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.

Demucs v4

Separate a recording locally into drums, bass, other accompaniment and vocals with Meta Hybrid Transformer Demucs. Produces complete stereo float WAV stems.

Desktop inferencePrivate · Runs on your computer
About this model

Separate a recording locally into drums, bass, other accompaniment and vocals with Meta Hybrid Transformer Demucs. Produces complete stereo float WAV stems.

Created by
Meta
License
MIT
Model released
Unknown
Source repository created
Oct 25, 2019
Model weights · FP32
84.1 MB

Best for

  • Separating vocals and backing tracks
  • Music practice and remix preparation with authorized recordings
  • Inspecting individual instrument stems

Keep in mind

  • Stems can contain bleeding, missing transients, distortion and musical artifacts; not guaranteed isolated studio tracks.
  • Trained for music; spoken audio, unusual instruments, reverberation and overlapping sources can separate poorly.
  • Other is the combined remaining accompaniment; not independent guitar/piano tracks.
  • Not speaker diarization or voice identification; do not use output as forensic ground truth.
  • Input contract is normalized44.1kHz mono/stereo WAV; the app decodes supported audio formats and resamples. No product duration cap, but full WAV input/output and IPC use memory; long files take time.
  • Only FP32 CPU/Metal is verified on the measured Mac. Metal is partial: complex spectral operations execute on CPU by the official runtime.
  • Windows CUDA path is implemented with a separately installed compatible pinned runtime but unverified; other Mac/OS/hardware and AMD/Intel GPU backends are unverified.
  • The app downloads its own pinned CPython and isolated packages on request; installed Python, user packages and Python/pip configuration are not used. Automatic model-package provisioning is verified on Apple Silicon Mac; other platform stacks remain unverified.
  • Synthetic fixture quality gates do not establish MUSDB/natural-music separation accuracy.
  • Respect rights in input recordings and generated stems. Upstream Meta repository is archived; pinned runtime dependencies need maintenance.

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://github.com/facebookresearch/demucs
Release verification
Unknown
Release evidence
Exact original checkpoint public release date has not been verified. Repository creation, training, paper, artifact conversion and update dates are not release evidence.
Source updated
Nov 16, 2023
Native model files
https://github.com/facebookresearch/demucs
Files updated
Oct 26, 2022
Artifact revision
8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4

Desktop app

Sample result

HTDemucs — Music Source Separation

Run Demucs v4 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 / FP32Tested8.0 GiBRAM estimate
macOS Apple SiliconMPS / FP32Tested8.0 GiBRAM estimate
Windows Intel / AMDCPU / FP32Unverified8.0 GiBRAM estimate
macOS Intel / AMDCPU / FP32Unverified8.0 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.

