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.

Qwen3-TTS

Generate local speech with nine preset voices using Qwen3-TTS 0.6B CustomVoice. Supply text and choose a voice; returns complete 24 kHz mono WAV audio.

Desktop inferencePrivate · Runs on your computer
About this model

Generate local speech with nine preset voices using Qwen3-TTS 0.6B CustomVoice. Supply text and choose a voice; returns complete 24 kHz mono WAV audio.

License
Apache-2.0
Model released
Jan 21, 2026
Source repository created
Jan 21, 2026
Model weights · FP32
2.5 GB

Best for

  • Local preset-voice narration and accessibility prototypes
  • Reviewed synthetic speech production

Keep in mind

  • Pronunciation, prosody or words may be wrong; review generated speech. Synthetic audio is not an authentic recording or proof of identity.
  • Short English Aiden phrases passed independent ASR gates. Other offered voices/languages and long-form or subjective quality remain unverified.
  • Style instructions are not supported by the pinned0.6B implementation. Reference cloning and VoiceDesign use separate checkpoints.
  • Pinned qwen-tts0.1.1 generate_custom_voice returns complete audio; its non_streaming_mode flag does not expose actual incremental audio generation.
  • Seeded outputs can differ across devices and warm runs. Incomplete EOS, empty or nonfinite audio fails visibly.
  • GPU neural execution uses Metal; CPU text preparation/WAV encoding remain. Metal FP16 produces nonfinite sampling; Metal FP32 omitted speech and failed intelligibility. Neither is advertised. Other Mac chips/OS, Windows and other GPU providers remain unverified.
  • Downloads use a private app-owned runtime; no system Python or global packages.

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/Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice
Release verification
Verified
Release evidence
Exact original CustomVoice checkpoint weights were publicly uploaded January21,2026 at commit84120d3482174bce7756cd4bed7a78a57286cbc6; the series announcement followed January22. Repository creation and later README updates are distinct.
Source updated
Jan 29, 2026
Native model files
https://huggingface.co/Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice
Files updated
Jan 29, 2026
Artifact revision
85e237c12c027371202489a0ec509ded67b5e4b5

Desktop app

Sample result

Qwen3-TTS Desktop — 0.6B CustomVoice

Run Qwen3-TTS 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 / FP32Tested16.0 GiBRAM estimate
macOS Apple SiliconMPS / BF16Tested16.0 GiBRAM estimate
Windows Intel / AMDCPU / FP32Unverified16.0 GiBRAM estimate
macOS Intel / AMDCPU / FP32Unverified16.0 GiBRAM estimate
If your device falls short

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

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

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

Benchmark evidence

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

Qwen3-TTS Desktop — 0.6B CustomVoice

42 configurations tested
Passed27
Failed15
TestedNov 4, 2026
Runtimeqwen-tts 0.1.1

Quality criteria

No explicit quality gate recorded. Review per-case evidence; full certification is pending.

