 367.56391602102667,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 469.8025830439292,
      "inferenceMs": 469.23062502173707,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.14663457031929283,
      "audioSecondsPerSecond": 6.819674227042961,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 836.7945410427637,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.11654200730845,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 361.50116595672444,
      "inferenceMs": 358.7863749708049,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.11212074217837653,
      "audioSecondsPerSecond": 8.918956301672242,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 698.9029169781134,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.11654200730845,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 408.9785000542179,
      "inferenceMs": 408.35187502671033,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.11096518343117129,
      "audioSecondsPerSecond": 9.011835686463522,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12491369247436523,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 748.4684170340188,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.11654200730845,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 304.01516699930653,
      "inferenceMs": 303.3679170184769,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.11491208977972608,
      "audioSecondsPerSecond": 8.702304534856957,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.044472020119428635,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 643.4844590257853,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.11654200730845,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 947.0938749727793,
      "inferenceMs": 945.7581659662537,
      "audioSeconds": 8.16,
      "realTimeFactor": 0.11590173602527619,
      "audioSecondsPerSecond": 8.62799846054003,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1261340230703354,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1285.8747079735622,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.11654200730845,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 352.2601669537835,
      "inferenceMs": 351.6288329847157,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.10988401030772366,
      "audioSecondsPerSecond": 9.1005051344555,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 691.7453749920242,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.11654200730845,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 356.72054201131687,
      "inferenceMs": 356.11287504434586,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.11128527345135808,
      "audioSecondsPerSecond": 8.985914928241536,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 696.2294170516543,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.11654200730845,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 339.38604197464883,
      "inferenceMs": 338.77804200164974,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.10586813812551554,
      "audioSecondsPerSecond": 9.445712541146387,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 678.8945840089582,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.0602500187233,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 361.4995409734547,
      "inferenceMs": 358.6381249479018,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.11207441404621932,
      "audioSecondsPerSecond": 8.92264312519717,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 698.6983749666251,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.0602500187233,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 403.05366698885337,
      "inferenceMs": 401.30650001810864,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.10905067935274691,
      "audioSecondsPerSecond": 9.170048329229513,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12491369247436523,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 741.366750036832,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.0602500187233,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 287.10587503155693,
      "inferenceMs": 286.58224997343495,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.10855388256569506,
      "audioSecondsPerSecond": 9.212015050634566,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.044472020119428635,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 626.6424999921583,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.0602500187233,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 867.6982920151204,
      "inferenceMs": 866.478374984581,
      "audioSeconds": 8.16,
      "realTimeFactor": 0.10618607536575748,
      "audioSecondsPerSecond": 9.417430642911553,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1261340230703354,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1206.5386250033043,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.0602500187233,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 353.7027499987744,
      "inferenceMs": 353.14566700253636,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.11035802093829261,
      "audioSecondsPerSecond": 9.061416573962996,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 693.2059170212597,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.0602500187233,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 357.98550001345575,
      "inferenceMs": 357.38291702000424,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.11168216156875133,
      "audioSecondsPerSecond": 8.95398142329473,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 697.4431670387276,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 340.0602500187233,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 334.7653750097379,
      "inferenceMs": 334.1857090126723,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.1044330340664601,
      "audioSecondsPerSecond": 9.575514193752243,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 674.2459590313956,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.3294579857029,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 397.3917090333998,
      "inferenceMs": 394.15291702607647,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.1231727865706489,
      "audioSecondsPerSecond": 8.118676436912667,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 725.4823750117794,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.3294579857029,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 550.7126250304282,
      "inferenceMs": 549.8943750280887,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.14942781930111107,
      "audioSecondsPerSecond": 6.692194296062812,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12491369247436523,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 881.2238330137916,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.3294579857029,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 381.8132500164211,