HTDemucs — Music Source Separation

32 configurations tested
Passed32
Failed0
TestedNov 1, 2026
Runtimedemucs 4.0.1

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 · aarch64423 ms / 0 ms0.070 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64425 ms / 0 ms0.071 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64425 ms / 0 ms0.071 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64425 ms / 0 ms0.071 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64428 ms / 0 ms0.071 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64430 ms / 0 ms0.072 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64436 ms / 0 ms0.073 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64438 ms / 0 ms0.073 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64457 ms / 0 ms0.076 RTF
FP32mpsAcceleratedPassedVery fastmacos · aarch64498 ms / 0 ms0.083 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641763 ms / 271 ms0.147 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641407 ms / 269 ms0.235 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641467 ms / 271 ms0.245 RTF
FP32mpsAcceleratedPassedFastmacos · aarch641490 ms / 318 ms0.248 RTF
FP32cpuPassedVery fastmacos · aarch640 ms / 0 ms0.000 RTF
FP32cpuPassedFastmacos · aarch644308 ms / 144 ms0.359 RTF
FP32cpuPassedFastmacos · aarch642617 ms / 0 ms0.436 RTF
FP32cpuPassedFastmacos · aarch642686 ms / 0 ms0.448 RTF
FP32cpuPassedFastmacos · aarch642709 ms / 0 ms0.451 RTF
FP32cpuPassedFastmacos · aarch642748 ms / 0 ms0.458 RTF
FP32cpuPassedFastmacos · aarch642750 ms / 0 ms0.458 RTF
FP32cpuPassedFastmacos · aarch642774 ms / 0 ms0.462 RTF
FP32cpuPassedFastmacos · aarch642781 ms / 0 ms0.464 RTF
FP32cpuPassedFastmacos · aarch642864 ms / 0 ms0.477 RTF
FP32cpuPassedFastmacos · aarch642893 ms / 146 ms0.482 RTF
FP32cpuPassedFastmacos · aarch642908 ms / 0 ms0.485 RTF
FP32cpuPassedFastmacos · aarch642958 ms / 144 ms0.493 RTF
FP32cpuPassedFastmacos · aarch643062 ms / 0 ms0.510 RTF
FP32cpuPassedFastmacos · aarch643217 ms / 145 ms0.536 RTF
FP32cpuPassedModeratemacos · aarch646560 ms / 0 ms1.093 RTF
FP32cpuPassedModeratemacos · aarch646808 ms / 152 ms1.135 RTF
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-01T17:23:11.376147+00:00
Environment
{ "hardware": "Apple M3 Max", "ramBytes": 103079215104, "os": "macOS-27.2-arm64-arm-64bit-Mach-O", "python": "3.13.1", "runtime": "Demucs4.0.1/Torch2.6.0/Torchaudio2.6.0" }
Runtime
demucs 4.0.1
Artifact revision
8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4
Fixture
{
  "file": "model-samples/desktop/demucs/mixture.wav",
  "size": 2116888,
  "sha1": "393be305b010d1c356774ec91949951528d693be",
  "sha256": "069ba9486fea5798c50d89425391784ce8e6f57ceb05a03ddbdce541503d2c4e",
  "url": "https://assets.localai.tools/samples/v1/files/069ba9486fea5798c50d89425391784ce8e6f57ceb05a03ddbdce541503d2c4e/mixture.wav"
}
Rerun when
checkpoint hash; runtime/driver/backend changes; input/config/preprocessing/quality gates/evaluator changes
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995468139648438,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407799601555,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 151.63066699824412,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 6807.888416002243,
      "durationMs": 6807.888416002243,
      "realTimeFactor": 1.134648069333707,
      "audioSecondsPerSecond": 0.8813305438286466,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 6959.519083000487,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995468139648438,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407799601555,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 6559.595208000246,
      "durationMs": 6559.595208000246,
      "realTimeFactor": 1.093265868000041,
      "audioSecondsPerSecond": 0.9146905883281106,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 6559.595208000246,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "moderate"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995466232299805,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 143.52770800178405,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 2958.21849999993,
      "durationMs": 2958.21849999993,
      "realTimeFactor": 0.493036416666655,
      "audioSecondsPerSecond": 2.0282477443772806,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 3101.746208001714,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.644685745239258,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.009105682373047,
        "bass": 16.03553581237793,
        "other": 23.863502502441406,
        "vocals": 11.331764221191406
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.2059218883514404,
        "bass": -0.41152554750442505,
        "other": -6.680319786071777,
        "vocals": -21.509140014648438
      },
      "stemSnrImprovementDb": {
        "drums": 17.215027570724487,
        "bass": 16.447061359882355,
        "other": 30.543822288513184,
        "vocals": 32.840904235839844
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.03418053314089775,
        "bass": 0.03638255223631859,
        "other": 0.024187276139855385,
        "vocals": 0.004705666098743677
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 2708.980082999915,
      "durationMs": 2708.980082999915,
      "realTimeFactor": 0.4514966804999858,
      "audioSecondsPerSecond": 2.214855708114185,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2708.980082999915,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 29.665307998657227,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": -111.55569458007812,
        "bass": 44.849735260009766,
        "other": -110.82048034667969,
        "vocals": -111.9467544555664
      },
      "mixtureBaselineSnrDb": {
        "drums": -151.89328002929688,
        "bass": 151.89328002929688,
        "other": -151.89328002929688,
        "vocals": -151.89328002929688
      },
      "stemSnrImprovementDb": {
        "drums": 40.33758544921875,
        "bass": -107.04354476928711,
        "other": 41.07279968261719,
        "vocals": 39.94652557373047
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.0005199654842726886,
        "bass": 0.05405852943658829,
        "other": 0.00047776481369510293,
        "vocals": 0.0005439108936116099
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 2750.441457999841,
      "durationMs": 2750.441457999841,
      "realTimeFactor": 0.45840690966664016,
