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.

Whisper 3 Desktop

Transcribe WAV audio locally with word-level timestamps in seconds.

Desktop inferencePrivate · Runs on your computer
About this model

Transcribe WAV audio locally with word-level timestamps in seconds.

Created by
OpenAI
License
MIT
Model released
Oct 1, 2024
Source repository created
Oct 1, 2024
Model weights · FP32
1.6 GB

Best for

  • Offline transcription
  • Subtitle editing with manual review

Keep in mind

  • Noise, overlapping speech, silence and accents can degrade transcripts and timestamp alignment. Review names, numbers, omissions and timing before relying on output.
  • Word timestamps are estimates. This adapter provides offline transcription without translation or speaker labels.
  • English short-clip suite only; multilingual accuracy is not established by these measurements.
  • Complete WAV recordings only: 25 ms–30 seconds, mono/stereo, 8–96 kHz. Quiet audio below RMS 1e-5 is skipped; this heuristic can omit very quiet speech.
  • Streaming emits decoded text; estimated word timestamps become available with the final result. No translation, diarization or live microphone streaming.
  • CPU speed limitation: a short 3.4-second synthetic fixture was transcribed exactly with valid timestamps but took about 4.08 seconds (RTF 1.20). CPU fallback is not guaranteed real-time for every fixture; prefer the measured Metal profile when latency matters.

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

Target platforms · verification required

  • macos · aarch64, x86_64
  • windows · x86_64

Recommended configuration

FP32 · mps (verified platform only)

Sources & technical details
Original model
https://huggingface.co/openai/whisper-large-v3-turbo
Release verification
Verified
Release evidence
OpenAI official announcement for the exact large-v3-turbo checkpoint on October 1, 2024; repository/update dates are separate.
Source updated
Oct 4, 2024
Native model files
https://huggingface.co/openai/whisper-large-v3-turbo
Files updated
Oct 4, 2024
Artifact revision
41f01f3fe87f28c78e2fbf8b568835947dd65ed9

Desktop app

Sample result

Whisper 3 Desktop

Run Whisper 3 Desktop 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 / FP32Tested16.0 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.

Whisper 3 Desktop

41 configurations tested
Passed40
Failed1
TestedNov 4, 2026
Runtimetransformers 4.57.6