PrecisionDeviceResultPlatformInference / loadTask metric
BF16mpsAcceleratedPassedModeratemacos · aarch646744 ms / 0 ms2.107 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch647471 ms / 0 ms2.224 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch647516 ms / 0 ms2.237 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch647524 ms / 0 ms2.239 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch647563 ms / 0 ms2.251 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch647670 ms / 0 ms2.283 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch647811 ms / 0 ms2.325 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch6412886 ms / 993 ms2.334 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch648609 ms / 0 ms2.339 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch6413695 ms / 1058 ms2.481 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch6413717 ms / 1063 ms2.485 RTF
FP32cpuPassedModeratemacos · aarch644717 ms / 0 ms1.404 RTF
FP32cpuPassedModeratemacos · aarch643950 ms / 0 ms1.411 RTF
FP32cpuPassedModeratemacos · aarch643962 ms / 0 ms1.415 RTF
FP32cpuPassedModeratemacos · aarch643981 ms / 0 ms1.422 RTF
FP32cpuPassedModeratemacos · aarch644015 ms / 0 ms1.434 RTF
FP32cpuPassedModeratemacos · aarch644018 ms / 0 ms1.435 RTF
FP32cpuPassedModeratemacos · aarch644036 ms / 0 ms1.441 RTF
FP32cpuPassedModeratemacos · aarch644040 ms / 0 ms1.443 RTF
FP32cpuPassedModeratemacos · aarch644065 ms / 628 ms1.452 RTF
FP32cpuPassedModeratemacos · aarch644301 ms / 0 ms1.453 RTF
FP32cpuPassedModeratemacos · aarch644088 ms / 621 ms1.460 RTF
FP32cpuPassedModeratemacos · aarch644321 ms / 0 ms1.460 RTF
FP32cpuPassedModeratemacos · aarch644925 ms / 0 ms1.466 RTF
FP32cpuPassedModeratemacos · aarch644136 ms / 0 ms1.477 RTF
FP32cpuPassedModeratemacos · aarch644168 ms / 651 ms1.489 RTF
FP32cpuPassedModeratemacos · aarch644438 ms / 944 ms1.585 RTF
FP16mpsFailed—Not recorded— / ——
FP32mpsFailed—macos · aarch647210 ms / 1158 ms2.504 RTF
BF16mpsFailed—macos · aarch6412040 ms / 0 ms2.062 RTF
BF16mpsFailed—macos · aarch6411604 ms / 0 ms1.987 RTF
BF16mpsFailed—macos · aarch6411943 ms / 0 ms2.045 RTF
BF16mpsFailed—macos · aarch6411773 ms / 0 ms2.016 RTF
BF16mpsFailed—macos · aarch6412315 ms / 0 ms2.109 RTF
BF16mpsFailed—macos · aarch6411970 ms / 0 ms2.050 RTF
BF16mpsFailed—macos · aarch6422091 ms / 1044 ms2.321 RTF
FP32mpsFailed—macos · aarch6412091 ms / 1154 ms3.977 RTF
FP32mpsFailed—macos · aarch6412926 ms / 0 ms3.438 RTF
FP32mpsFailed—macos · aarch6419258 ms / 0 ms3.439 RTF
FP32mpsFailed—macos · aarch647018 ms / 0 ms3.509 RTF
FP32mpsFailed—macos · aarch646735 ms / 0 ms3.367 RTF
FP32mpsFailed—macos · aarch647607 ms / 0 ms2.113 RTF
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-04T11:42:29.466947+00:00
Environment
{ "hardware": "Apple M3 Max", "os": "macOS-27.2-arm64-arm-64bit-Mach-O", "runtime": "qwen-tts0.1.1/Transformers4.57.3/Torch2.10.0", "interpreterScope": "Existing app-owned pinned private Qwen probe venv; normal per-model provisioning tested separately" }
Runtime
qwen-tts 0.1.1
Artifact revision
85e237c12c027371202489a0ec509ded67b5e4b5
Fixture
{
  "file": "model-samples/desktop/qwen3-tts/morning.txt",
  "size": 39,
  "sha1": "e1c3512efd9e8669d9d3027e2b20cdea5e875d6e",
  "sha256": "1335d38acab68b7e9bb398ef2ceaf8249a48ec10a0788514a7d91e14f821d0d5",
  "url": "https://assets.localai.tools/samples/v1/files/1335d38acab68b7e9bb398ef2ceaf8249a48ec10a0788514a7d91e14f821d0d5/morning.txt"
}
Rerun when
artifact revision/checksum; qwen-tts/Transformers/Torch versions, preprocessing or sampling; OS/device/backend behavior; fixture text, voice/language selections or independent ASR/quality/performance gates
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 944.2220409982838,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4443.348375003552,
      "inferenceMs": 4438.48004200845,
      "realTimeFactor": 1.5851714435744466,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 5387.570416001836,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.6308465901612967,
      "peakSampledHelperRssBytes": 5713543168,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Please save this file before closing the window.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.608071+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4925.707500020508,
      "inferenceMs": 4924.814541009255,
      "realTimeFactor": 1.4657186133956115,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.17639891803264618,
      "peak": 0.7897162437438965,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 4925.707500020508,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.6822591941322986,
      "peakSampledHelperRssBytes": 5826412544,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "The blue box contains three small books.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.690041+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4302.010625018738,
      "inferenceMs": 4301.217416999862,
      "realTimeFactor": 1.453113992229683,
      "audioSeconds": 2.96,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 38,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 8,
      "maxNewTokens": 512,
      "rms": 0.10349362343549728,
      "peak": 0.7155910134315491,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 4302.010625018738,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.6881772561184845,
      "peakSampledHelperRssBytes": 5831639040,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4136.532708012965,
      "inferenceMs": 4135.797833994729,
      "realTimeFactor": 1.4770706549981178,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 4136.532708012965,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.677015684128715,
      "peakSampledHelperRssBytes": 5831639040,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4036.3774170109536,
      "inferenceMs": 4035.6358750141226,
      "realTimeFactor": 1.441298526790758,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 4036.3774170109536,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.6938187900785775,
      "peakSampledHelperRssBytes": 5831639040,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 650.8436670119409,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4172.994749998907,
      "inferenceMs": 4168.060249998234,
      "realTimeFactor": 1.4885929464279408,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4823.8384170108475,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.6717753180274172,
      "peakSampledHelperRssBytes": 5464571904,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Please save this file before closing the window.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.608071+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4718.140917015262,
      "inferenceMs": 4717.260334000457,
      "realTimeFactor": 1.4039465279763266,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.17639891803264618,
      "peak": 0.7897162437438965,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4718.140917015262,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.7122778396990787,
      "peakSampledHelperRssBytes": 5517508608,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "The blue box contains three small books.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.690041+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4322.318291990086,
      "inferenceMs": 4321.41854200745,
      "realTimeFactor": 1.4599386966241386,
      "audioSeconds": 2.96,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 38,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 8,
      "maxNewTokens": 512,
      "rms": 0.10349362343549728,
      "peak": 0.7155910134315491,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4322.318291990086,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.6849602673813161,
      "peakSampledHelperRssBytes": 5835358208,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 3950.286041974323,
      "inferenceMs": 3949.5460839825682,
      "realTimeFactor": 1.4105521728509174,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 3950.286041974323,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.7089422279070077,
      "peakSampledHelperRssBytes": 5835456512,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 3962.3884170141537,
      "inferenceMs": 3961.624874995323,
      "realTimeFactor": 1.414866026784044,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 3962.3884170141537,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.7067806994228107,
      "peakSampledHelperRssBytes": 5835456512,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 628.0749580182601,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4070.3229999926407,
      "inferenceMs": 4065.393708006013,
      "realTimeFactor": 1.451926324287862,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4698.397958010901,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.688740181421036,
      "peakSampledHelperRssBytes": 5455855616,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4015.81158299814,
      "inferenceMs": 4014.9961249844637,
      "realTimeFactor": 1.4339271874944515,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4015.81158299814,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.6973854800447247,
      "peakSampledHelperRssBytes": 5465243648,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4041.0304999968503,
      "inferenceMs": 4040.260458015837,
      "realTimeFactor": 1.4429501635770847,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4041.0304999968503,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.6930246277674568,
      "peakSampledHelperRssBytes": 5474025472,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 621.1847079975996,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4092.62562499498,
      "inferenceMs": 4087.531333992956,
      "realTimeFactor": 1.4598326192831987,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4713.81033299258,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.6850100393639759,
      "peakSampledHelperRssBytes": 4972150784,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4019.125457998598,
      "inferenceMs": 4018.3474159857724,
      "realTimeFactor": 1.435124077137776,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4019.125457998598,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.6968038624189267,
      "peakSampledHelperRssBytes": 5470502912,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.519176+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 3981.704042002093,
      "inferenceMs": 3980.974874983076,
      "realTimeFactor": 1.4217767410653843,
      "audioSeconds": 2.8,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 36,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08865950256586075,
      "peak": 0.5263413786888123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 3981.704042002093,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.7033453081041873,
      "peakSampledHelperRssBytes": 5477777408,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.788412+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 993.3586249826476,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 12891.231083020102,
      "inferenceMs": 12886.217416991713,
      "realTimeFactor": 2.3344596769912522,
      "audioSeconds": 5.52,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 70,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.16219741106033325,
      "peak": 0.8671875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3190013952,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 13884.58970800275,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.42836464894045256,
      "peakSampledHelperRssBytes": 1429176320,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Please save this file before closing the window.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.875722+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 6744.507124996744,
      "inferenceMs": 6743.6192499881145,
      "realTimeFactor": 2.1073810156212858,
      "audioSeconds": 3.2,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 41,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.14652307331562042,
      "peak": 0.80859375,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3191144448,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 6744.507124996744,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.47452263856765636,
      "peakSampledHelperRssBytes": 1503707136,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "The blue box contains three small books.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.957953+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 8609.927791025257,
      "inferenceMs": 8609.022667020326,
      "realTimeFactor": 2.3394083334294367,
      "audioSeconds": 3.68,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 47,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 8,
      "maxNewTokens": 512,
      "rms": 0.12260419875383377,
      "peak": 0.79296875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3192586240,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 8609.927791025257,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.42745850979083166,
      "peakSampledHelperRssBytes": 1600159744,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:10.043362+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 7670.93483399367,
      "inferenceMs": 7670.082042022841,
      "realTimeFactor": 2.282762512506798,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.13558422029018402,
      "peak": 0.796875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3193733120,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 7670.93483399367,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.4380657184096903,
      "peakSampledHelperRssBytes": 1674969088,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:10.043362+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 7811.599208012922,
      "inferenceMs": 7810.7803750026505,
      "realTimeFactor": 2.3246370163698367,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.13558422029018402,
      "peak": 0.796875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3193733120,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 7811.599208012922,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.4301746866104732,
      "peakSampledHelperRssBytes": 1688010752,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.788412+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1057.6847080083098,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 13700.652041006833,
      "inferenceMs": 13695.359959005145,
      "realTimeFactor": 2.4810434708342655,
      "audioSeconds": 5.52,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 70,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.16219741106033325,
      "peak": 0.8671875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3190013952,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 14758.336749015143,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.40305621878674464,
      "peakSampledHelperRssBytes": 1434189824,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:10.043362+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 7472.342249995563,
      "inferenceMs": 7471.438000007765,
      "realTimeFactor": 2.2236422619070733,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.13558422029018402,
      "peak": 0.796875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3191160832,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 7472.342249995563,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.44971262560119046,
      "peakSampledHelperRssBytes": 1510096896,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:10.043362+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 7516.54079198488,
      "inferenceMs": 7515.75429100194,
      "realTimeFactor": 2.236831634226768,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.13558422029018402,
      "peak": 0.796875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3191160832,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 7516.54079198488,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.44706091629720796,
      "peakSampledHelperRssBytes": 1523662848,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:09.788412+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1063.2034169975668,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 13721.999542001868,
      "inferenceMs": 13716.71870801947,
      "realTimeFactor": 2.4849128094238173,
      "audioSeconds": 5.52,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 70,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.16219741106033325,
      "peak": 0.8671875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3190013952,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 14785.202958999434,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.40242860683384396,
      "peakSampledHelperRssBytes": 1428832256,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:10.043362+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 7563.585791998776,
      "inferenceMs": 7562.734624982113,
      "realTimeFactor": 2.2508138764827716,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.13558422029018402,
      "peak": 0.796875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3191160832,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 7563.585791998776,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.4442837368510662,
      "peakSampledHelperRssBytes": 1487126528,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:46:10.043362+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 7524.681457987754,
