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.

Chatterbox Turbo English

Generate English speech in your browser from text and a required reference voice. Returns complete mono24kHz PCM audio.

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

Generate English speech in your browser from text and a required reference voice. Returns complete mono24kHz PCM audio.

Created by
Resemble AI
License
MIT
Model released
Dec 15, 2025
Source repository created
Dec 2, 2025
Model weights · Q4F16
2.2 GB

Best for

  • Short English narration in a verified WebGPU browser
  • Reference-conditioned speech using authorized reference recordings

Keep in mind

  • English only; output can omit, repeat or mispronounce words.
  • Use a clear single-speaker reference in the same language; voice similarity is approximate.
  • Large downloads and GPU memory use; performance and support depend on the measured browser/device.
  • Use reference voices only with appropriate permission. Generated speech must not be represented as an authentic recording.
  • The selected ONNX runtime returns a complete waveform; incremental audio streaming is unavailable.
  • Precision selects only the language-model graph; embeddings, reference encoder and speech decoder remain FP32. Quantizing those auxiliary graphs failed retained intelligibility probes.
  • Browser reference input is required; native built-in conditioning is not exported through this artifact/runtime interface. Browser output does not apply native Perth watermarking.
  • Upstream paralinguistic text tags and broad text/accent/voice-identity quality are unverified; the bounded authored suite does not certify them.
  • Current measured hardware/browser only. WASM candidates are unavailable because of kernel or speed failures.

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://huggingface.co/ResembleAI/chatterbox-turbo
Release verification
Verified
Release evidence
Official Chatterbox Turbo 350M release commit dated 2025-12-15 introduces Turbo in the public README and links its public weights. This is an explicit release announcement, not the Hugging Face repository creation date.
Source updated
Dec 15, 2025
Browser model files
ResembleAI/chatterbox-turbo-ONNX
Files updated
Dec 15, 2025
Artifact revision
d21799bd0354adb85e348b8a0442a8405110a2cf

Input

AUDIO

Sample result

Chatterbox Turbo English
Record a clip, then run inference. This does not send live audio to the model.
Advanced settings 1
Caps generated speech tokens; output may be truncated when reached. Higher values allow longer speech at greater time and memory cost. Lower values may cut speech short.

Output

Your generated speech will appear here

Enter text and choose a reference recording, or try the sample.

Benchmark evidence

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

Chatterbox Turbo English

8 configurations tested
Passed4
Failed4
TestedNov 6, 2026
Runtimetransformers.js 4.3.2

Quality criteria

Frozen three English utterances: WER<=0.35, no excessive repetition, finite audible mono24kHz PCM, clipping<=2%,1–30seconds; every case RTF<=4.