      "audioSecondsPerSecond": 2.181467990365184,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2750.441457999841,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 0,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "mixtureBaselineSnrDb": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "stemSnrImprovementDb": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 0.07375000132014975,
      "durationMs": 0.07375000132014975,
      "realTimeFactor": 0.00007375000132014975,
      "audioSecondsPerSecond": 13559.321791181894,
      "activationPlacement": {},
      "silenceFastPath": true,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 0.07375000132014975,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995466232299805,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 2907.594209002127,
      "durationMs": 2907.594209002127,
      "realTimeFactor": 0.4845990348336879,
      "audioSecondsPerSecond": 2.0635616832030945,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2907.594209002127,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995466232299805,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 2864.0679589989304,
      "durationMs": 2864.0679589989304,
      "realTimeFactor": 0.47734465983315505,
      "audioSecondsPerSecond": 2.0949223572534095,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2864.0679589989304,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995466232299805,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 3062.4404169975605,
      "durationMs": 3062.4404169975605,
      "realTimeFactor": 0.5104067361662601,
      "audioSecondsPerSecond": 1.959221791450377,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 3062.4404169975605,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995466232299805,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 146.06900000217138,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 2892.867374997877,
      "durationMs": 2892.867374997877,
      "realTimeFactor": 0.48214456249964616,
      "audioSecondsPerSecond": 2.0740667380247264,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 3038.9363750000484,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995466232299805,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 2616.8638749986712,
      "durationMs": 2616.8638749986712,
      "realTimeFactor": 0.4361439791664452,
      "audioSecondsPerSecond": 2.2928208292848424,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2616.8638749986712,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995466232299805,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 2781.370667002193,
      "durationMs": 2781.370667002193,
      "realTimeFactor": 0.46356177783369884,
      "audioSecondsPerSecond": 2.157209778323757,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2781.370667002193,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995466232299805,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 144.99100000102771,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 3217.4237920007727,
      "durationMs": 3217.4237920007727,
      "realTimeFactor": 0.5362372986667955,
      "audioSecondsPerSecond": 1.8648460345563824,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 3362.4147920018004,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995466232299805,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 2747.551750002458,
      "durationMs": 2747.551750002458,
      "realTimeFactor": 0.45792529166707635,
      "audioSecondsPerSecond": 2.1837623258577867,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2747.551750002458,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995466232299805,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120024681091309,
        "bass": 16.32056427001953,
        "other": 20.95989418029785,
        "vocals": 16.642501831054688
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.843475580215454,
        "bass": 17.388606905937195,
        "other": 28.039381980895996,
        "vocals": 27.27028179168701
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.04858150705695152,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 2685.9365420023096,
      "durationMs": 2685.9365420023096,
      "realTimeFactor": 0.4476560903337183,
      "audioSecondsPerSecond": 2.233857690296408,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2685.9365420023096,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995464324951172,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120025634765625,
        "bass": 16.320568084716797,
        "other": 20.95989227294922,
        "vocals": 16.642498016357422
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.84347653388977,
        "bass": 17.38861072063446,
        "other": 28.039380073547363,
        "vocals": 27.270277976989746
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.048581503331661224,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 268.878582999605,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 1407.4253340004361,
      "durationMs": 1407.4253340004361,
      "realTimeFactor": 0.2345708890000727,
      "audioSecondsPerSecond": 4.263103594238798,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1676.303917000041,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.644685745239258,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.009109497070312,
        "bass": 16.03553581237793,
        "other": 23.863502502441406,
        "vocals": 11.331762313842773
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.2059218883514404,
        "bass": -0.41152554750442505,
        "other": -6.680319786071777,
        "vocals": -21.509140014648438
      },
      "stemSnrImprovementDb": {
        "drums": 17.215031385421753,
        "bass": 16.447061359882355,
        "other": 30.543822288513184,
        "vocals": 32.84090232849121
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.034180525690317154,
        "bass": 0.03638255223631859,
        "other": 0.024187276139855385,
        "vocals": 0.004705666098743677
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 422.87225000109174,
      "durationMs": 422.87225000109174,
      "realTimeFactor": 0.07047870833351529,
      "audioSecondsPerSecond": 14.188682279304233,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 422.87225000109174,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 29.665306091308594,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": -111.55569458007812,
        "bass": 44.849769592285156,
        "other": -110.82049560546875,
        "vocals": -111.9467544555664
      },