      "inferenceMs": 381.2996660126373,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.14443169167145353,
      "audioSecondsPerSecond": 6.923688204627232,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.044472020119428635,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 712.6291239983402,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.3294579857029,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1048.255917034112,
      "inferenceMs": 1046.4616670506075,
      "audioSeconds": 8.16,
      "realTimeFactor": 0.12824285135424113,
      "audioSecondsPerSecond": 7.797705598713897,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1261340230703354,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1377.7911250363104,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.3294579857029,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 442.77787499595433,
      "inferenceMs": 441.8696669745259,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.13808427092953934,
      "audioSecondsPerSecond": 7.241954447587104,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 773.1991249602288,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.3294579857029,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 425.3769579809159,
      "inferenceMs": 424.62333297589794,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.1326947915549681,
      "audioSecondsPerSecond": 7.536090816237919,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 755.9527909616008,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.3294579857029,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 375.7300419965759,
      "inferenceMs": 375.1495420001447,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.11723423187504523,
      "audioSecondsPerSecond": 8.52993177850865,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 706.4789999858476,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.76716696470976,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 401.0621670167893,
      "inferenceMs": 397.9602080071345,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.12436256500222953,
      "audioSecondsPerSecond": 8.041004943747119,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 729.7273749718443,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.76716696470976,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 472.4350000033155,
      "inferenceMs": 471.6736249974929,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.12817218070584047,
      "audioSecondsPerSecond": 7.802005041133179,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12491369247436523,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 803.4407919622026,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.76716696470976,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 391.61404100013897,
      "inferenceMs": 390.9526250208728,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.14808811553820939,
      "audioSecondsPerSecond": 6.752736344612219,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.044472020119428635,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 722.7197919855826,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.76716696470976,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1173.6592500237748,
      "inferenceMs": 1172.241541033145,
      "audioSeconds": 8.16,
      "realTimeFactor": 0.14365705159719916,
      "audioSecondsPerSecond": 6.961022719607986,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1261340230703354,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1504.0087079978548,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.76716696470976,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 402.6769580086693,
      "inferenceMs": 401.98274998692796,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.125619609370915,
      "audioSecondsPerSecond": 7.96054059559536,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 733.7499169516377,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.76716696470976,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 414.87325000343844,
      "inferenceMs": 414.2886250047013,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.12946519531396916,
      "audioSecondsPerSecond": 7.724083662600408,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 746.0557919694111,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 331.76716696470976,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 427.8793329722248,
      "inferenceMs": 427.24504094803706,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.13351407529626158,
      "audioSecondsPerSecond": 7.4898470275964995,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 759.0122079127468,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.5510409530252,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 403.9479999919422,
      "inferenceMs": 400.910250027664,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.125284453133645,
      "audioSecondsPerSecond": 7.981836333142369,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 719.4612909806892,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.5510409530252,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 454.3551249662414,
      "inferenceMs": 453.562582959421,
      "audioSeconds": 3.68,
      "realTimeFactor": 0.12325070189114701,
      "audioSecondsPerSecond": 8.113544058217077,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12491369247436523,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 772.1136239124462,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.5510409530252,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 350.4849169985391,
      "inferenceMs": 349.7366670053452,
      "audioSeconds": 2.64,
      "realTimeFactor": 0.13247601022929745,
      "audioSecondsPerSecond": 7.548536510641738,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "marius",
      "rms": 0.044472020119428635,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 668.2877079583704,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.5510409530252,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1089.640500023961,
      "inferenceMs": 1088.3673329954036,
      "audioSeconds": 8.16,
      "realTimeFactor": 0.13337834963178966,
      "audioSecondsPerSecond": 7.49746868783911,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.1261340230703354,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1406.9183739484288,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.5510409530252,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 466.8097500107251,
      "inferenceMs": 466.1114169866778,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.1456598178083368,
      "audioSecondsPerSecond": 6.865311346989515,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 784.662457939703,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.5510409530252,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 399.2012920207344,