PrecisionDeviceResult
Q4F16GPU · WebGPUAcceleratedPassedFast
Q4GPU · WebGPUAcceleratedPassedModerate
FP16GPU · WebGPUAcceleratedPassedModerate
FP32GPU · WebGPUAcceleratedPassedModerate
FP32CPU · WASMFailed—
FP16CPU · WASMFailed—
Q4CPU · WASMFailed—
Q4F16CPU · WASMFailed—
Test setup & sample details
Quality gate
Frozen three English utterances: WER<=0.35, no excessive repetition, finite audible mono24kHz PCM, clipping<=2%,1–30seconds; every case RTF<=4.
Tested
2026-11-06T20:15:49.048Z
Environment
Apple M3 Max/96GiB/macOS27.2, Chrome157.0.8103.39, Transformers.js4.3.2 and pinned ONNX Runtime Web1.31.0-dev.20260914-8d85527a0. Fresh browser/profile/module worker per pair, immutable SHA-verified local model-file transport; load is not Internet download time. No competing synthesis/ASR/builds; background archive transfers active.
Runtime
transformers.js 4.3.2
Artifact revision
d21799bd0354adb85e348b8a0442a8405110a2cf
Fixture
{
  "file": "model-samples/chatterbox-turbo-standard-input.json",
  "size": 615,
  "sha1": "499356c0caa0dd612ae00fdcfd0e1fc5114a1060",
  "sha256": "56acb62653012225630fd2fbbcf605202c68f84469e46c028bd5dc5f97061b45",
  "url": "https://assets.localai.tools/samples/v1/files/56acb62653012225630fd2fbbcf605202c68f84469e46c028bd5dc5f97061b45/chatterbox-turbo-standard-input.json"
}
Rerun when
artifact revision; Transformers.js/ONNX Runtime version; fixture or quality gate; browser/backend behavior
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "passed": true,
      "scope": "1 complete authored English results; intelligibility only",
      "worstWordErrorRate": 0
    },
    "runtime": {
      "loadMs": 7765.789999999106,
      "inferenceMs": 14789.870000001043,
      "totalMs": 22555.66000000015,
      "audioSeconds": 3.24,
      "realTimeFactor": 4.5647746913583465,
      "rms": 0.028958523657733944,
      "clippedFraction": 0,
      "audioSecondsPerSecond": 0.2190688626742339,
      "grade": "slow",
      "matrixTotalMs": 22582.704999998212
    },
    "failure": {
      "message": "CPU first utterance exceeded the predefined real-time factor 4 usability gate; further utterances skipped."
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "scope": "3 complete authored English results; intelligibility only",
      "worstWordErrorRate": 0.2
    },
    "runtime": {
      "loadMs": 6720,
      "inferenceMs": 2756.9850000031292,
      "totalMs": 9476.98500000313,
      "audioSeconds": 3.24,
      "realTimeFactor": 0.8509212962972621,
      "rms": 0.026651632592519077,
      "clippedFraction": 0,
      "audioSecondsPerSecond": 1.1751968182620953,
      "grade": "good",
      "matrixTotalMs": 14572.815000001341
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "passed": true,
      "scope": "1 complete authored English results; intelligibility only",
      "worstWordErrorRate": 0
    },
    "runtime": {
      "loadMs": 8466.920000001788,
      "inferenceMs": 13876.130000002682,
      "totalMs": 22343.05000000447,
      "audioSeconds": 3.24,
      "realTimeFactor": 4.282756172840334,
      "rms": 0.029103699346706886,
      "clippedFraction": 0,
      "audioSecondsPerSecond": 0.2334944973850327,
      "grade": "slow",
      "matrixTotalMs": 22372.050000000745
    },
    "failure": {
      "message": "CPU first utterance exceeded the predefined real-time factor 4 usability gate; further utterances skipped."
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "scope": "3 complete authored English results; intelligibility only",
      "worstWordErrorRate": 0.2
    },
    "runtime": {
      "loadMs": 7602.919999998063,
      "inferenceMs": 2653.5,
      "totalMs": 10256.419999998063,
      "audioSeconds": 3.28,
      "realTimeFactor": 0.8089939024390245,
      "rms": 0.026666687036945722,
      "clippedFraction": 0,
      "audioSecondsPerSecond": 1.236103259845487,
      "grade": "good",
      "matrixTotalMs": 14699.460000000894
    }
  },
  {
    "precision": "q4",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "passed": false,
      "scope": "0 complete authored English results; intelligibility only",
      "worstWordErrorRate": null
    },
    "runtime": "Not embedded",
    "failure": {
      "message": "Error: Can't create a session. ERROR_CODE: 9, ERROR_MESSAGE: Could not find an implementation for GatherBlockQuantized(1) node with name '/model/wpe/Gather_Quant'",
      "stack": "Error: Can't create a session. ERROR_CODE: 9, ERROR_MESSAGE: Could not find an implementation for GatherBlockQuantized(1) node with name '/model/wpe/Gather_Quant'\n    at ze (https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.2:15:31333)\n    at pm (https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.2:16:22185)\n    at async uS.loadModel (https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.2:16:35074)\n    at async Jd.createInferenceSessionHandler (https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.2:16:36795)\n    at async RT.create (https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.2:11:49080)"
    }
  },
  {
    "precision": "q4",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "scope": "3 complete authored English results; intelligibility only",
      "worstWordErrorRate": 0.2
    },
    "runtime": {
      "loadMs": 6447.544999998063,
      "inferenceMs": 1736.9749999977648,
      "totalMs": 8184.519999995828,
      "audioSeconds": 3.04,
      "realTimeFactor": 0.5713733552624226,
      "rms": 0.02823716751615756,
      "clippedFraction": 0,
      "audioSecondsPerSecond": 1.7501691158502064,
      "grade": "good",
      "matrixTotalMs": 10574.42499999702
    }
  },
  {
    "precision": "q4f16",
    "device": "wasm",
    "status": "fail",
    "quality": {
      "passed": false,
      "scope": "0 complete authored English results; intelligibility only",
      "worstWordErrorRate": null
    },
    "runtime": "Not embedded",
    "failure": {
      "message": "Error: Can't create a session. ERROR_CODE: 9, ERROR_MESSAGE: Failed to find kernel for com.microsoft.GatherBlockQuantized(1) (node:'/model/wpe/Gather_Quant' ep:'CPUExecutionProvider'). Kernel not found",
      "stack": "Error: Can't create a session. ERROR_CODE: 9, ERROR_MESSAGE: Failed to find kernel for com.microsoft.GatherBlockQuantized(1) (node:'/model/wpe/Gather_Quant' ep:'CPUExecutionProvider'). Kernel not found\n    at ze (https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.2:15:31333)\n    at pm (https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.2:16:22185)\n    at async uS.loadModel (https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.2:16:35074)\n    at async Jd.createInferenceSessionHandler (https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.2:16:36795)\n    at async RT.create (https://cdn.jsdelivr.net/npm/@huggingface/transformers@4.3.2:11:49080)"
    }
  },
  {
    "precision": "q4f16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "passed": true,
      "scope": "3 complete authored English results; intelligibility only",
      "worstWordErrorRate": 0.2
    },
    "runtime": {
      "loadMs": 7151.25,
      "inferenceMs": 1387.9900000020862,
      "totalMs": 8539.240000002086,
      "audioSeconds": 3.08,
      "realTimeFactor": 0.4506461038967812,
      "rms": 0.030149302311589053,
      "clippedFraction": 0,
      "audioSecondsPerSecond": 2.219036160199548,
      "grade": "good",
      "matrixTotalMs": 10890.580000001937
    }
  }
]

Something not quite right?

Tell us what happened with Chatterbox Turbo. 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.