      "mixtureBaselineSnrDb": {
        "drums": -151.89328002929688,
        "bass": 151.89328002929688,
        "other": -151.89328002929688,
        "vocals": -151.89328002929688
      },
      "stemSnrImprovementDb": {
        "drums": 40.33758544921875,
        "bass": -107.04351043701172,
        "other": 41.072784423828125,
        "vocals": 39.94652557373047
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.0005199655424803495,
        "bass": 0.05405852943658829,
        "other": 0.0004777655703946948,
        "vocals": 0.0005439109518192708
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 430.1116250026098,
      "durationMs": 430.1116250026098,
      "realTimeFactor": 0.0716852708337683,
      "audioSecondsPerSecond": 13.949867083838047,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 430.1116250026098,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 0,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "mixtureBaselineSnrDb": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "stemSnrImprovementDb": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 0.09249999857274815,
      "durationMs": 0.09249999857274815,
      "realTimeFactor": 0.00009249999857274815,
      "audioSecondsPerSecond": 10810.810977618918,
      "activationPlacement": {},
      "silenceFastPath": true,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 0.09249999857274815,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995464324951172,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120025634765625,
        "bass": 16.320568084716797,
        "other": 20.95989227294922,
        "vocals": 16.642498016357422
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.84347653388977,
        "bass": 17.38861072063446,
        "other": 28.039380073547363,
        "vocals": 27.270277976989746
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.048581503331661224,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 425.400083000568,
      "durationMs": 425.400083000568,
      "realTimeFactor": 0.07090001383342799,
      "audioSecondsPerSecond": 14.104369603500968,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 425.400083000568,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995464324951172,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120025634765625,
        "bass": 16.320568084716797,
        "other": 20.95989227294922,
        "vocals": 16.642498016357422
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.84347653388977,
        "bass": 17.38861072063446,
        "other": 28.039380073547363,
        "vocals": 27.270277976989746
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.048581503331661224,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 436.4345409994712,
      "durationMs": 436.4345409994712,
      "realTimeFactor": 0.07273909016657853,
      "audioSecondsPerSecond": 13.74776612836258,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 436.4345409994712,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995464324951172,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120025634765625,
        "bass": 16.320568084716797,
        "other": 20.95989227294922,
        "vocals": 16.642498016357422
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.84347653388977,
        "bass": 17.38861072063446,
        "other": 28.039380073547363,
        "vocals": 27.270277976989746
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.048581503331661224,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 425.1729579991661,
      "durationMs": 425.1729579991661,
      "realTimeFactor": 0.07086215966652769,
      "audioSecondsPerSecond": 14.111904078367486,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 425.1729579991661,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995464324951172,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120025634765625,
        "bass": 16.320568084716797,
        "other": 20.95989227294922,
        "vocals": 16.642498016357422
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.84347653388977,
        "bass": 17.38861072063446,
        "other": 28.039380073547363,
        "vocals": 27.270277976989746
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.048581503331661224,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 270.93629099908867,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 1467.0566659988253,
      "durationMs": 1467.0566659988253,
      "realTimeFactor": 0.24450944433313757,
      "audioSecondsPerSecond": 4.089821572035176,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1737.992956997914,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995464324951172,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120025634765625,
        "bass": 16.320568084716797,
        "other": 20.95989227294922,
        "vocals": 16.642498016357422
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.84347653388977,
        "bass": 17.38861072063446,
        "other": 28.039380073547363,
        "vocals": 27.270277976989746
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.048581503331661224,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 438.3467500010738,
      "durationMs": 438.3467500010738,
      "realTimeFactor": 0.07305779166684563,
      "audioSecondsPerSecond": 13.687793966729084,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 438.3467500010738,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995464324951172,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120025634765625,
        "bass": 16.320568084716797,
        "other": 20.95989227294922,
        "vocals": 16.642498016357422
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.84347653388977,
        "bass": 17.38861072063446,
        "other": 28.039380073547363,
        "vocals": 27.270277976989746
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.048581503331661224,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 497.97320799916633,
      "durationMs": 497.97320799916633,
      "realTimeFactor": 0.08299553466652772,
      "audioSecondsPerSecond": 12.048840989071936,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 497.97320799916633,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995464324951172,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120025634765625,
        "bass": 16.320568084716797,
        "other": 20.95989227294922,
        "vocals": 16.642498016357422
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.84347653388977,
        "bass": 17.38861072063446,
        "other": 28.039380073547363,