      "inferenceMs": 398.6423750175163,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.12457574219297385,
      "audioSecondsPerSecond": 8.027244970782126,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 717.1934159705415,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 0,
      "metalScheduling": "not applicable",
      "gpuWarmupMs": 0,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 318.5510409530252,
      "provider": "PyTorch CPU",
      "placement": {
        "cpu/torch.float32": 210
      },
      "acceleration": "cpu",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 396.3209579815157,
      "inferenceMs": 395.7262079929933,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.1236644399978104,
      "audioSecondsPerSecond": 8.086398968189288,
      "activationPlacement": {
        "flowOutput": [
          "cpu"
        ]
      },
      "waveformDevice": "cpu",
      "seed": 42,
      "voice": "alba",
      "rms": 0.11645898222923279,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": null,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 714.2772489460185,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1016.4255420095287,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1463.3879999746569,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1952.3768750368617,
      "inferenceMs": 1948.8565000356175,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.6090176562611305,
      "audioSecondsPerSecond": 1.6419885199046296,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1171734528,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3412.2445000102744,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1016.4255420095287,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1463.3879999746569,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1085.2790419594385,
      "inferenceMs": 1084.3557919724844,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.30120994221457903,
      "audioSecondsPerSecond": 3.3199435338944085,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.13206949830055237,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1176551424,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2547.7437919471413,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1016.4255420095287,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1463.3879999746569,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 725.94987496268,
      "inferenceMs": 725.0914580072276,
      "audioSeconds": 2.8,
      "realTimeFactor": 0.2589612350025813,
      "audioSecondsPerSecond": 3.861581830925513,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "marius",
      "rms": 0.0325266569852829,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1176551424,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2188.4794579818845,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1016.4255420095287,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1463.3879999746569,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 4804.906249977648,
      "inferenceMs": 4803.15004196018,
      "audioSeconds": 8.4,
      "realTimeFactor": 0.571803576423831,
      "audioSecondsPerSecond": 1.74885230038992,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12636812031269073,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 6266.538041934837,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1016.4255420095287,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1463.3879999746569,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 942.9136660182849,
      "inferenceMs": 941.0458750207908,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.29407683594399714,
      "audioSecondsPerSecond": 3.400471841959141,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2404.4338749954477,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1016.4255420095287,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1463.3879999746569,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 791.5613750228658,
      "inferenceMs": 790.7672909786925,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.2471147784308414,
      "audioSecondsPerSecond": 4.04670253373723,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2254.1552909533493,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 1016.4255420095287,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1463.3879999746569,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 857.1355830063112,
      "inferenceMs": 856.3467910280451,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.2676083721962641,
      "audioSecondsPerSecond": 3.736803866758696,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2319.734791002702,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 889.6802090457641,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1332.683167012874,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1845.003040973097,
      "inferenceMs": 1841.2355000036769,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.575386093751149,
      "audioSecondsPerSecond": 1.7379634489958562,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1171734528,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3173.918667016551,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 889.6802090457641,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1332.683167012874,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1121.7542919912376,
      "inferenceMs": 1120.811207976658,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.31133644666018273,
      "audioSecondsPerSecond": 3.211959315163249,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.13206949830055237,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1176551424,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2453.494374989532,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 889.6802090457641,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1332.683167012874,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 666.0069160279818,
      "inferenceMs": 665.2249160106294,
      "audioSeconds": 2.8,
      "realTimeFactor": 0.23758032714665336,
      "audioSecondsPerSecond": 4.209102714900805,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "marius",
      "rms": 0.0325266569852829,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1176551424,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1997.9080830235034,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 889.6802090457641,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1332.683167012874,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 5473.766041977797,
      "inferenceMs": 5472.040749969892,
      "audioSeconds": 8.4,
      "realTimeFactor": 0.6514334226154633,
      "audioSecondsPerSecond": 1.5350762875890898,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12636812031269073,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 6804.723916982766,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 889.6802090457641,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1332.683167012874,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 796.761334000621,