      "inferenceMs": 7523.823499999708,
      "realTimeFactor": 2.2392331845237226,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.13558422029018402,
      "peak": 0.796875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3191160832,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 7524.681457987754,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.4465814489135917,
      "peakSampledHelperRssBytes": 1519681536,
      "resourceMeasurementScope": "200ms sampled helper RSS, process lifetime high-water sample; not a validated minimum or GPU peak."
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "{'ok': False, 'error': 'probability tensor contains either `inf`, `nan` or element < 0', 'code': 'runtime_error', 'traceback': 'Traceback (most recent call last):\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/crates/desktopai/src/models/qwen_tts/runtime.py\", line 53, in <module>\\n    with contextlib.redirect_stdout(sys.stderr):data=init(r) if r[\\'type\\']==\\'init\\' else {\\'alive\\':True,\\'pid\\':os.getpid()} if r[\\'type\\']==\\'health\\' else generate(r)\\n                                                                                                                                                    ~~~~~~~~^^^\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/crates/desktopai/src/models/qwen_tts/runtime.py\", line 42, in generate\\n    with torch.inference_mode():wavs,sr=model.generate_custom_voice(text=text,language=language,speaker=speaker,non_streaming_mode=True,max_new_tokens=maximum,do_sample=True,subtalker_dosample=True,temperature=0.9,top_k=50,top_p=1.0,repetition_penalty=1.05)\\n                                        ~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/torch/utils/_contextlib.py\", line 124, in decorate_context\\n    return func(*args, **kwargs)\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/qwen_tts/inference/qwen3_tts_model.py\", line 829, in generate_custom_voice\\n    talker_codes_list, _ = self.model.generate(\\n                           ~~~~~~~~~~~~~~~~~~~^\\n        input_ids=input_ids,\\n        ^^^^^^^^^^^^^^^^^^^^\\n    ...<4 lines>...\\n        **gen_kwargs,\\n        ^^^^^^^^^^^^^\\n    )\\n    ^\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/torch/utils/_contextlib.py\", line 124, in decorate_context\\n    return func(*args, **kwargs)\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/qwen_tts/core/models/modeling_qwen3_tts.py\", line 2272, in generate\\n    talker_result = self.talker.generate(\\n        inputs_embeds=talker_input_embeds,\\n    ...<3 lines>...\\n        **talker_kwargs,\\n    )\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/crates/desktopai/src/models/qwen_tts/runtime.py\", line 25, in observed\\n    result=original(*args,**kwargs);sequences=result.sequences\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/torch/utils/_contextlib.py\", line 124, in decorate_context\\n    return func(*args, **kwargs)\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/transformers/generation/utils.py\", line 2564, in generate\\n    result = decoding_method(\\n        self,\\n    ...<5 lines>...\\n        **model_kwargs,\\n    )\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/transformers/generation/utils.py\", line 2787, in _sample\\n    outputs = model_forward(**model_inputs, return_dict=True)\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/torch/nn/modules/module.py\", line 1776, in _wrapped_call_impl\\n    return self._call_impl(*args, **kwargs)\\n           ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/torch/nn/modules/module.py\", line 1787, in _call_impl\\n    return forward_call(*args, **kwargs)\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/transformers/utils/generic.py\", line 918, in wrapper\\n    output = func(self, *args, **kwargs)\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/qwen_tts/core/models/modeling_qwen3_tts.py\", line 1671, in forward\\n    predictor_result = self.code_predictor.generate(\\n        inputs_embeds=torch.cat((past_hidden, last_id_hidden), dim=1),\\n    ...<6 lines>...\\n        return_dict_in_generate=True,\\n    )\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/torch/utils/_contextlib.py\", line 124, in decorate_context\\n    return func(*args, **kwargs)\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/transformers/generation/utils.py\", line 2564, in generate\\n    result = decoding_method(\\n        self,\\n    ...<5 lines>...\\n        **model_kwargs,\\n    )\\n  File \"/Users/pengzhang/Desktop/hubters-webai/webai-js/localai-desktop/.cache/qwen-voice-probe/venv/lib/python3.13/site-packages/transformers/generation/utils.py\", line 2829, in _sample\\n    next_tokens = torch.multinomial(probs, num_samples=1).squeeze(1)\\n                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^\\nRuntimeError: probability tensor contains either `inf`, `nan` or element < 0\\n'}"
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "wordErrorRate": 0.2857142857142857,
      "transcript": "The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:28:35.895740+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1158.2479999924544,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "durationMs": 7214.914374984801,
      "inferenceMs": 7210.173625004245,
      "realTimeFactor": 2.5035325086820297,
      "audioSeconds": 2.88,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 37,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.06079854816198349,
      "peak": 0.41497504711151123,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 5727633408,
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 8373.162374977255,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.3994355961155241,
      "peakSampledHelperRssBytes": 3560357888
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "wordErrorRate": 0.2857142857142857,
      "transcript": "Uh huh. Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:33:51.166387+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "durationMs": 12041.508540976793,
      "inferenceMs": 12040.462000004482,
      "realTimeFactor": 2.061722945206247,
      "audioSeconds": 5.84,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 74,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08339852839708328,
      "peak": 0.546875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3185917952,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 12041.508540976793,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.4850312222236842,
      "peakSampledHelperRssBytes": 1698496512
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "wordErrorRate": 0.2857142857142857,
      "transcript": "Uh huh. Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:33:51.166387+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "durationMs": 11604.526707989862,
      "inferenceMs": 11603.5159160092,
      "realTimeFactor": 1.986903410275548,
      "audioSeconds": 5.84,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 74,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08339852839708328,
      "peak": 0.546875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3185917952,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 11604.526707989862,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.5032957288353126,
      "peakSampledHelperRssBytes": 1728823296
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "wordErrorRate": 0.2857142857142857,
      "transcript": "Uh huh. Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:33:51.166387+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "durationMs": 11944.045832991833,
      "inferenceMs": 11942.955917009385,
      "realTimeFactor": 2.0450266981180456,
      "audioSeconds": 5.84,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 74,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08339852839708328,
      "peak": 0.546875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3183296512,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 11944.045832991833,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.48899117107872436,
      "peakSampledHelperRssBytes": 1526218752
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "wordErrorRate": 0.2857142857142857,
      "transcript": "Uh huh. Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:33:51.166387+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "durationMs": 11773.998374992516,
      "inferenceMs": 11773.008999996819,
      "realTimeFactor": 2.015926198629592,
      "audioSeconds": 5.84,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 74,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08339852839708328,
      "peak": 0.546875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3183296512,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 11773.998374992516,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.4960499053386928,
      "peakSampledHelperRssBytes": 1569865728
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "wordErrorRate": 0.2857142857142857,
      "transcript": "Uh huh. Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:33:51.166387+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "durationMs": 12315.832499996759,
      "inferenceMs": 12314.80029100203,
      "realTimeFactor": 2.108698679966101,
      "audioSeconds": 5.84,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 74,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08339852839708328,
      "peak": 0.546875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3183296512,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 12315.832499996759,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.4742261232012891,
      "peakSampledHelperRssBytes": 1535098880
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "wordErrorRate": 0.2857142857142857,
      "transcript": "Uh huh. Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:33:51.166387+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "durationMs": 11971.489499992458,
      "inferenceMs": 11970.487959013553,
      "realTimeFactor": 2.049741088872184,
      "audioSeconds": 5.84,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 74,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08339852839708328,
      "peak": 0.546875,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3183296512,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 11971.489499992458,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.4878664946655403,
      "peakSampledHelperRssBytes": 1556889600
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "wordErrorRate": 0.2857142857142857,
      "transcript": "D but good morning, the train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:36:41.292676+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1043.6494160094298,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "durationMs": 22097.087709000334,
      "inferenceMs": 22091.48187498795,
      "realTimeFactor": 2.320533810397894,
      "audioSeconds": 9.52,
      "seed": 0,
      "eosReached": true,
      "generatedTokens": 120,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.036851946264505386,
      "peak": 0.34765625,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 3269312512,
      "precision": "bf16",
      "cpuThreads": 4,
      "totalMs": 23140.737125009764,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.43093532855206856,
      "peakSampledHelperRssBytes": 1756823552
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": true,
      "wordErrorRate": 0,
      "transcript": "Good morning. The train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:38:21.225979+00:00",
      "structuralPassed": true,
      "performancePassed": false
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1153.8627919799183,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "eager",
      "durationMs": 12096.490583993727,
      "inferenceMs": 12091.324209002778,
      "realTimeFactor": 3.9774092792772295,
      "audioSeconds": 3.04,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 39,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08722323179244995,
      "peak": 0.6686762571334839,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 5736005632,
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 13250.353375973646,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "slow",
      "audioSecondsPerSecond": 0.2514199394088302,
      "peakSampledHelperRssBytes": 3055747072
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.25,
      "transcript": "One bait. Please save this file before closing the window.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:38:21.317741+00:00",
      "structuralPassed": true,
      "performancePassed": false
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "eager",
      "durationMs": 12926.560332998633,
      "inferenceMs": 12925.634916988201,
      "realTimeFactor": 3.4376688609011175,
      "audioSeconds": 3.76,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 48,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.08904044330120087,
      "peak": 0.5335569381713867,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 5746065408,
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 12926.560332998633,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "slow",
      "audioSecondsPerSecond": 0.2908948012339588,
      "peakSampledHelperRssBytes": 3055747072
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "wordErrorRate": 0.5714285714285714,
      "transcript": "When Brom's the Blue Box contains three small books, uh.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:38:21.417511+00:00",
      "structuralPassed": true,
      "performancePassed": false
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "eager",
      "durationMs": 19259.12524998421,
      "inferenceMs": 19258.111083006952,
      "realTimeFactor": 3.438948407679813,
      "audioSeconds": 5.6,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 71,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 8,
      "maxNewTokens": 512,
      "rms": 0.06005222350358963,
      "peak": 0.4930441975593567,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 5757321216,
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 19259.12524998421,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "slow",
      "audioSecondsPerSecond": 0.290786566546568,
      "peakSampledHelperRssBytes": 3055747072
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.14285714285714285,
      "transcript": "Morning, the train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:38:21.493349+00:00",
      "structuralPassed": true,
      "performancePassed": false
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "eager",
      "durationMs": 7018.734832992777,
      "inferenceMs": 7017.955791001441,
      "realTimeFactor": 3.5089778955007205,
      "audioSeconds": 2,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 26,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.04995442181825638,
      "peak": 0.389910489320755,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 5758353408,
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 7018.734832992777,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "slow",
      "audioSecondsPerSecond": 0.28498327141992524,
      "peakSampledHelperRssBytes": 3055747072
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": true,
      "wordErrorRate": 0.14285714285714285,
      "transcript": "Morning, the train leaves at nine.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:38:21.493349+00:00",
      "structuralPassed": true,
      "performancePassed": false
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "eager",
      "durationMs": 6735.496207984397,
      "inferenceMs": 6734.785541018937,
      "realTimeFactor": 3.3673927705094684,
      "audioSeconds": 2,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 26,
      "generationPlacement": "mps:0",
      "speaker": "Ryan",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.04995442181825638,
      "peak": 0.389910489320755,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 5758353408,
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 6735.496207984397,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "slow",
      "audioSecondsPerSecond": 0.2969656550782181,
      "peakSampledHelperRssBytes": 3055747072
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "wordErrorRate": 0.5714285714285714,
      "transcript": "Once runes known to the blue box contains three small books.",
      "repetitive": false,
      "maximumWordErrorRate": 0.25,
      "evaluator": "Independent Qwen3-ASR0.6B MLXQ4, revision313d850181767edf09f00a9c289becca70e58cd0 / mlx-audio0.6.3 / mlx0.32.4",
      "evaluatedAt": "2026-11-04T11:39:51.874648+00:00",
      "structuralPassed": true,
      "performancePassed": true
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 0,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 7607.696624996606,
      "inferenceMs": 7606.8151670042425,
      "realTimeFactor": 2.113004213056734,
      "audioSeconds": 3.6,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 46,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 8,
      "maxNewTokens": 512,
      "rms": 0.12580694258213043,
      "peak": 0.8521938323974609,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "driverAllocatedMemoryBytes": 5679300608,
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 7607.696624996606,
      "artifactRevision": "85e237c12c027371202489a0ec509ded67b5e4b5",
      "fallback": false,
      "grade": "moderate",
      "audioSecondsPerSecond": 0.47325982306176784,
      "peakSampledHelperRssBytes": 2958917632
    }
  }
]