Quality criteria

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

PrecisionDeviceResultPlatformInference / loadTask metric
FP32mpsAcceleratedPassedVery fastmacos · aarch640 ms / 0 ms0.000 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch640 ms / 0 ms0.000 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641904 ms / 0 ms0.166 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641831 ms / 0 ms0.166 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641849 ms / 0 ms0.168 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641888 ms / 0 ms0.172 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641900 ms / 0 ms0.173 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641907 ms / 0 ms0.173 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641995 ms / 0 ms0.174 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641925 ms / 0 ms0.175 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641927 ms / 0 ms0.175 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641929 ms / 0 ms0.175 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641955 ms / 0 ms0.178 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641997 ms / 0 ms0.182 RTF
FP32mpsAcceleratedPassedFastmacos · aarch642008 ms / 0 ms0.183 RTF
FP32mpsAcceleratedPassedFastmacos · aarch642313 ms / 0 ms0.210 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch642846 ms / 781 ms0.259 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch642896 ms / 764 ms0.263 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch642970 ms / 824 ms0.270 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch643216 ms / 787 ms0.292 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch643253 ms / 852 ms0.296 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch643319 ms / 759 ms0.302 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch641454 ms / 0 ms0.428 RTF
FP32mpsAcceleratedPassedModeratemacos · aarch641455 ms / 0 ms0.428 RTF
FP32cpuPassedVery fastmacos · aarch640 ms / 0 ms0.000 RTF
FP32cpuPassedModeratemacos · aarch643131 ms / 0 ms0.272 RTF
FP32cpuPassedModeratemacos · aarch643098 ms / 0 ms0.282 RTF
FP32cpuPassedModeratemacos · aarch643119 ms / 429 ms0.284 RTF
FP32cpuPassedModeratemacos · aarch643158 ms / 0 ms0.287 RTF
FP32cpuPassedModeratemacos · aarch643175 ms / 0 ms0.289 RTF
FP32cpuPassedModeratemacos · aarch643263 ms / 0 ms0.297 RTF
FP32cpuPassedModeratemacos · aarch643273 ms / 401 ms0.298 RTF
FP32cpuPassedModeratemacos · aarch643351 ms / 0 ms0.305 RTF
FP32cpuPassedModeratemacos · aarch643557 ms / 418 ms0.323 RTF
FP32cpuPassedModeratemacos · aarch643598 ms / 0 ms0.327 RTF
FP32cpuPassedModeratemacos · aarch643882 ms / 0 ms0.338 RTF
FP32cpuPassedModeratemacos · aarch643883 ms / 0 ms0.353 RTF
FP32cpuPassedModeratemacos · aarch643990 ms / 429 ms0.363 RTF
FP32cpuPassedModeratemacos · aarch649165 ms / 574 ms0.833 RTF
FP32cpuPassedModeratemacos · aarch642953 ms / 0 ms0.869 RTF
FP32cpuFailed—macos · aarch644084 ms / 0 ms1.201 RTF
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-04T00:31:38.746102+00:00
Environment
{ "os": "macos", "arch": "aarch64", "hardware": "Apple M3 Max", "osVersion": "27.2", "runtime": "Transformers 4.57.6 / Torch 2.6.0", "interpreter": "App-owned CPython 3.13.18 model-private managed environment" }
Runtime
transformers 4.57.6
Artifact revision
41f01f3fe87f28c78e2fbf8b568835947dd65ed9
Fixture
{
  "file": "model-samples/desktop/qwen3-forced-aligner/clean.wav",
  "size": 352044,
  "sha1": "ce87ae892606eecb8ef65803ed7c4d1ea13ca437",
  "sha256": "ba1a561773d6b6e55c7b9d326d9aa792f6d3a6f430556b83a78f78a2a4cb25e1",
  "url": "https://assets.localai.tools/samples/v1/files/ba1a561773d6b6e55c7b9d326d9aa792f6d3a6f430556b83a78f78a2a4cb25e1/clean.wav"
}
Rerun when
Artifact or runtime revision; Audio preprocessing, decoding or output changes; Fixture, reference or quality gate changes; Relevant OS/backend behavior
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 574.3930840108078,
      "inferenceMs": 9165.481542004272,
      "realTimeFactor": 0.8332255947276611,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 9739.87462601508,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 428.7040420167614,
      "inferenceMs": 3989.5574580004904,
      "realTimeFactor": 0.36268704163640825,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4418.261500017252,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 3883.4094580088276,
      "realTimeFactor": 0.353037223455348,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 3883.4094580088276,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "medianShiftErrorSeconds": 0,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 3882.4215000204276,
      "realTimeFactor": 0.33760186956699373,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 3882.4215000204276,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "fail",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 4083.7409579835366,
      "realTimeFactor": 1.2011002817598637,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4083.7409579835366,
      "grade": "slow",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 429.41129198879935,
      "inferenceMs": 3119.11416699877,
      "realTimeFactor": 0.2835558333635246,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3548.5254589875694,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 3097.916290978901,
      "realTimeFactor": 0.2816287537253546,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3097.916290978901,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "medianShiftErrorSeconds": 0,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 3131.4391670166515,
      "realTimeFactor": 0.27229905800144794,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3131.4391670166515,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 2953.236332978122,
      "realTimeFactor": 0.8685989214641535,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 2953.236332978122,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 0.031042000045999885,
      "realTimeFactor": 0.000015958488802425563,
      "silenceSkipped": true,
      "observedPlacement": {},
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 0.031042000045999885,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 3350.5777919781394,
      "realTimeFactor": 0.3045979810889218,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3350.5777919781394,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 3598.413875006372,
      "realTimeFactor": 0.3271285340914884,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3598.413875006372,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 3157.9976249777246,
      "realTimeFactor": 0.28709069317979313,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3157.9976249777246,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 3263.384416990448,
      "realTimeFactor": 0.29667131063549523,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3263.384416990448,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "streamTextParity": true,
      "streamNativeParity": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 3174.95250000502,
      "realTimeFactor": 0.2886320454550018,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3174.95250000502,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "streamTextParity": true,
      "streamNativeParity": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 418.2338330138009,
      "inferenceMs": 3557.1242919832002,
      "realTimeFactor": 0.3233749356348364,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3975.358124997001,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "streamTextParity": true,