        "vocals": 27.270277976989746
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.048581503331661224,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 317.7285839992692,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 1490.078499999072,
      "durationMs": 1490.078499999072,
      "realTimeFactor": 0.248346416666512,
      "audioSecondsPerSecond": 4.026633496157241,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 1807.807083998341,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995464324951172,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120025634765625,
        "bass": 16.320568084716797,
        "other": 20.95989227294922,
        "vocals": 16.642498016357422
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.84347653388977,
        "bass": 17.38861072063446,
        "other": 28.039380073547363,
        "vocals": 27.270277976989746
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.048581503331661224,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 427.9361670014623,
      "durationMs": 427.9361670014623,
      "realTimeFactor": 0.07132269450024371,
      "audioSecondsPerSecond": 14.020782683646127,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 427.9361670014623,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "metric": "Ground-truth stem SNR improvement + mixture reconstruction + finite/duration/negative checks",
      "reconstructionSnrDb": 25.995464324951172,
      "minimumReconstructionSnrDb": 12,
      "stemSnrDb": {
        "drums": 14.120025634765625,
        "bass": 16.320568084716797,
        "other": 20.95989227294922,
        "vocals": 16.642498016357422
      },
      "mixtureBaselineSnrDb": {
        "drums": -3.7234508991241455,
        "bass": -1.0680426359176636,
        "other": -7.0794878005981445,
        "vocals": -10.627779960632324
      },
      "stemSnrImprovementDb": {
        "drums": 17.84347653388977,
        "bass": 17.38861072063446,
        "other": 28.039380073547363,
        "vocals": 27.270277976989746
      },
      "minimumMeanImprovementDb": 0,
      "stemRms": {
        "drums": 0.048581503331661224,
        "bass": 0.05213407427072525,
        "other": 0.03451025113463402,
        "vocals": 0.023751571774482727
      },
      "scope": "Synthetic authored sources, not natural-music SDR certification"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 424.97875000117347,
      "durationMs": 424.97875000117347,
      "realTimeFactor": 0.07082979166686225,
      "audioSecondsPerSecond": 14.11835297643337,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 424.97875000117347,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 25.816709518432617,
      "minimumReconstructionSnrDb": 12,
      "stemSnrImprovementDb": {
        "drums": 17.70212197303772,
        "bass": 17.233455538749695,
        "other": 27.927823066711426,
        "vocals": 27.20649814605713
      },
      "minimumMeanImprovementDb": 0,
      "scope": "Synthetic ground-truth chunk boundary and shifted averaging"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 143.90774999992573,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 4308.371958999487,
      "durationMs": 4308.371958999487,
      "realTimeFactor": 0.3590309965832906,
      "audioSecondsPerSecond": 2.785274835645041,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 4452.279708999413,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 26.040983200073242,
      "minimumReconstructionSnrDb": 12,
      "stemSnrImprovementDb": {
        "drums": 18.007781267166138,
        "bass": 17.551740527153015,
        "other": 27.350289344787598,
        "vocals": 26.992804527282715
      },
      "minimumMeanImprovementDb": 0,
      "scope": "Synthetic ground-truth chunk boundary and shifted averaging"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 2773.979875000805,
      "durationMs": 2773.979875000805,
      "realTimeFactor": 0.4623299791668008,
      "audioSecondsPerSecond": 2.1629572925428158,
      "activationPlacement": {
        "input": "cpu",
        "output": "cpu"
      },
      "silenceFastPath": false,
      "overlap": 0.5,
      "shifts": 1,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2773.979875000805,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 25.816699981689453,
      "minimumReconstructionSnrDb": 12,
      "stemSnrImprovementDb": {
        "drums": 17.702121019363403,
        "bass": 17.233455538749695,
        "other": 27.927823066711426,
        "vocals": 27.206496238708496
      },
      "minimumMeanImprovementDb": 0,
      "scope": "Synthetic ground-truth chunk boundary and shifted averaging"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 270.56795799944666,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 1762.9492499982007,
      "durationMs": 1762.9492499982007,
      "realTimeFactor": 0.14691243749985006,
      "audioSecondsPerSecond": 6.806775634643055,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 2033.5172079976473,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  },
  {
    "precision": "fp32",
    "device": "mps",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 26.04097557067871,
      "minimumReconstructionSnrDb": 12,
      "stemSnrImprovementDb": {
        "drums": 18.007786989212036,
        "bass": 17.551746249198914,
        "other": 27.35029125213623,
        "vocals": 26.992802619934082
      },
      "minimumMeanImprovementDb": 0,
      "scope": "Synthetic ground-truth chunk boundary and shifted averaging"
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "mps:0/torch.float32": 533
      },
      "parameterDtype": "torch.float32",
      "provider": "Metal/MPS",
      "acceleration": "partial",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 456.58095800172305,
      "durationMs": 456.58095800172305,
      "realTimeFactor": 0.07609682633362051,
      "audioSecondsPerSecond": 13.141152505044587,
      "activationPlacement": {
        "input": "mps:0",
        "output": "mps:0"
      },
      "silenceFastPath": false,
      "overlap": 0.5,
      "shifts": 1,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "driverAllocatedMemoryBytes": 2591473664,
      "memoryMeasurementScope": "Post-inference driver allocation, not peak",
      "device": "mps",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 456.58095800172305,
      "artifactRevision": "8726e21a993978c7ba086d3872e7608d7d5bfca646ca4aca459ffda844faa8b4",
      "fallback": false,
      "grade": "fast"
    }
  }
]