Qwen3-TTS 1.7B · CustomVoice — Controlled Speech

40 configurations tested
Passed36
Failed4
TestedNov 6, 2026
Runtimemlx-audio 0.6.3

Quality criteria

{ "maximumWordErrorRate": 0.25, "minimumSpeakerCosine": null, "minimumRms": 0.005, "maximumClippedFraction": 0.02, "eosRequired": true, "maximumRealTimeFactor": 3, "maximumRequestMs": 45000 }

PrecisionDeviceResultPlatformInference / loadTask metric
BF16mpsAcceleratedPassedFastmacos · aarch641197 ms / 310 ms0.468 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641124 ms / 310 ms0.468 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641425 ms / 312 ms0.469 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641204 ms / 312 ms0.470 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641129 ms / 310 ms0.470 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641430 ms / 310 ms0.471 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641243 ms / 312 ms0.471 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641135 ms / 312 ms0.473 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641248 ms / 310 ms0.473 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641135 ms / 312 ms0.473 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641148 ms / 312 ms0.478 RTF
BF16mpsAcceleratedPassedFastmacos · aarch641163 ms / 310 ms0.485 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch641573 ms / 332 ms0.517 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch641245 ms / 332 ms0.519 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch641247 ms / 332 ms0.520 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch641453 ms / 332 ms0.550 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch641540 ms / 332 ms0.601 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch641859 ms / 332 ms0.774 RTF
FP32cpuPassedModeratemacos · aarch646469 ms / 974 ms2.128 RTF
FP32cpuPassedModeratemacos · aarch644944 ms / 925 ms2.131 RTF
FP32cpuPassedModeratemacos · aarch646482 ms / 925 ms2.132 RTF
FP32cpuPassedModeratemacos · aarch644955 ms / 914 ms2.136 RTF
FP32cpuPassedModeratemacos · aarch646497 ms / 914 ms2.137 RTF
FP32cpuPassedModeratemacos · aarch645301 ms / 974 ms2.137 RTF
FP32cpuPassedModeratemacos · aarch644971 ms / 914 ms2.142 RTF
FP32cpuPassedModeratemacos · aarch644971 ms / 974 ms2.142 RTF
FP32cpuPassedModeratemacos · aarch644973 ms / 925 ms2.144 RTF
FP32cpuPassedModeratemacos · aarch644981 ms / 914 ms2.147 RTF
FP32cpuPassedModeratemacos · aarch644641 ms / 925 ms2.148 RTF
FP32cpuPassedModeratemacos · aarch644659 ms / 914 ms2.157 RTF
FP32cpuPassedModeratemacos · aarch645363 ms / 914 ms2.163 RTF
FP32cpuPassedModeratemacos · aarch645373 ms / 925 ms2.167 RTF
FP32cpuPassedModeratemacos · aarch645073 ms / 974 ms2.187 RTF
FP32cpuPassedModeratemacos · aarch645097 ms / 974 ms2.197 RTF
FP32cpuPassedModeratemacos · aarch644750 ms / 974 ms2.199 RTF
FP32cpuPassedModeratemacos · aarch645114 ms / 925 ms2.204 RTF
BF16mpsFailed—macos · aarch6416008 ms / 1627 ms5.885 RTF
FP32mpsFailed—macos · aarch649529 ms / 1574 ms4.254 RTF
FP16mpsFailed—macos · aarch6410812 ms / 1412 ms4.360 RTF
FP32mpsFailed—macos · aarch6414303 ms / 1623 ms4.832 RTF
Test setup & sample details
Quality gate
{ "maximumWordErrorRate": 0.25, "minimumSpeakerCosine": null, "minimumRms": 0.005, "maximumClippedFraction": 0.02, "eosRequired": true, "maximumRealTimeFactor": 3, "maximumRequestMs": 45000 }
Tested
2026-11-06T14:34:13.031046+00:00
Environment
{ "os": "macos", "arch": "aarch64", "hardware": "Apple M3 Max", "osVersion": "27.2", "runtime": "App-owned CPython3.13.18; CPU qwen-tts0.1.1/Transformers4.57.3/Torch2.10.0; Metal mlx-audio0.6.3/MLX0.32.4/Transformers5.22.1; original identical checkpoint files" }
Runtime
mlx-audio 0.6.3
Artifact revision
0c0e3051f131929182e2c023b9537f8b1c68adfe
Fixture
{
  "file": "model-samples/desktop/qwen3-tts-family/customvoice-morning-input.json",
  "url": "https://assets.localai.tools/samples/v1/files/44decdf28f478b082c7088eebe2beec02d39eaa0bc76e8dd10d23112e0a439fc/customvoice-morning-input.json",
  "size": 234,
  "sha1": "4e3c80615fb763a1a3fac3c165dbe529e820a05c",
  "sha256": "44decdf28f478b082c7088eebe2beec02d39eaa0bc76e8dd10d23112e0a439fc"
}
Rerun when
artifact revision/checksum; qwen-tts/Transformers/Torch versions, preprocessing or sampling; OS/device/backend behavior; fixture text, voice/language selections or independent ASR/quality/performance gates
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.11956550180912018,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:37.448742+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.41802436113357544
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 973.9282919999823,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5101.3998329999595,
      "inferenceMs": 5096.5938750014175,
      "realTimeFactor": 2.1968077047419903,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.11956549435853958,
      "peak": 0.6288480758666992,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 6083.301667000342,
      "audioSecondsPerSecond": 0.4552059781297278,
      "grade": "moderate",
      "residentMemoryBytes": 9379840000,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.10222581028938293,
      "audioSeconds": 2.48,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:37.558922+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3600354790687561
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 973.9282919999823,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5301.646542000526,
      "inferenceMs": 5300.851415999205,
      "realTimeFactor": 2.1374400870964534,
      "audioSeconds": 2.48,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 32,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.10222581773996353,
      "peak": 0.5643832087516785,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 6283.967458999541,
      "audioSecondsPerSecond": 0.4678493708604588,
      "grade": "moderate",
      "residentMemoryBytes": 9675259904,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "The blue box contains three small books.",
      "wordErrorRate": 0,
      "rms": 0.12879793345928192,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:37.665807+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4279877543449402
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 973.9282919999823,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5073.718000001463,
      "inferenceMs": 5072.972957999809,
      "realTimeFactor": 2.186626274999918,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 8,
      "maxNewTokens": 512,
      "rms": 0.12879793345928192,
      "peak": 0.7126675844192505,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 6055.527874999825,
      "audioSecondsPerSecond": 0.4573255207957462,
      "grade": "moderate",
      "residentMemoryBytes": 9675161600,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.1544686257839203,
      "audioSeconds": 3.04,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:37.779011+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.38538694381713867
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 973.9282919999823,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 6469.744042000457,
      "inferenceMs": 6468.830499999967,
      "realTimeFactor": 2.127904769736831,
      "audioSeconds": 3.04,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 39,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.1544685661792755,
      "peak": 0.6981966495513916,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "Speak slowly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 7453.931417001513,
      "audioSecondsPerSecond": 0.4699458426063282,
      "grade": "moderate",
      "residentMemoryBytes": 9986244608,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.1282392144203186,
      "audioSeconds": 2.16,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:37.888374+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.38728201389312744
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 973.9282919999823,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4750.806417001513,
      "inferenceMs": 4750.051792001614,
      "realTimeFactor": 2.199098051852599,
      "audioSeconds": 2.16,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 28,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.1282392293214798,
      "peak": 0.8515697717666626,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "Speak quickly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 5732.169332999547,
      "audioSecondsPerSecond": 0.4547318839001127,
      "grade": "moderate",
      "residentMemoryBytes": 9986981888,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.11956550180912018,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:37.999289+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.41802436113357544
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 973.9282919999823,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4971.3132079996285,
      "inferenceMs": 4970.540416999938,
      "realTimeFactor": 2.1424743176723875,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.11956549435853958,
      "peak": 0.6288480758666992,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 5952.824374999182,
      "audioSecondsPerSecond": 0.4667500523816842,
      "grade": "moderate",
      "residentMemoryBytes": 9987162112,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.11956550180912018,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:38.108187+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.41802436113357544
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 925.3592499990191,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4948.787875000562,
      "inferenceMs": 4943.862791998981,
      "realTimeFactor": 2.1309753413788712,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.11956549435853958,
      "peak": 0.6288480758666992,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 5881.7525829999795,
      "audioSecondsPerSecond": 0.4692686867755771,
      "grade": "moderate",
      "residentMemoryBytes": 9383198720,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.10222581028938293,
      "audioSeconds": 2.48,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:38.216653+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3600354790687561
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 925.3592499990191,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5374.14762499975,
      "inferenceMs": 5373.307290999946,
      "realTimeFactor": 2.166656165725785,
      "audioSeconds": 2.48,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 32,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.10222581773996353,
      "peak": 0.5643832087516785,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 6307.126666999466,
      "audioSecondsPerSecond": 0.46154069843611795,
      "grade": "moderate",
      "residentMemoryBytes": 9678225408,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "The blue box contains three small books.",
      "wordErrorRate": 0,
      "rms": 0.12879793345928192,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:38.326336+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4279877543449402
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 925.3592499990191,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4973.953500000789,
      "inferenceMs": 4973.1902920011635,
      "realTimeFactor": 2.1436165051729152,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 8,
      "maxNewTokens": 512,
      "rms": 0.12879793345928192,
      "peak": 0.7126675844192505,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 5907.225582999672,
      "audioSecondsPerSecond": 0.4665013530110577,
      "grade": "moderate",
      "residentMemoryBytes": 9686499328,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.1544686257839203,
      "audioSeconds": 3.04,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:38.440991+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.38538694381713867
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 925.3592499990191,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 6483.118916999956,
      "inferenceMs": 6482.2800000001735,
      "realTimeFactor": 2.132328947368478,
      "audioSeconds": 3.04,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 39,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.1544685661792755,
      "peak": 0.6981966495513916,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "Speak slowly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 7418.684416998076,
      "audioSecondsPerSecond": 0.4689707942267102,
      "grade": "moderate",
      "residentMemoryBytes": 9981935616,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.1282392144203186,
      "audioSeconds": 2.16,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:38.550631+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.38728201389312744
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 925.3592499990191,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4641.575083000134,
      "inferenceMs": 4640.665417000491,
      "realTimeFactor": 2.1484562115743016,
      "audioSeconds": 2.16,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 28,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.1282392293214798,
      "peak": 0.8515697717666626,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "Speak quickly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 5574.149332998786,
      "audioSecondsPerSecond": 0.4654504916659393,
      "grade": "moderate",
      "residentMemoryBytes": 9982869504,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.11956550180912018,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:38.657722+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.41802436113357544
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 925.3592499990191,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5114.679457999955,
      "inferenceMs": 5113.961332999679,
      "realTimeFactor": 2.204293678017103,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.11956549435853958,
      "peak": 0.6288480758666992,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 6047.808999999688,