      "streamNativeParity": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "cpu": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 400.6914999918081,
      "inferenceMs": 3273.2874999928754,
      "realTimeFactor": 0.2975715909084432,
      "observedPlacement": {
        "encoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "cpu",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3673.9789999846835,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 787.2843330260366,
      "inferenceMs": 3216.1587919981685,
      "realTimeFactor": 0.2923780719998335,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4003.443125024205,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 2008.36924999021,
      "realTimeFactor": 0.1825790227263827,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2008.36924999021,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "medianShiftErrorSeconds": 0,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1903.5319160029758,
      "realTimeFactor": 0.16552451443504138,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1903.5319160029758,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1453.6135829985142,
      "realTimeFactor": 0.4275334067642689,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1453.6135829985142,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 0.031333009246736765,
      "realTimeFactor": 0.000016228994354605675,
      "silenceSkipped": true,
      "observedPlacement": {},
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 0.031333009246736765,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1906.9446249923203,
      "realTimeFactor": 0.17335860227202912,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1906.9446249923203,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1929.0799589944072,
      "realTimeFactor": 0.17537090536312794,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1929.0799589944072,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1954.8102080007084,
      "realTimeFactor": 0.1777100189091553,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1954.8102080007084,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1887.7060000086203,
      "realTimeFactor": 0.17160963636442003,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1887.7060000086203,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "streamTextParity": true,
      "streamNativeParity": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 2312.9814580024686,
      "realTimeFactor": 0.21027104163658805,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2312.9814580024686,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "streamTextParity": true,
      "streamNativeParity": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 759.2415830004029,
      "inferenceMs": 3318.9707080018707,
      "realTimeFactor": 0.30172460981835186,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4078.2122910022736,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "streamTextParity": true,
      "streamNativeParity": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 851.6915419895668,
      "inferenceMs": 3252.6913750043605,
      "realTimeFactor": 0.29569921590948733,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4104.382916993927,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 763.8015830016229,
      "inferenceMs": 2895.8172079874203,
      "realTimeFactor": 0.26325610981703823,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3659.618790989043,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1927.183583000442,
      "realTimeFactor": 0.17519850754549474,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 1927.183583000442,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "medianShiftErrorSeconds": 0,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1995.2511249866802,
      "realTimeFactor": 0.17350009782492873,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 1995.2511249866802,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1454.695417021867,
      "realTimeFactor": 0.42785159324172556,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 1454.695417021867,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 0.032791984267532825,
      "realTimeFactor": 0.000016958496416918933,
      "silenceSkipped": true,
      "observedPlacement": {},
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 0.032791984267532825,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1900.2377909782808,
      "realTimeFactor": 0.17274889008893463,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 1900.2377909782808,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1849.2450419871602,
      "realTimeFactor": 0.1681131856351964,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 1849.2450419871602,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1831.0840420017485,
      "realTimeFactor": 0.1664621856365226,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 1831.0840420017485,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1924.8114580113906,
      "realTimeFactor": 0.17498285981921732,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 1924.8114580113906,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "streamTextParity": true,
      "streamNativeParity": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 0,
      "inferenceMs": 1997.0966670080088,
      "realTimeFactor": 0.18155424245527352,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 1997.0966670080088,
      "grade": "fast",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "streamTextParity": true,
      "streamNativeParity": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 781.0826670029201,
      "inferenceMs": 2845.882082998287,
      "realTimeFactor": 0.25871655299984425,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3626.964750001207,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "wordErrorRate": 0,
      "validWordTimes": true,
      "completeReconstruction": true,
      "silenceEmpty": true,
      "streamTextParity": true,
      "streamNativeParity": true,
      "passed": true
    },
    "runtime": {
      "placement": {
        "mps:0": 587
      },
      "version": "4.57.6",
      "torchVersion": "2.6.0",
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Audio preprocessing and text decoding on CPU",
      "loadMs": 824.3621660221834,
      "inferenceMs": 2969.819459016435,
      "realTimeFactor": 0.2699835871833123,
      "observedPlacement": {
        "encoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        },
        "decoder": {
          "device": "mps:0",
          "dtype": "torch.float32"
        }
      },
      "silenceSkipped": false,
      "attentionImplementation": "encoder-sdpa/decoder-eager",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 8,
      "totalMs": 3794.181625038618,
      "grade": "moderate",
      "fallback": false,
      "engine": "transformers",
      "artifactRevision": "41f01f3fe87f28c78e2fbf8b568835947dd65ed9",
      "loadTimingScope": "Model construction and device placement after Python/runtime imports; zero for reused warm model. Excludes artifact download, interpreter/package setup and helper process startup.",
      "totalTimingScope": "Measured model load plus inference; not end-to-end first-use setup. Public and packaged reports separately retain full init loadMs."
    }
  }
]

Something not quite right?

Tell us what happened with Whisper 3 Desktop. 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.