HTDemucs Fine-tuned — Music Source Separation

10 configurations tested
Passed10
Failed0
TestedNov 5, 2026
Runtimedemucs 4.0.1

Quality criteria

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

PrecisionDeviceResultPlatformInference / loadTask metric
FP32cpuPassedVery fastmacos · aarch640 ms / 0 ms0.000 RTF
FP32cpuPassedVery fastmacos · aarch640 ms / 0 ms0.000 RTF
FP32cpuPassedModeratemacos · aarch6419638 ms / 0 ms1.637 RTF
FP32cpuPassedModeratemacos · aarch6413286 ms / 593 ms2.214 RTF
FP32cpuPassedModeratemacos · aarch6413827 ms / 0 ms2.304 RTF
FP32cpuPassedModeratemacos · aarch6413951 ms / 0 ms2.325 RTF
FP32cpuPassedSlowmacos · aarch6441541 ms / 0 ms3.462 RTF
FP32cpuPassedSlowmacos · aarch6427380 ms / 0 ms4.563 RTF
FP32cpuPassedSlowmacos · aarch6427479 ms / 0 ms4.580 RTF
FP32cpuPassedSlowmacos · aarch6428534 ms / 659 ms4.756 RTF
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-05T19:57:09.951358+00:00
Environment
{ "hardware": "Apple M3 Max", "ramBytes": 103079215104, "os": "macOS-27.2-arm64-arm-64bit-Mach-O", "python": "3.13.1", "runtime": "Demucs4.0.1/Torch2.6.0/Torchaudio2.6.0" }
Runtime
demucs 4.0.1
Artifact revision
c9f602f3858524952bbdd216d18c94a8deef16f81e70ea2bb2a30747ae2f4ee6
Fixture
{
  "file": "model-samples/desktop/demucs/mixture.wav",
  "size": 2116888,
  "sha1": "393be305b010d1c356774ec91949951528d693be",
  "sha256": "069ba9486fea5798c50d89425391784ce8e6f57ceb05a03ddbdce541503d2c4e",
  "url": "https://assets.localai.tools/samples/v1/files/069ba9486fea5798c50d89425391784ce8e6f57ceb05a03ddbdce541503d2c4e/mixture.wav"
}
Rerun when
checkpoint hash; runtime/driver/backend changes; input/config/preprocessing/quality gates/evaluator changes
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 25.70563316345215,
      "stemSnrDb": {
        "drums": 17.779651641845703,
        "bass": 19.223316192626953,
        "other": 22.066192626953125,
        "vocals": 17.560930252075195
      },
      "stemSnrImprovementDb": {
        "drums": 21.50310254096985,
        "bass": 20.291358828544617,
        "other": 29.14568042755127,
        "vocals": 28.18871021270752
      },
      "minimumReconstructionSnrDb": 12,
      "minimumBassOrNamedInstrumentSnrDb": 6,
      "minimumMeanImprovementDb": 0
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 659.2849590233527,
      "placement": {
        "cpu/torch.float32": 2132
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 28534.049416950438,
      "durationMs": 28534.049416950438,
      "realTimeFactor": 4.755674902825073,
      "audioSecondsPerSecond": 0.21027509668626793,
      "activationPlacement": {
        "0": {
          "input": "cpu",
          "output": "cpu"
        },
        "1": {
          "input": "cpu",
          "output": "cpu"
        },
        "2": {
          "input": "cpu",
          "output": "cpu"
        },
        "3": {
          "input": "cpu",
          "output": "cpu"
        }
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 29193.33437597379,
      "grade": "slow",
      "artifactRevision": "c9f602f3858524952bbdd216d18c94a8deef16f81e70ea2bb2a30747ae2f4ee6",
      "fallback": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 25.418880462646484,
      "stemSnrDb": {
        "drums": 18.02065658569336,
        "bass": 18.87936782836914,
        "other": 24.838966369628906,
        "vocals": 12.057100296020508
      },
      "stemSnrImprovementDb": {
        "drums": 21.2265784740448,
        "bass": 19.290893375873566,
        "other": 31.519286155700684,
        "vocals": 33.566240310668945
      },
      "minimumReconstructionSnrDb": 12,
      "minimumBassOrNamedInstrumentSnrDb": 6,
      "minimumMeanImprovementDb": 0
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 2132
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 27379.6385829919,
      "durationMs": 27379.6385829919,
      "realTimeFactor": 4.563273097165317,
      "audioSecondsPerSecond": 0.21914094964449862,
      "activationPlacement": {
        "0": {
          "input": "cpu",
          "output": "cpu"
        },
        "1": {
          "input": "cpu",
          "output": "cpu"
        },
        "2": {
          "input": "cpu",
          "output": "cpu"
        },
        "3": {
          "input": "cpu",
          "output": "cpu"
        }
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 27379.6385829919,
      "grade": "slow",
      "artifactRevision": "c9f602f3858524952bbdd216d18c94a8deef16f81e70ea2bb2a30747ae2f4ee6",
      "fallback": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 30.046594619750977,
      "stemSnrDb": {
        "drums": -110.99014282226562,
        "bass": 43.67986297607422,
        "other": -111.33708953857422,
        "vocals": -111.74966430664062
      },
      "stemSnrImprovementDb": {
        "drums": 40.90313720703125,
        "bass": -108.21341705322266,
        "other": 40.556190490722656,
        "vocals": 40.14361572265625
      },
      "minimumReconstructionSnrDb": 12,
      "minimumBassOrNamedInstrumentSnrDb": 6,
      "minimumMeanImprovementDb": 0
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 2132
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 27479.2546670069,