      "audioSecondsPerSecond": 0.4536600589897627,
      "grade": "moderate",
      "residentMemoryBytes": 9982869504,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.11956550180912018,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:38.769815+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.41802436113357544
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 913.5980409992044,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4975.732834000155,
      "inferenceMs": 4970.532625000487,
      "realTimeFactor": 2.142470959051934,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.11956549435853958,
      "peak": 0.6288480758666992,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 5897.185373998582,
      "audioSecondsPerSecond": 0.46675078407714354,
      "grade": "moderate",
      "residentMemoryBytes": 9382969344,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.10222581028938293,
      "audioSeconds": 2.48,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:38.878033+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3600354790687561
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 913.5980409992044,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5364.141542000652,
      "inferenceMs": 5363.327791999836,
      "realTimeFactor": 2.162632174193482,
      "audioSeconds": 2.48,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 32,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.10222581773996353,
      "peak": 0.5643832087516785,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 6285.884748998797,
      "audioSecondsPerSecond": 0.46239948333929387,
      "grade": "moderate",
      "residentMemoryBytes": 9684058112,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "The blue box contains three small books.",
      "wordErrorRate": 0,
      "rms": 0.12879793345928192,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:38.981939+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4279877543449402
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 913.5980409992044,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4956.122416999278,
      "inferenceMs": 4955.36416700088,
      "realTimeFactor": 2.135932830603828,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 8,
      "maxNewTokens": 512,
      "rms": 0.12879793345928192,
      "peak": 0.7126675844192505,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 5877.000290998694,
      "audioSecondsPerSecond": 0.4681795165428026,
      "grade": "moderate",
      "residentMemoryBytes": 9684975616,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.1544686257839203,
      "audioSeconds": 3.04,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:39.095017+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.38538694381713867
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 913.5980409992044,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 6498.445458000788,
      "inferenceMs": 6497.485375000906,
      "realTimeFactor": 2.1373307154608243,
      "audioSeconds": 3.04,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 39,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.1544685661792755,
      "peak": 0.6981966495513916,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "Speak slowly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 7422.254415998395,
      "audioSecondsPerSecond": 0.46787331168091717,
      "grade": "moderate",
      "residentMemoryBytes": 9979428864,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.1282392144203186,
      "audioSeconds": 2.16,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:39.202797+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.38728201389312744
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 913.5980409992044,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4660.242541998741,
      "inferenceMs": 4659.402291999868,
      "realTimeFactor": 2.1571306907406793,
      "audioSeconds": 2.16,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 28,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.1282392293214798,
      "peak": 0.8515697717666626,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "Speak quickly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 5581.313874999978,
      "audioSecondsPerSecond": 0.46357877354970867,
      "grade": "moderate",
      "residentMemoryBytes": 9979953152,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.11956550180912018,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:39.309547+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.41802436113357544
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 913.5980409992044,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 4981.399374999455,
      "inferenceMs": 4980.639249999513,
      "realTimeFactor": 2.1468272629308247,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.11956549435853958,
      "peak": 0.6288480758666992,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 5902.578831999563,
      "audioSecondsPerSecond": 0.46580366164849757,
      "grade": "moderate",
      "residentMemoryBytes": 9979985920,
      "supportsStreaming": false
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.1649700105190277,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:39.415478+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3360149562358856
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 332.1784170002502,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1859.382333999747,
      "inferenceMs": 1858.7289170009171,
      "realTimeFactor": 0.7744703820837155,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "maxNewTokens": 512,
      "promptTokens": 9,
      "rms": 0.16496999561786652,
      "peak": 0.8428813219070435,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 469.610707999891,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583183708,
      "peakMemoryBytes": 5237400109,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2203.878166999857,
      "audioSecondsPerSecond": 1.2912049616532735,
      "grade": "moderate",
      "residentMemoryBytes": 4893425664
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.09189192205667496,
      "audioSeconds": 2.56,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:39.525977+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3798825144767761
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 332.1784170002502,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1540.2527499991265,
      "inferenceMs": 1539.71808300048,
      "realTimeFactor": 0.6014523761720625,
      "audioSeconds": 2.56,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 33,
      "maxNewTokens": 512,
      "promptTokens": 9,
      "rms": 0.09189192205667496,
      "peak": 0.525258481502533,
      "clippedFraction": 0,
      "chunks": 6,
      "firstAudioMs": 332.0858329989278,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583200100,
      "peakMemoryBytes": 5237400109,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1881.582250000065,
      "audioSecondsPerSecond": 1.6626420305535907,
      "grade": "moderate",
      "residentMemoryBytes": 4895375360
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "The blue box contains three small books.",
      "wordErrorRate": 0,
      "rms": 0.10940630733966827,
      "audioSeconds": 2.64,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:39.741611+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.46773505210876465
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 332.1784170002502,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1452.8746249998221,
      "inferenceMs": 1452.6106250013981,
      "realTimeFactor": 0.5502312973490144,
      "audioSeconds": 2.64,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 34,
      "maxNewTokens": 512,
      "promptTokens": 8,
      "rms": 0.10940626263618469,
      "peak": 0.7453674674034119,
      "clippedFraction": 0,
      "chunks": 6,
      "firstAudioMs": 302.7335000006133,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583216488,
      "peakMemoryBytes": 5237400109,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1794.3940830009524,
      "audioSecondsPerSecond": 1.8174175202645644,
      "grade": "moderate",
      "residentMemoryBytes": 4895571968
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.10392040014266968,
      "audioSeconds": 3.04,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:39.851080+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3559342622756958
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 332.1784170002502,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1573.1981249991804,
      "inferenceMs": 1572.9596660003153,
      "realTimeFactor": 0.5174209427632617,
      "audioSeconds": 3.04,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 39,
      "maxNewTokens": 512,
      "promptTokens": 14,
      "rms": 0.10392042994499207,
      "peak": 0.6027777194976807,
      "clippedFraction": 0,
      "chunks": 7,
      "firstAudioMs": 304.58741599977657,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583297404,
      "peakMemoryBytes": 5237400109,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "Speak slowly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1916.3867919996846,
      "audioSecondsPerSecond": 1.9326623979685633,
      "grade": "moderate",
      "residentMemoryBytes": 4895768576
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.10883376747369766,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:39.957360+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.37157905101776123
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 332.1784170002502,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1244.8362079994695,
      "inferenceMs": 1244.5335419997718,
      "realTimeFactor": 0.5185556424999049,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "maxNewTokens": 512,
      "promptTokens": 14,
      "rms": 0.10883378237485886,
      "peak": 0.6034771800041199,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 267.50766699842643,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583232860,
      "peakMemoryBytes": 5237400109,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "Speak quickly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1585.9787080007663,
      "audioSecondsPerSecond": 1.9284333599748331,
      "grade": "moderate",
      "residentMemoryBytes": 4895457280
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.1649700105190277,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:40.068681+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3360149562358856
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 332.1784170002502,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1247.24229200001,
      "inferenceMs": 1247.0442920002824,
      "realTimeFactor": 0.519601788333451,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "maxNewTokens": 512,
      "promptTokens": 9,
      "rms": 0.16496999561786652,
      "peak": 0.8428813219070435,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 275.8472499990603,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583183708,
      "peakMemoryBytes": 5237400109,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1588.2446670002537,
      "audioSecondsPerSecond": 1.9245507279860563,
      "grade": "moderate",
      "residentMemoryBytes": 4896718848
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.1649700105190277,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:40.176323+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3360149562358856
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 309.9599999986822,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1163.4656250007538,
      "inferenceMs": 1162.8724170004716,
      "realTimeFactor": 0.4845301737501965,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "maxNewTokens": 512,
      "promptTokens": 9,
      "rms": 0.16496999561786652,
      "peak": 0.8428813219070435,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 265.27337500010617,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583183708,
      "peakMemoryBytes": 5237259287,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1484.5112499988318,
      "audioSecondsPerSecond": 2.0638549551210366,
      "grade": "fast",
      "residentMemoryBytes": 4893425664
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.09189192205667496,
      "audioSeconds": 2.56,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:40.283484+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3798825144767761
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 309.9599999986822,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1197.3001249989466,
      "inferenceMs": 1196.9590839999,
      "realTimeFactor": 0.46756214218746095,
      "audioSeconds": 2.56,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 33,
      "maxNewTokens": 512,
      "promptTokens": 9,
      "rms": 0.09189192205667496,
      "peak": 0.525258481502533,
      "clippedFraction": 0,
      "chunks": 6,
      "firstAudioMs": 231.13362500043877,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583200100,
      "peakMemoryBytes": 5237259287,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1516.6909999989002,
      "audioSecondsPerSecond": 2.138753140537771,
      "grade": "fast",
      "residentMemoryBytes": 4894457856
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "The blue box contains three small books.",
      "wordErrorRate": 0,
      "rms": 0.10940630733966827,
      "audioSeconds": 2.64,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:40.394653+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.46773505210876465
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 309.9599999986822,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1248.6763329998212,
      "inferenceMs": 1248.4351660004904,
      "realTimeFactor": 0.4728921083335191,
      "audioSeconds": 2.64,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 34,
      "maxNewTokens": 512,
      "promptTokens": 8,
      "rms": 0.10940626263618469,
      "peak": 0.7453674674034119,
      "clippedFraction": 0,
      "chunks": 6,