      "inferenceMs": 796.0289170150645,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.24875903656720766,
      "audioSecondsPerSecond": 4.0199544659750615,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2128.7120840279385,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 889.6802090457641,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1332.683167012874,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 882.0339170051739,
      "inferenceMs": 880.975459003821,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.27530483093869407,
      "audioSecondsPerSecond": 3.6323372771569122,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2213.658626016695,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 889.6802090457641,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1332.683167012874,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 810.9728749841452,
      "inferenceMs": 809.949874994345,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.2531093359357328,
      "audioSecondsPerSecond": 3.950861774035513,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2142.633042007219,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 862.2577919741161,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1296.8209579703398,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1845.1472090091556,
      "inferenceMs": 1841.9541670009494,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.5756106771877967,
      "audioSecondsPerSecond": 1.7372853555906915,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1171734528,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 3138.775124971289,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The meeting starts at 9.30. Please bring three notebooks.",
      "wer": 0.2,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0.2
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 862.2577919741161,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1296.8209579703398,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 1267.2294999938458,
      "inferenceMs": 1266.3071670103818,
      "audioSeconds": 3.6,
      "realTimeFactor": 0.35175199083621717,
      "audioSecondsPerSecond": 2.842912125735829,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.13206949830055237,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1176551424,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2563.1281249807216,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "The quick brown fox jumps over the lazy dog.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 862.2577919741161,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1296.8209579703398,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 695.6491669989191,
      "inferenceMs": 694.8663329822011,
      "audioSeconds": 2.8,
      "realTimeFactor": 0.24816654749364325,
      "audioSecondsPerSecond": 4.029551968625484,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "marius",
      "rms": 0.0325266569852829,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1176551424,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 1991.687290952541,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Welcome to the local audio library. You can listen to this message without sending your text to a remote service. Please review the generated speech before sharing it.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 862.2577919741161,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1296.8209579703398,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 5345.918457955122,
      "inferenceMs": 5343.88391696848,
      "audioSeconds": 8.4,
      "realTimeFactor": 0.6361766567819619,
      "audioSecondsPerSecond": 1.5718904322242873,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.12636812031269073,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 6640.70487493882,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 862.2577919741161,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1296.8209579703398,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 808.0845839576796,
      "inferenceMs": 807.3197919875383,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.25228743499610573,
      "audioSecondsPerSecond": 3.963732874827618,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2104.140749957878,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 862.2577919741161,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1296.8209579703398,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 874.274207977578,
      "inferenceMs": 873.4738330240361,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.2729605728200113,
      "audioSecondsPerSecond": 3.6635327573824905,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2170.294790994376,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "text": "Hello, this is a local speech synthesis test.",
      "wer": 0,
      "maximumRepeatedWords": 1,
      "passed": true,
      "wordErrorRate": 0
    },
    "runtime": {
      "metalDecoderFramesPerCall": 1,
      "metalScheduling": "serialized FlowLM step/Mimi single-frame decoder",
      "gpuWarmupMs": 862.2577919741161,
      "gpuWarmupScope": "Real short preset synthesis included in measured loadMs, never hidden or counted as a passing task case",
      "engine": "pocket-tts",
      "version": "4.0.1",
      "torchVersion": "2.10.0",
      "loadMs": 1296.8209579703398,
      "provider": "Metal/MPS",
      "placement": {
        "mps:0/torch.float32": 210
      },
      "acceleration": "partial",
      "accelerationDetails": "FlowLM/Mimi neural parameters and returned waveform on selected device; text/token processing and WAV encoding CPU",
      "voiceCloning": false,
      "temperature": 0.3,
      "sampleRate": 24000,
      "streaming": false,
      "streamChunkCount": 0,
      "firstChunkMs": null,
      "durationMs": 792.2329169814475,
      "inferenceMs": 791.4027919759974,
      "audioSeconds": 3.2,
      "realTimeFactor": 0.24731337249249918,
      "audioSecondsPerSecond": 4.043453008309647,
      "activationPlacement": {
        "flowOutput": [
          "mps:0"
        ]
      },
      "waveformDevice": "mps:0",
      "seed": 42,
      "voice": "alba",
      "rms": 0.10643884539604187,
      "clippedFraction": 0,
      "driverAllocatedMemoryBytes": 1270628352,
      "memoryMeasurementScope": "Post-inference Metal driver allocation, not peak VRAM",
      "artifactRevision": "a7056f505a4a8ca72d8b5fb1440071d6f1a35739",
      "fallback": false,
      "totalMs": 2088.223749946337,
      "grade": "fast"
    }
  }
]
Release validation still required
  • Packaged app acceptance and archive/source recovery tracked in the September investigation.

Something not quite right?

Tell us what happened with Pocket TTS September2026. 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.