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.

Sherpa-ONNX — Speaker Diarization

Identify who spoke when in mono16kHz audio, locally in a browser module worker. Combines pyannote segmentation3.0,3D-Speaker ERes2Net embeddings and agglomerative clustering, returning the complete native timeline.

Browser inferencePowered by WebAI · Private · Unlimited runs
About this model

Identify who spoke when in mono16kHz audio, locally in a browser module worker. Combines pyannote segmentation3.0,3D-Speaker ERes2Net embeddings and agglomerative clustering, returning the complete native timeline.

Model released
Unknown
Source repository created
Sep 1, 2022
Model weights · FP32
45.6 MB

Best for

  • Exploring speaker turns in authorized recordings
  • Preparing local speaker timelines for separate ASR alignment

Keep in mind

  • Speaker labels are anonymous recording-local clusters, not identified people; they can swap, split or merge.
  • Noise, overlap, short turns, similar voices and non-speech can cause errors. Synthetic fixture results do not certify natural meetings.
  • The embedding checkpoint was trained on Chinese3D-Speaker data; quality on other languages/domains requires validation.
  • WASM SIMD CPU only; this build has no WebGPU backend. The runtime initially allocates512MiB and grows; long recordings require substantial memory.
  • Input must be decoded mono16kHz Float32Array PCM; file decoding/downmix/resampling happens in the application/sample helper.
  • This is offline diarization, not live streaming or transcription. Segments can overlap and confidence probabilities are not provided by this native API.
  • Original runtime/model licenses and component attribution apply; source and converter are recorded separately.
  • Automatic speaker count is sensitive to resampling and clustering threshold. Initial0.5/0.6 settings split an extra cluster on browser-prepared synthetic audio; the retained measured0.7 setting uses unchanged DER/count gates. Known speaker count can help but can also force wrong merges/splits.

Processed on your deviceInference runs in this browser. Input/output sharing requires consent for each run. Basic run logs include device information and approximate connection location. Model and runtime files download from external hosts. Saved downloads may be cleared by your browser.

Sources & technical details
Original model
https://github.com/k2-fsa/sherpa-onnx
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
Sep 1, 2026
Browser model files
Transcripta/sherpa-onnx-speaker-diarization-wasm
Files updated
Sep 6, 2026
Artifact revision
e978abbb67d995f48e26a9950db4025e297574d5

Input

AUDIO

Sample result

Sherpa-ONNX — Speaker Diarization
Record a clip, then run inference. This does not send live audio to the model.
Advanced settings 4
Null estimates speaker count; a positive int32 integer fixes clustering count. Frontend selection maximum20 is advisory, not a model cap. A wrong fixed count can force merges or splits.
Native cosine-dissimilarity clustering threshold. The browser-prepared synthetic fixture suite uses0.7; other values require domain tuning. Adjusts how readily speaker clusters merge; domain tuning requires validation.
Frontend selection range for native minimum speech segment duration in seconds. Increasing suppresses short detections and can remove brief turns.
Frontend selection range for native minimum silence gaps in seconds. Increasing bridges short gaps and changes turn boundaries.

Output

Explore the tone of your audio

Choose a recording or try the sample to compare vocal emotion scores.

Benchmark evidence

Saved worker evidence is separate from benchmarks run on your device. It does not guarantee speed or quality on every browser.

Sherpa-ONNX — Speaker Diarization

23 configurations tested
Passed22
Failed1
TestedNov 2, 2026
Runtimesherpa-onnx 1.13.7-webai-esm1

Quality criteria

{ "maximumDiarizationErrorRate": 0.25, "collarSeconds": 0.25, "evaluationStepSeconds": 0.02, "exactSpeakerCount": true, "silence": "Zero returned turns", "scope": "Five cleared synthetic fixtures; no broad natural meeting/language/overlap certification" }