      "firstAudioMs": 241.4589159998286,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583216488,
      "peakMemoryBytes": 5237259287,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1567.6933339982497,
      "audioSecondsPerSecond": 2.1146472575404553,
      "grade": "fast",
      "residentMemoryBytes": 4895768576
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.10392040014266968,
      "audioSeconds": 3.04,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:40.506835+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3559342622756958
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 309.9599999986822,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1430.6194170003437,
      "inferenceMs": 1430.3737080008432,
      "realTimeFactor": 0.4705176671055405,
      "audioSeconds": 3.04,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 39,
      "maxNewTokens": 512,
      "promptTokens": 14,
      "rms": 0.10392042994499207,
      "peak": 0.6027777194976807,
      "clippedFraction": 0,
      "chunks": 7,
      "firstAudioMs": 241.00308299966855,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583281020,
      "peakMemoryBytes": 5237259287,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "Speak slowly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1751.0307499996998,
      "audioSecondsPerSecond": 2.1253187072690576,
      "grade": "fast",
      "residentMemoryBytes": 4896342016
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.10883376747369766,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:40.613470+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.37157905101776123
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 309.9599999986822,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1129.1795420002018,
      "inferenceMs": 1128.9415419996658,
      "realTimeFactor": 0.4703923091665274,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "maxNewTokens": 512,
      "promptTokens": 14,
      "rms": 0.10883378237485886,
      "peak": 0.6034771800041199,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 239.95954199926928,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583232860,
      "peakMemoryBytes": 5237259287,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "Speak quickly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1447.2084589979204,
      "audioSecondsPerSecond": 2.1258850974240353,
      "grade": "fast",
      "residentMemoryBytes": 4896129024
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.1649700105190277,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:40.721913+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3360149562358856
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 309.9599999986822,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1124.5804159989348,
      "inferenceMs": 1124.3720000002213,
      "realTimeFactor": 0.46848833333342554,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "maxNewTokens": 512,
      "promptTokens": 9,
      "rms": 0.16496999561786652,
      "peak": 0.8428813219070435,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 231.59229200064146,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583183708,
      "peakMemoryBytes": 5237377425,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1442.943624997497,
      "audioSecondsPerSecond": 2.1345248725506574,
      "grade": "fast",
      "residentMemoryBytes": 4896129024
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.1649700105190277,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:40.830665+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3360149562358856
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 311.9535839996388,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1148.6714579987165,
      "inferenceMs": 1148.0654589995538,
      "realTimeFactor": 0.4783606079164808,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "maxNewTokens": 512,
      "promptTokens": 9,
      "rms": 0.16496999561786652,
      "peak": 0.8428813219070435,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 261.75858399983554,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583183708,
      "peakMemoryBytes": 5237325363,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1470.9914590002882,
      "audioSecondsPerSecond": 2.0904731356445527,
      "grade": "fast",
      "residentMemoryBytes": 4895670272
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.09189192205667496,
      "audioSeconds": 2.56,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:40.939071+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3798825144767761
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 311.9535839996388,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1203.9577909999934,
      "inferenceMs": 1203.647667000041,
      "realTimeFactor": 0.47017486992189106,
      "audioSeconds": 2.56,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 33,
      "maxNewTokens": 512,
      "promptTokens": 9,
      "rms": 0.09189192205667496,
      "peak": 0.525258481502533,
      "clippedFraction": 0,
      "chunks": 6,
      "firstAudioMs": 234.7428339999169,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583200100,
      "peakMemoryBytes": 5237325363,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1525.4904169996735,
      "audioSecondsPerSecond": 2.1268682440772033,
      "grade": "fast",
      "residentMemoryBytes": 4897406976
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "The blue box contains three small books.",
      "wordErrorRate": 0,
      "rms": 0.10940630733966827,
      "audioSeconds": 2.64,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:41.044019+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.46773505210876465
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 311.9535839996388,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1243.3321669996076,
      "inferenceMs": 1243.0634579995967,
      "realTimeFactor": 0.4708573704543927,
      "audioSeconds": 2.64,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 34,
      "maxNewTokens": 512,
      "promptTokens": 8,
      "rms": 0.10940626263618469,
      "peak": 0.7453674674034119,
      "clippedFraction": 0,
      "chunks": 6,
      "firstAudioMs": 229.58366599959845,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583216488,
      "peakMemoryBytes": 5237325363,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1564.4410839995544,
      "audioSecondsPerSecond": 2.1237853811972136,
      "grade": "fast",
      "residentMemoryBytes": 4898455552
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.10392040014266968,
      "audioSeconds": 3.04,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:41.157607+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3559342622756958
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 311.9535839996388,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1425.6837500015536,
      "inferenceMs": 1425.4284170001483,
      "realTimeFactor": 0.46889092664478565,
      "audioSeconds": 3.04,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 39,
      "maxNewTokens": 512,
      "promptTokens": 14,
      "rms": 0.10392042994499207,
      "peak": 0.6027777194976807,
      "clippedFraction": 0,
      "chunks": 7,
      "firstAudioMs": 240.68145899946103,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583297404,
      "peakMemoryBytes": 5237325363,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "Speak slowly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1748.2165839992376,
      "audioSecondsPerSecond": 2.1326921532810186,
      "grade": "fast",
      "residentMemoryBytes": 4898340864
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.10883376747369766,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:41.266771+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.37157905101776123
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 311.9535839996388,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1135.5378329990344,
      "inferenceMs": 1135.2879579990258,
      "realTimeFactor": 0.47303664916626076,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "maxNewTokens": 512,
      "promptTokens": 14,
      "rms": 0.10883378237485886,
      "peak": 0.6034771800041199,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 236.07204099971568,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583232860,
      "peakMemoryBytes": 5237325363,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "Speak quickly and clearly.",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1455.4647500008286,
      "audioSecondsPerSecond": 2.114001107023157,
      "grade": "fast",
      "residentMemoryBytes": 4898357248
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.1649700105190277,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:41.373601+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3360149562358856
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 311.9535839996388,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1134.85770799889,
      "inferenceMs": 1134.6415420011908,
      "realTimeFactor": 0.47276730916716286,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "maxNewTokens": 512,
      "promptTokens": 9,
      "rms": 0.16496999561786652,
      "peak": 0.8428813219070435,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 242.55087500023365,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583183708,
      "peakMemoryBytes": 5237325363,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "custom_voice",
      "instruct": "",
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 1454.9524590001965,
      "audioSecondsPerSecond": 2.1152054734106334,
      "grade": "fast",
      "residentMemoryBytes": 4898979840
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.14905516803264618,
      "audioSeconds": 2.72,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:28:57.674754+00:00",
      "structuralPassed": true,
      "performancePassed": false,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.33941158652305603
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1626.769709000655,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.bfloat16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.bfloat16",
      "attentionImplementation": "sdpa",
      "durationMs": 16012.851916999352,
      "inferenceMs": 16007.618292000188,
      "realTimeFactor": 5.885153783823598,
      "audioSeconds": 2.72,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 35,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.14905516803264618,
      "peak": 0.83203125,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "driverAllocatedMemoryBytes": 5175934976,
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 17648.293876001844,
      "audioSecondsPerSecond": 0.16991909417026274,
      "grade": "slow",
      "residentMemoryBytes": 1181237248,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.12418601661920547,
      "audioSeconds": 2.24,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:28:57.910852+00:00",
      "structuralPassed": true,
      "performancePassed": false,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.3386712074279785
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1574.0860829992016,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 9536.391583000295,
      "inferenceMs": 9528.927832998306,
      "realTimeFactor": 4.253985639731386,
      "audioSeconds": 2.24,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 29,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.12418593466281891,
      "peak": 0.8332152366638184,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "driverAllocatedMemoryBytes": 9745317888,
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 11121.686000000409,
      "audioSecondsPerSecond": 0.23507366613093317,
      "grade": "slow",
      "residentMemoryBytes": 1192116224,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.13684940338134766,
      "audioSeconds": 2.48,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:28:58.115460+00:00",
      "structuralPassed": true,
      "performancePassed": false,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.38086578249931335
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1412.1661669996683,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float16",
      "attentionImplementation": "sdpa",
      "durationMs": 10818.091000001004,
      "inferenceMs": 10811.847584000134,
      "realTimeFactor": 4.359615961290377,
      "audioSeconds": 2.48,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 32,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.13684940338134766,
      "peak": 0.83349609375,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "driverAllocatedMemoryBytes": 5175754752,
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 12239.815082999485,
      "audioSecondsPerSecond": 0.22937800230092192,
      "grade": "slow",
      "residentMemoryBytes": 1178779648,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.10639488697052002,
      "audioSeconds": 2.96,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:29:03.288988+00:00",
      "structuralPassed": true,
      "performancePassed": false,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.40386563539505005
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1622.8039170000557,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "eager",
      "durationMs": 14309.05412500033,
      "inferenceMs": 14303.209292000247,
      "realTimeFactor": 4.832165301351435,
      "audioSeconds": 2.96,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 38,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.10639490932226181,
      "peak": 0.8179582953453064,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "custom_voice",
      "instruct": "",
      "driverAllocatedMemoryBytes": 9779249152,
      "artifactRevision": "0c0e3051f131929182e2c023b9537f8b1c68adfe",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 15942.233916999612,
      "audioSecondsPerSecond": 0.20694656280080592,
      "grade": "slow",
      "residentMemoryBytes": 1334312960,
      "supportsStreaming": false
    }
  }
]