      "durationMs": 27479.2546670069,
      "realTimeFactor": 4.579875777834483,
      "audioSecondsPerSecond": 0.21834653351074798,
      "activationPlacement": {
        "0": {
          "input": "cpu",
          "output": "cpu"
        },
        "1": {
          "input": "cpu",
          "output": "cpu"
        },
        "2": {
          "input": "cpu",
          "output": "cpu"
        },
        "3": {
          "input": "cpu",
          "output": "cpu"
        }
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 27479.2546670069,
      "grade": "slow",
      "artifactRevision": "c9f602f3858524952bbdd216d18c94a8deef16f81e70ea2bb2a30747ae2f4ee6",
      "fallback": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 0,
      "stemSnrDb": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "stemSnrImprovementDb": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "minimumReconstructionSnrDb": 12,
      "minimumBassOrNamedInstrumentSnrDb": 6,
      "minimumMeanImprovementDb": 0
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 2132
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 0.13633305206894875,
      "durationMs": 0.13633305206894875,
      "realTimeFactor": 0.00013633305206894875,
      "audioSecondsPerSecond": 7334.978457712972,
      "activationPlacement": {},
      "silenceFastPath": true,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 0.13633305206894875,
      "grade": "fast",
      "artifactRevision": "c9f602f3858524952bbdd216d18c94a8deef16f81e70ea2bb2a30747ae2f4ee6",
      "fallback": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 25.56863021850586,
      "stemSnrDb": {
        "drums": 18.012126922607422,
        "bass": 19.68528175354004,
        "other": 22.16707992553711,
        "vocals": 17.67275619506836
      },
      "stemSnrImprovementDb": {
        "drums": 21.735577821731567,
        "bass": 20.753324389457703,
        "other": 29.246567726135254,
        "vocals": 28.300536155700684
      },
      "minimumReconstructionSnrDb": 12,
      "minimumBassOrNamedInstrumentSnrDb": 6,
      "minimumMeanImprovementDb": 0
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 2132
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 41540.56587500963,
      "durationMs": 41540.56587500963,
      "realTimeFactor": 3.4617138229174693,
      "audioSecondsPerSecond": 0.28887425453246107,
      "activationPlacement": {
        "0": {
          "input": "cpu",
          "output": "cpu"
        },
        "1": {
          "input": "cpu",
          "output": "cpu"
        },
        "2": {
          "input": "cpu",
          "output": "cpu"
        },
        "3": {
          "input": "cpu",
          "output": "cpu"
        }
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 1,
      "totalMs": 41540.56587500963,
      "grade": "slow",
      "artifactRevision": "c9f602f3858524952bbdd216d18c94a8deef16f81e70ea2bb2a30747ae2f4ee6",
      "fallback": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 25.70563316345215,
      "stemSnrDb": {
        "drums": 17.779653549194336,
        "bass": 19.22332191467285,
        "other": 22.066192626953125,
        "vocals": 17.560932159423828
      },
      "stemSnrImprovementDb": {
        "drums": 21.50310444831848,
        "bass": 20.291364550590515,
        "other": 29.14568042755127,
        "vocals": 28.188712120056152
      },
      "minimumReconstructionSnrDb": 12,
      "minimumBassOrNamedInstrumentSnrDb": 6,
      "minimumMeanImprovementDb": 0
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 592.5954999984242,
      "placement": {
        "cpu/torch.float32": 2132
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 13285.981166991405,
      "durationMs": 13285.981166991405,
      "realTimeFactor": 2.2143301944985674,
      "audioSecondsPerSecond": 0.4516038314811711,
      "activationPlacement": {
        "0": {
          "input": "cpu",
          "output": "cpu"
        },
        "1": {
          "input": "cpu",
          "output": "cpu"
        },
        "2": {
          "input": "cpu",
          "output": "cpu"
        },
        "3": {
          "input": "cpu",
          "output": "cpu"
        }
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 13878.57666698983,
      "grade": "moderate",
      "artifactRevision": "c9f602f3858524952bbdd216d18c94a8deef16f81e70ea2bb2a30747ae2f4ee6",
      "fallback": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 25.41887664794922,
      "stemSnrDb": {
        "drums": 18.02066421508789,
        "bass": 18.879371643066406,
        "other": 24.838966369628906,
        "vocals": 12.057099342346191
      },
      "stemSnrImprovementDb": {
        "drums": 21.22658610343933,
        "bass": 19.29089719057083,
        "other": 31.519286155700684,
        "vocals": 33.56623935699463
      },
      "minimumReconstructionSnrDb": 12,
      "minimumBassOrNamedInstrumentSnrDb": 6,
      "minimumMeanImprovementDb": 0
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 2132
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 13950.687166012358,
      "durationMs": 13950.687166012358,
      "realTimeFactor": 2.325114527668726,