PrecisionDeviceResult
FP32CPU · WASMPassedVery fast
FP32CPU · WASMPassedVery fast
FP32CPU · WASMPassedVery fast
FP32CPU · WASMPassedFast
FP32CPU · WASMPassedFast
FP32CPU · WASMPassedFast
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32CPU · WASMPassedModerate
FP32GPU · WebGPUFailed—
Test setup & sample details
Quality gate
{ "maximumDiarizationErrorRate": 0.25, "collarSeconds": 0.25, "evaluationStepSeconds": 0.02, "exactSpeakerCount": true, "silence": "Zero returned turns", "scope": "Five cleared synthetic fixtures; no broad natural meeting/language/overlap certification" }
Tested
2026-11-02T13:20:01.101Z
Environment
Chrome 156.0.8078.79; Apple M3 Max; Darwin 27.2.0 arm64; fresh disposable disk-backed profiles, module worker, immutable remote WASM/data URLs
Runtime
sherpa-onnx 1.13.7-webai-esm1
Artifact revision
e978abbb67d995f48e26a9950db4025e297574d5
Fixture
{
  "file": "model-samples/desktop/pyannote/pyannote-long-two-speakers.wav",
  "url": "https://assets.localai.tools/samples/v1/files/e600857dcab01289b7134862a783827b689da90990ece36b5e5c030b97d285aa/pyannote-long-two-speakers.wav",
  "size": 2204200,
  "sha1": "438e52c06caa4ce65108d946d91c6ba3bbfce793",
  "sha256": "e600857dcab01289b7134862a783827b689da90990ece36b5e5c030b97d285aa"
}
Rerun when
Pinned runtime/artifact revision or preprocessing changes; Fixture checksum/config/quality gate changes; Relevant browser/WASM behavior changes
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": {
      "loadMs": null,
      "inferenceMs": null,
      "totalMs": 0.20000000298023224,
      "grade": "failed"
    },
    "failure": "Error: [WorkerError:initialization_failed] Device 'webgpu' is not supported for precision 'fp32'. Supported devices for this precision are: wasm"
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 14483.29999999702,
      "inferenceMs": 19101.29999999702,
      "totalMs": 33584.59999999404,
      "realTimeFactor": 0.5546355007893442,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 19731.90000000596,
      "totalMs": 19731.90000000596,
      "realTimeFactor": 0.5729459376079258,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.03072812291249165,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 19602.29999999702,
      "totalMs": 19602.29999999702,
      "realTimeFactor": 0.5691828031140814,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 1,
      "hypothesisSpeakers": 1,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 7917.600000008941,
      "totalMs": 7917.600000008941,
      "realTimeFactor": 0.42600952355378874,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 0,
      "hypothesisSpeakers": 0,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 67.3999999910593,
      "totalMs": 67.3999999910593,
      "realTimeFactor": 0.022466666663686435,
      "grade": "excellent"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 19386.5,
      "totalMs": 19386.5,
      "realTimeFactor": 0.5629167195978441,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 19424,
      "totalMs": 19424,
      "realTimeFactor": 0.5640055895323304,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 7175.0999999940395,
      "inferenceMs": 19907,
      "totalMs": 27082.09999999404,
      "realTimeFactor": 0.5780302342885142,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 19697.30000001192,
      "totalMs": 19697.30000001192,
      "realTimeFactor": 0.5719412736152129,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.03072812291249165,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 19420.20000000298,
      "totalMs": 19420.20000000298,
      "realTimeFactor": 0.563895250712389,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 1,
      "hypothesisSpeakers": 1,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 7823,
      "totalMs": 7823,
      "realTimeFactor": 0.4209195340453579,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 0,
      "hypothesisSpeakers": 0,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 65.5,
      "totalMs": 65.5,
      "realTimeFactor": 0.021833333333333333,
      "grade": "excellent"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 19129.59999999404,
      "totalMs": 19129.59999999404,
      "realTimeFactor": 0.55545723463315,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 18925.60000000894,
      "totalMs": 18925.60000000894,
      "realTimeFactor": 0.5495337821899771,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 9124.39999999106,
      "inferenceMs": 18938.39999999106,
      "totalMs": 28062.79999998212,
      "realTimeFactor": 0.5499054497937625,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 19079.79999999702,
      "totalMs": 19079.79999999702,
      "realTimeFactor": 0.5540112153602387,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.03072812291249165,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 19101.90000000596,
      "totalMs": 19101.90000000596,
      "realTimeFactor": 0.5546529227085556,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 1,
      "hypothesisSpeakers": 1,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 7784.70000000298,
      "totalMs": 7784.70000000298,
      "realTimeFactor": 0.4188587877648156,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 0,
      "hypothesisSpeakers": 0,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 67,
      "totalMs": 67,
      "realTimeFactor": 0.022333333333333334,
      "grade": "excellent"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 19643.59999999404,
      "totalMs": 19643.59999999404,
      "realTimeFactor": 0.5703820118685092,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 0,
      "inferenceMs": 19471.5,
      "totalMs": 19471.5,
      "realTimeFactor": 0.5653848247826797,
      "grade": "good"
    }
  },
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "passed": true,
      "diarizationErrorRate": 0.08483633934535738,
      "maximumDiarizationErrorRate": 0.25,
      "referenceSpeakers": 2,
      "hypothesisSpeakers": 2,
      "collarSeconds": 0.25,
      "evaluationStepSeconds": 0.02
    },
    "runtime": {
      "loadMs": 9135.5,
      "inferenceMs": 19445.80000001192,
      "totalMs": 28581.30000001192,
      "realTimeFactor": 0.5646385859212579,
      "grade": "good"
    }
  }
]

Something not quite right?

Tell us what happened with Sherpa-ONNX. 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.