Qwen3-TTS 1.7B · VoiceDesign — Described Voices

28 configurations tested
Passed24
Failed4
TestedNov 6, 2026
Runtimemlx-audio 0.6.3

Quality criteria

{ "maximumWordErrorRate": 0.25, "minimumSpeakerCosine": null, "minimumRms": 0.005, "maximumClippedFraction": 0.02, "eosRequired": true, "maximumRealTimeFactor": 3, "maximumRequestMs": 45000 }

PrecisionDeviceResultPlatformInference / loadTask metric
BF16mpsAcceleratedPassedModeratemacos · aarch641962 ms / 324 ms0.584 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch642115 ms / 321 ms0.630 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch642121 ms / 324 ms0.631 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch642142 ms / 321 ms0.637 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch641741 ms / 321 ms0.680 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch641745 ms / 400 ms0.682 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch642312 ms / 400 ms0.688 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch641933 ms / 324 ms0.755 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch641768 ms / 321 ms0.762 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch642689 ms / 400 ms0.800 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch641903 ms / 400 ms0.820 RTF
BF16mpsAcceleratedPassedModeratemacos · aarch642197 ms / 324 ms0.947 RTF
FP32cpuPassedModeratemacos · aarch645500 ms / 1136 ms2.148 RTF
FP32cpuPassedModeratemacos · aarch645524 ms / 1526 ms2.158 RTF
FP32cpuPassedModeratemacos · aarch646910 ms / 1526 ms2.159 RTF
FP32cpuPassedModeratemacos · aarch645573 ms / 2875 ms2.177 RTF
FP32cpuPassedModeratemacos · aarch646982 ms / 2875 ms2.182 RTF
FP32cpuPassedModeratemacos · aarch646994 ms / 1136 ms2.186 RTF
FP32cpuPassedModeratemacos · aarch646998 ms / 1526 ms2.187 RTF
FP32cpuPassedModeratemacos · aarch647034 ms / 1136 ms2.198 RTF
FP32cpuPassedModeratemacos · aarch647139 ms / 2875 ms2.231 RTF
FP32cpuPassedModeratemacos · aarch645402 ms / 1526 ms2.251 RTF
FP32cpuPassedModeratemacos · aarch645412 ms / 2875 ms2.255 RTF
FP32cpuPassedModeratemacos · aarch645553 ms / 1136 ms2.314 RTF
FP32cpuFailed—macos · aarch647990 ms / 945 ms3.329 RTF
FP32cpuFailed—macos · aarch6410018 ms / 945 ms3.130 RTF
FP16mpsFailed—macos · aarch648884 ms / 1359 ms3.001 RTF
FP32mpsFailed—macos · aarch6415231 ms / 1619 ms4.231 RTF
Test setup & sample details
Quality gate
{ "maximumWordErrorRate": 0.25, "minimumSpeakerCosine": null, "minimumRms": 0.005, "maximumClippedFraction": 0.02, "eosRequired": true, "maximumRealTimeFactor": 3, "maximumRequestMs": 45000 }
Tested
2026-11-06T14:34:13.031046+00:00
Environment
{ "os": "macos", "arch": "aarch64", "hardware": "Apple M3 Max", "osVersion": "27.2", "runtime": "App-owned CPython3.13.18; CPU qwen-tts0.1.1/Transformers4.57.3/Torch2.10.0; Metal mlx-audio0.6.3/MLX0.32.4/Transformers5.22.1; original identical checkpoint files" }
Runtime
mlx-audio 0.6.3
Artifact revision
5ecdb67327fd37bb2e042aab12ff7391903235d3
Fixture
{
  "file": "model-samples/desktop/qwen3-tts-family/voicedesign-low-input.json",
  "url": "https://assets.localai.tools/samples/v1/files/eadae4dc62232f2fad097aa8ea3eb8b591aca626a0b4bd1495f907feb345e2f7/voicedesign-low-input.json",
  "size": 319,
  "sha1": "3b42d13dc8eb04fa9a98afe920da4b5339816fcd",
  "sha256": "eadae4dc62232f2fad097aa8ea3eb8b591aca626a0b4bd1495f907feb345e2f7"
}
Rerun when
artifact revision/checksum; qwen-tts/Transformers/Torch versions, preprocessing or sampling; OS/device/backend behavior; fixture text, voice/language selections or independent ASR/quality/performance gates
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "passed": false,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.1320529729127884,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:44.457236+00:00",
      "structuralPassed": true,
      "performancePassed": false,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4736924171447754
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 945.3067080012261,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 7991.344707999815,
      "inferenceMs": 7990.082292000807,
      "realTimeFactor": 3.3292009550003363,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.13205301761627197,
      "peak": 0.7125951647758484,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 8950.60954100154,
      "audioSecondsPerSecond": 0.3003723756891386,
      "grade": "slow",
      "residentMemoryBytes": 9637494784,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "passed": false,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.0951128676533699,
      "audioSeconds": 3.2,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:44.772603+00:00",
      "structuralPassed": true,
      "performancePassed": false,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4283345639705658
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 945.3067080012261,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 10019.167708000168,
      "inferenceMs": 10017.56770799875,
      "realTimeFactor": 3.1304899087496096,
      "audioSeconds": 3.2,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 41,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.0951128825545311,
      "peak": 0.6588311195373535,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 10997.836249000102,
      "audioSecondsPerSecond": 0.3194388192100652,
      "grade": "slow",
      "residentMemoryBytes": 9670164480,
      "supportsStreaming": false
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.12029409408569336,
      "audioSeconds": 3.36,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:46.340434+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5384488105773926
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 399.66295900012483,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 2690.5820419997326,
      "inferenceMs": 2688.8882080002077,
      "realTimeFactor": 0.8002643476191095,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.12029409408569336,
      "peak": 0.7411079406738281,
      "clippedFraction": 0,
      "chunks": 7,
      "firstAudioMs": 451.3404580011411,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583232908,
      "peakMemoryBytes": 5237470947,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 3109.20945900034,
      "audioSecondsPerSecond": 1.2495870932837756,
      "grade": "moderate",
      "residentMemoryBytes": 4893720576
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.10172471404075623,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:46.512416+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5870595574378967
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 399.66295900012483,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1903.4165000011853,
      "inferenceMs": 1902.9447920001985,
      "realTimeFactor": 0.8202348241380166,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.10172471404075623,
      "peak": 0.6221702694892883,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 562.5405840000894,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583134552,
      "peakMemoryBytes": 5237470947,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2317.4331250011164,
      "audioSecondsPerSecond": 1.2191630622985292,
      "grade": "moderate",
      "residentMemoryBytes": 4895719424
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.11006369441747665,
      "audioSeconds": 2.56,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:46.664741+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5889490842819214
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 399.66295900012483,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1745.6557500008785,
      "inferenceMs": 1745.2004169990687,
      "realTimeFactor": 0.6817189128902612,
      "audioSeconds": 2.56,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 33,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.11006369441747665,
      "peak": 0.560906708240509,
      "clippedFraction": 0,
      "chunks": 6,
      "firstAudioMs": 321.07941699905496,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583150948,
      "peakMemoryBytes": 5237470947,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2160.7656680007494,
      "audioSecondsPerSecond": 1.4668802362550468,
      "grade": "moderate",
      "residentMemoryBytes": 4896014336
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.12029409408569336,
      "audioSeconds": 3.36,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:46.825960+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5384488105773926
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 399.66295900012483,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 2312.5957920001383,
      "inferenceMs": 2312.056292001216,
      "realTimeFactor": 0.6881119916670286,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.12029409408569336,
      "peak": 0.7411079406738281,
      "clippedFraction": 0,
      "chunks": 7,
      "firstAudioMs": 342.13933300088684,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583232908,
      "peakMemoryBytes": 5237470947,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2730.5923759995494,
      "audioSecondsPerSecond": 1.4532518138179624,
      "grade": "moderate",
      "residentMemoryBytes": 4895899648
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.12029409408569336,
      "audioSeconds": 3.36,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:47.002470+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5384488105773926
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 320.93075000011595,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 2115.962167001271,
      "inferenceMs": 2115.226750000147,
      "realTimeFactor": 0.6295317708333772,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.12029409408569336,
      "peak": 0.7411079406738281,
      "clippedFraction": 0,
      "chunks": 7,
      "firstAudioMs": 274.1260000002512,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583282060,
      "peakMemoryBytes": 5237782271,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2457.0431249994726,
      "audioSecondsPerSecond": 1.588482180456429,
      "grade": "moderate",
      "residentMemoryBytes": 4894162944
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.10172471404075623,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:47.169490+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5870595574378967
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 320.93075000011595,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1768.2426669998677,
      "inferenceMs": 1767.605791999813,
      "realTimeFactor": 0.7618990482757815,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.10172471404075623,
      "peak": 0.6221702694892883,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 492.14095800016366,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583183704,
      "peakMemoryBytes": 5237782271,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2102.360625001893,
      "audioSecondsPerSecond": 1.312509842692485,
      "grade": "moderate",
      "residentMemoryBytes": 4896423936
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.11006369441747665,
      "audioSeconds": 2.56,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:47.319802+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5889490842819214
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 320.93075000011595,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1741.5440420008963,
      "inferenceMs": 1741.0897499994462,
      "realTimeFactor": 0.6801131835935337,
      "audioSeconds": 2.56,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 33,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.11006369441747665,
      "peak": 0.560906708240509,
      "clippedFraction": 0,
      "chunks": 6,
      "firstAudioMs": 332.33279200067045,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583200100,
      "peakMemoryBytes": 5237782271,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2075.6133750001027,
      "audioSecondsPerSecond": 1.470343501821669,
      "grade": "moderate",
      "residentMemoryBytes": 4896653312
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.12029409408569336,
      "audioSeconds": 3.36,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:47.540015+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5384488105773926
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 320.93075000011595,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 2142.1050829994783,
      "inferenceMs": 2141.7555839998386,
      "realTimeFactor": 0.6374272571428091,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.12029409408569336,
      "peak": 0.7411079406738281,
      "clippedFraction": 0,
      "chunks": 7,
      "firstAudioMs": 304.4497089995275,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583282060,
      "peakMemoryBytes": 5237782271,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2478.1757079999807,
      "audioSecondsPerSecond": 1.5688064619049702,
      "grade": "moderate",
      "residentMemoryBytes": 4898177024
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.12029409408569336,
      "audioSeconds": 3.36,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:47.726086+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5384488105773926
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 324.4161669990717,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1962.460791999547,
      "inferenceMs": 1961.6782080011035,
      "realTimeFactor": 0.5838328000003284,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.12029409408569336,
      "peak": 0.7411079406738281,
      "clippedFraction": 0,
      "chunks": 7,
      "firstAudioMs": 268.0612080002902,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583282060,
      "peakMemoryBytes": 5237665105,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2324.9722509990534,
      "audioSecondsPerSecond": 1.71281914959118,
      "grade": "moderate",
      "residentMemoryBytes": 4894081024
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.10172471404075623,
      "audioSeconds": 2.32,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:47.903799+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5870595574378967
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 324.4161669990717,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 2198.0918339995696,
      "inferenceMs": 2197.0570419998694,
      "realTimeFactor": 0.947007345689599,
      "audioSeconds": 2.32,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 30,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.10172471404075623,
      "peak": 0.6221702694892883,
      "clippedFraction": 0,
      "chunks": 5,
      "firstAudioMs": 317.72470800024166,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583183704,
      "peakMemoryBytes": 5237665105,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2555.3507919994445,
      "audioSecondsPerSecond": 1.0559580182261548,
      "grade": "moderate",
      "residentMemoryBytes": 4896161792
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.11006369441747665,
      "audioSeconds": 2.56,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:48.055471+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5889490842819214
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 324.4161669990717,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 1934.399540999948,
      "inferenceMs": 1932.7620410003874,
      "realTimeFactor": 0.7549851722657763,
      "audioSeconds": 2.56,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 33,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.11006369441747665,
      "peak": 0.560906708240509,
      "clippedFraction": 0,
      "chunks": 6,
      "firstAudioMs": 368.61995800063596,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583200100,
      "peakMemoryBytes": 5237665105,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2283.8211669986777,
      "audioSecondsPerSecond": 1.3245293241970737,
      "grade": "moderate",
      "residentMemoryBytes": 4897488896
    }
  },
  {
    "precision": "bf16",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.12029409408569336,
      "audioSeconds": 3.36,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:48.219368+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.5384488105773926