      "audioSecondsPerSecond": 0.4300863411673097,
      "activationPlacement": {
        "0": {
          "input": "cpu",
          "output": "cpu"
        },
        "1": {
          "input": "cpu",
          "output": "cpu"
        },
        "2": {
          "input": "cpu",
          "output": "cpu"
        },
        "3": {
          "input": "cpu",
          "output": "cpu"
        }
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 13950.687166012358,
      "grade": "moderate",
      "artifactRevision": "c9f602f3858524952bbdd216d18c94a8deef16f81e70ea2bb2a30747ae2f4ee6",
      "fallback": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 30.046594619750977,
      "stemSnrDb": {
        "drums": -110.99014282226562,
        "bass": 43.679866790771484,
        "other": -111.33709716796875,
        "vocals": -111.74966430664062
      },
      "stemSnrImprovementDb": {
        "drums": 40.90313720703125,
        "bass": -108.21341323852539,
        "other": 40.556182861328125,
        "vocals": 40.14361572265625
      },
      "minimumReconstructionSnrDb": 12,
      "minimumBassOrNamedInstrumentSnrDb": 6,
      "minimumMeanImprovementDb": 0
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 2132
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 13826.643000007607,
      "durationMs": 13826.643000007607,
      "realTimeFactor": 2.304440500001268,
      "audioSecondsPerSecond": 0.43394481220038,
      "activationPlacement": {
        "0": {
          "input": "cpu",
          "output": "cpu"
        },
        "1": {
          "input": "cpu",
          "output": "cpu"
        },
        "2": {
          "input": "cpu",
          "output": "cpu"
        },
        "3": {
          "input": "cpu",
          "output": "cpu"
        }
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 13826.643000007607,
      "grade": "moderate",
      "artifactRevision": "c9f602f3858524952bbdd216d18c94a8deef16f81e70ea2bb2a30747ae2f4ee6",
      "fallback": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 0,
      "stemSnrDb": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "stemSnrImprovementDb": {
        "drums": 0,
        "bass": 0,
        "other": 0,
        "vocals": 0
      },
      "minimumReconstructionSnrDb": 12,
      "minimumBassOrNamedInstrumentSnrDb": 6,
      "minimumMeanImprovementDb": 0
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 2132
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 0.07908296538516879,
      "durationMs": 0.07908296538516879,
      "realTimeFactor": 0.00007908296538516879,
      "audioSecondsPerSecond": 12644.948189911198,
      "activationPlacement": {},
      "silenceFastPath": true,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 0.07908296538516879,
      "grade": "fast",
      "artifactRevision": "c9f602f3858524952bbdd216d18c94a8deef16f81e70ea2bb2a30747ae2f4ee6",
      "fallback": false
    }
  },
  {
    "precision": "fp32",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "reconstructionSnrDb": 25.568628311157227,
      "stemSnrDb": {
        "drums": 18.012128829956055,
        "bass": 19.685279846191406,
        "other": 22.16707992553711,
        "vocals": 17.67275619506836
      },
      "stemSnrImprovementDb": {
        "drums": 21.7355797290802,
        "bass": 20.75332248210907,
        "other": 29.246567726135254,
        "vocals": 28.300536155700684
      },
      "minimumReconstructionSnrDb": 12,
      "minimumBassOrNamedInstrumentSnrDb": 6,
      "minimumMeanImprovementDb": 0
    },
    "runtime": {
      "engine": "demucs",
      "version": "4.0.1",
      "torchVersion": "2.6.0",
      "loadMs": 0,
      "placement": {
        "cpu/torch.float32": 2132
      },
      "parameterDtype": "torch.float32",
      "provider": "PyTorch CPU",
      "acceleration": "cpu",
      "accelerationDetails": "Model convolution/transformer parameters and activations on selected device. Official MPS path uses CPU STFT/iSTFT/complex reconstruction; WAV decode, normalization, chunk assembly and encoding also use CPU.",
      "inferenceMs": 19638.384375022724,
      "durationMs": 19638.384375022724,
      "realTimeFactor": 1.6365320312518936,
      "audioSecondsPerSecond": 0.611048229367703,
      "activationPlacement": {
        "0": {
          "input": "cpu",
          "output": "cpu"
        },
        "1": {
          "input": "cpu",
          "output": "cpu"
        },
        "2": {
          "input": "cpu",
          "output": "cpu"
        },
        "3": {
          "input": "cpu",
          "output": "cpu"
        }
      },
      "silenceFastPath": false,
      "overlap": 0.25,
      "shifts": 0,
      "seed": 42,
      "segmentSeconds": 7.8,
      "outputEncoding": "32bit float WAV; no independent stem clipping/normalization",
      "inputChannels": 2,
      "device": "cpu",
      "precision": "fp32",
      "cpuThreads": 4,
      "totalMs": 19638.384375022724,
      "grade": "moderate",
      "artifactRevision": "c9f602f3858524952bbdd216d18c94a8deef16f81e70ea2bb2a30747ae2f4ee6",
      "fallback": false
    }
  }
]

Something not quite right?

Tell us what happened with Demucs v4. 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.