    },
    "runtime": {
      "engine": "mlx-audio",
      "version": "0.6.3",
      "mlxVersion": "0.32.4",
      "transformersVersion": "5.22.1",
      "loadMs": 324.4161669990717,
      "device": "mps",
      "provider": "MLX Metal",
      "acceleration": "partial",
      "parameterDtype": "Original mixed BF16 talker / FP32 codec",
      "parameterDtypes": {
        "mlx.core.bfloat16": 404,
        "mlx.core.float32": 464
      },
      "placement": {
        "talker": "MLX default GPU",
        "speechTokenizer": "MLX default GPU",
        "speakerEncoder": "not applicable"
      },
      "accelerationDetails": "Neural tensor evaluation on explicit MLX GPU; CPU tokenizer, WAV/resampling and serialization; no CPU device fallback",
      "supportsStreaming": true,
      "strictCodecLoading": true,
      "restoredOriginalCodebookStateFlags": 32,
      "durationMs": 2121.943416999784,
      "inferenceMs": 2121.306167000512,
      "realTimeFactor": 0.6313411211311047,
      "audioSeconds": 3.36,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 43,
      "maxNewTokens": 512,
      "promptTokens": 29,
      "rms": 0.12029409408569336,
      "peak": 0.7411079406738281,
      "clippedFraction": 0,
      "chunks": 7,
      "firstAudioMs": 337.9477920007048,
      "generationPlacement": "MLX GPU",
      "activeMemoryBytes": 4583282060,
      "peakMemoryBytes": 5237665105,
      "sampling": "Seeded .9/50/1.0/1.05; main/subtalker defaults from pinned MLX implementation",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "bf16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 2467.543458998989,
      "audioSecondsPerSecond": 1.5839297750927572,
      "grade": "moderate",
      "residentMemoryBytes": 4896980992
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.0951128676533699,
      "audioSeconds": 3.2,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:48.385325+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4283345639705658
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1135.9806670006947,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 6999.244874999931,
      "inferenceMs": 6994.063833000837,
      "realTimeFactor": 2.1856449478127615,
      "audioSeconds": 3.2,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 41,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.0951128825545311,
      "peak": 0.6588311195373535,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 8146.110666999448,
      "audioSecondsPerSecond": 0.4575308542225621,
      "grade": "moderate",
      "residentMemoryBytes": 9603350528,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.1320529729127884,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:48.551474+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4736924171447754
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1135.9806670006947,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5553.613957999914,
      "inferenceMs": 5552.744041999176,
      "realTimeFactor": 2.31364335083299,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.13205301761627197,
      "peak": 0.7125951647758484,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 6697.437250000803,
      "audioSecondsPerSecond": 0.4322187339893878,
      "grade": "moderate",
      "residentMemoryBytes": 9621192704,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.09285616129636765,
      "audioSeconds": 2.56,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:48.698776+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.49859219789505005
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1135.9806670006947,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5500.997332999759,
      "inferenceMs": 5500.123750000057,
      "realTimeFactor": 2.1484858398437723,
      "audioSeconds": 2.56,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 33,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.09285619109869003,
      "peak": 0.5594902634620667,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 6645.552042000418,
      "audioSecondsPerSecond": 0.46544407296289897,
      "grade": "moderate",
      "residentMemoryBytes": 9654960128,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.0951128676533699,
      "audioSeconds": 3.2,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:48.863012+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4283345639705658
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1135.9806670006947,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 7034.582166999826,
      "inferenceMs": 7033.7211250007385,
      "realTimeFactor": 2.198037851562731,
      "audioSeconds": 3.2,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 41,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.0951128825545311,
      "peak": 0.6588311195373535,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 8180.908626001838,
      "audioSecondsPerSecond": 0.4549512190106434,
      "grade": "moderate",
      "residentMemoryBytes": 9658941440,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.0951128676533699,
      "audioSeconds": 3.2,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:49.027861+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4283345639705658
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1525.945083998522,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 6915.3922500008775,
      "inferenceMs": 6909.774749999997,
      "realTimeFactor": 2.159304609374999,
      "audioSeconds": 3.2,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 41,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.0951128825545311,
      "peak": 0.6588311195373535,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 8451.378624997233,
      "audioSecondsPerSecond": 0.4631120573069334,
      "grade": "moderate",
      "residentMemoryBytes": 9621962752,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.1320529729127884,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:49.182432+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4736924171447754
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1525.945083998522,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5403.293041999859,
      "inferenceMs": 5402.368625000236,
      "realTimeFactor": 2.2509869270834315,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.13205301761627197,
      "peak": 0.7125951647758484,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 6937.3372929985635,
      "audioSecondsPerSecond": 0.44424958135837966,
      "grade": "moderate",
      "residentMemoryBytes": 9636134912,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.09285616129636765,
      "audioSeconds": 2.56,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:49.327084+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.49859219789505005
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1525.945083998522,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5524.569958999564,
      "inferenceMs": 5523.696042000665,
      "realTimeFactor": 2.15769376640651,
      "audioSeconds": 2.56,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 33,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.09285619109869003,
      "peak": 0.5594902634620667,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 7058.979792998798,
      "audioSecondsPerSecond": 0.4634577971949333,
      "grade": "moderate",
      "residentMemoryBytes": 9668591616,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.0951128676533699,
      "audioSeconds": 3.2,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:49.482042+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4283345639705658
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1525.945083998522,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 6998.531291999825,
      "inferenceMs": 6997.636000000057,
      "realTimeFactor": 2.1867612500000178,
      "audioSeconds": 3.2,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 41,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.0951128825545311,
      "peak": 0.6588311195373535,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 8534.717791997537,
      "audioSecondsPerSecond": 0.4572972929715084,
      "grade": "moderate",
      "residentMemoryBytes": 9671770112,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.0951128676533699,
      "audioSeconds": 3.2,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:49.643961+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4283345639705658
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 2875.0193340001715,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 6986.804041000141,
      "inferenceMs": 6981.575957999667,
      "realTimeFactor": 2.181742486874896,
      "audioSeconds": 3.2,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 41,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.0951128825545311,
      "peak": 0.6588311195373535,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 9871.83354300032,
      "audioSecondsPerSecond": 0.45834923507970415,
      "grade": "moderate",
      "residentMemoryBytes": 9616031744,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.1320529729127884,
      "audioSeconds": 2.4,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:49.797642+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4736924171447754
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 2875.0193340001715,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5413.329917000738,
      "inferenceMs": 5412.345291999372,
      "realTimeFactor": 2.255143871666405,
      "audioSeconds": 2.4,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 31,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.13205301761627197,
      "peak": 0.7125951647758484,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 8296.505917000104,
      "audioSecondsPerSecond": 0.44343068864207974,
      "grade": "moderate",
      "residentMemoryBytes": 9623191552,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Please save this file before closing the window.",
      "wordErrorRate": 0,
      "rms": 0.09285616129636765,
      "audioSeconds": 2.56,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:49.943168+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.49859219789505005
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 2875.0193340001715,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 5573.498708999978,
      "inferenceMs": 5572.665957999561,
      "realTimeFactor": 2.1768226398435786,
      "audioSeconds": 2.56,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 33,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.09285619109869003,
      "peak": 0.5594902634620667,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult female speaker with a bright, high-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 8457.178084001498,
      "audioSecondsPerSecond": 0.4593851523300298,
      "grade": "moderate",
      "residentMemoryBytes": 9658433536,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.0951128676533699,
      "audioSeconds": 3.2,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:43:50.126428+00:00",
      "structuralPassed": true,
      "performancePassed": true,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4283345639705658
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 2875.0193340001715,
      "placement": {
        "talker": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "device": "cpu",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "sdpa",
      "durationMs": 7140.439374999914,
      "inferenceMs": 7139.469958001428,
      "realTimeFactor": 2.231084361875446,
      "audioSeconds": 3.2,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 41,
      "generationPlacement": "cpu",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.0951128825545311,
      "peak": 0.6588311195373535,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 2,
      "fallback": false,
      "totalMs": 10025.867124999422,
      "audioSecondsPerSecond": 0.4482125450242507,
      "grade": "moderate",
      "residentMemoryBytes": 9658679296,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp16",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.12078907340765,
      "audioSeconds": 2.96,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:29:03.049148+00:00",
      "structuralPassed": true,
      "performancePassed": false,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.46483319997787476
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1359.3380419988534,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float16"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float16"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float16",
      "attentionImplementation": "sdpa",
      "durationMs": 8889.750374999494,
      "inferenceMs": 8884.267333000025,
      "realTimeFactor": 3.0014416665540624,
      "audioSeconds": 2.96,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 38,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.12078910320997238,
      "peak": 0.6474609375,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "driverAllocatedMemoryBytes": 5184684032,
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp16",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 10259.047875997567,
      "audioSecondsPerSecond": 0.333173225101554,
      "grade": "slow",
      "residentMemoryBytes": 1217282048,
      "supportsStreaming": false
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "fail",
    "quality": {
      "passed": false,
      "transcript": "Good morning. The train leaves at nine.",
      "wordErrorRate": 0,
      "rms": 0.07820147275924683,
      "audioSeconds": 3.6,
      "repetitive": false,
      "gates": {
        "maximumWordErrorRate": 0.25,
        "minimumRms": 0.005,
        "maximumClippedFraction": 0.02,
        "durationSeconds": [
          0.5,
          20
        ],
        "eosRequired": true,
        "repetitionRejected": true,
        "scope": "Short-English intelligibility; no identity/similarity gate for preset or described voices"
      },
      "evaluator": "Pinned independent Qwen3-ASR0.6B MLXQ4; optional unrelated-reference cosine is diagnostic only, not a gate for this task",
      "evaluatedAt": "2026-11-06T14:29:03.662831+00:00",
      "structuralPassed": true,
      "performancePassed": false,
      "speakerSimilarityApplied": false,
      "diagnosticReferenceCosine": 0.4044334888458252
    },
    "runtime": {
      "engine": "qwen-tts",
      "version": "0.1.1",
      "transformersVersion": "4.57.3",
      "torchVersion": "2.10.0",
      "loadMs": 1619.334417000573,
      "placement": {
        "talker": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "speechTokenizer": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "device": "mps",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Talker and speech tokenizer parameters on selected device; CPU text preparation and WAV encoding remain",
      "parameterDtype": "torch.float32",
      "attentionImplementation": "eager",
      "durationMs": 15236.920375000409,
      "inferenceMs": 15230.602292000185,
      "realTimeFactor": 4.2307228588889405,
      "audioSeconds": 3.6,
      "seed": 42,
      "eosReached": true,
      "generatedTokens": 46,
      "generationPlacement": "mps:0",
      "speaker": "Aiden",
      "language": "English",
      "promptTokens": 9,
      "maxNewTokens": 512,
      "rms": 0.0782015472650528,
      "peak": 0.5488303303718567,
      "clippedFraction": 0,
      "sampling": "Official seeded main/subtalker sampling: temperature0.9, topK50, topP1.0, repetitionPenalty1.05",
      "taskMode": "voice_design",
      "instruct": "An adult male speaker with a deep, low-pitched voice, speaking clearly at a natural pace.",
      "driverAllocatedMemoryBytes": 9913958400,
      "artifactRevision": "5ecdb67327fd37bb2e042aab12ff7391903235d3",
      "precision": "fp32",
      "cpuThreads": 4,
      "fallback": false,
      "totalMs": 16870.36716700095,
      "audioSecondsPerSecond": 0.23636622708550967,
      "grade": "slow",
      "residentMemoryBytes": 1395851264,
      "supportsStreaming": false
    }
  }
]

Something not quite right?

Tell us what happened with Qwen3-TTS. 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.