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.

Kokoro 1

English text in. Speech out. Generated on your device.

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

Turn short English text into spoken audio using a selected voice. Returns 24 kHz mono speech for narration, previews, and reading text aloud.

Created by
Hexgrad
License
Apache-2.0
Model released
Jan 27, 2025
Source repository created
Dec 26, 2024
Model weights · FP32
325.5 MB

Best for

  • Local narration and accessibility prototypes.
  • Generating short English voice prompts without sending text to an inference server.

Keep in mind

  • The bundled phonemizer and advertised voices support English only.
  • Names, acronyms, numbers, and unusual punctuation may be pronounced incorrectly.
  • Long text is truncated to the model context; split long passages into sentences.
  • Synthetic speech can sound unnatural and must not be used to impersonate a person or mislead listeners.

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
hexgrad/Kokoro-82M
Release verification
Verified
Release evidence
Upstream version table dates v1.0; earlier v0.19 dates do not apply.
Source updated
Apr 10, 2025
Browser model files
onnx-community/Kokoro-82M-v1.0-ONNX
Files updated
Feb 8, 2025
Artifact revision
1939ad2a8e416c0acfeecc08a694d14ef25f2231

Input

TEXT → AUDIO
55 characters · 1 clips

Sample result

Kokoro 82M v1.0 — Text to Speech

Blank lines create separate clips · Up to 2,000 characters per clip.

Settings

Selects a preset synthetic speaking voice. Each listed value selects a different voice identity. There is no higher or lower quality ordering; pronunciation and sound vary by voice and text.
Changes the generated speaking rate. Lower values speak more slowly and usually produce longer audio. Higher values speak faster and may reduce intelligibility. A value of 1 uses the normal rate.
Not benchmarkedEstimated runtime

Output

Ready to give your words a voice

Enter text, choose a voice when available, and generate speech. Listen here or download a WAV file.

Benchmark evidence

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

Kokoro 82M v1.0 — Text to Speech

16 configurations tested
Passed10
Failed6
TestedSep 10, 2026
Runtimekokoro-js (Transformers.js adapter) 1.2.1 / 3.5.1

Quality criteria

Finite, non-silent 24 kHz audio of plausible duration, with waveform similarity to the checked-in fp32/WASM reference.

PrecisionDeviceResult
FP32GPU · WebGPUAcceleratedPassedFast
FP16GPU · WebGPUAcceleratedPassedFast
Q4GPU · WebGPUAcceleratedPassedFast
Q4F16GPU · WebGPUAcceleratedPassedFast
FP32CPU · WASMPassedSlow
FP16CPU · WASMPassedSlow
UINT8CPU · WASMPassedSlow
Q4CPU · WASMPassedSlow
Q4F16CPU · WASMPassedSlow
Q8CPU · WASMPassedSlow
Q8F16CPU · WASMFailed—
Q8F16GPU · WebGPUFailed—
Q8GPU · WebGPUFailed—
UINT8GPU · WebGPUFailed—
UINT8F16CPU · WASMFailed—
UINT8F16GPU · WebGPUFailed—
Test setup & sample details
Quality gate
Finite, non-silent 24 kHz audio of plausible duration, with waveform similarity to the checked-in fp32/WASM reference.
Tested
2026-09-10T16:36:45.442Z
Environment
Not recorded
Runtime
kokoro-js (Transformers.js adapter) 1.2.1 / 3.5.1
Artifact revision
1939ad2a8e416c0acfeecc08a694d14ef25f2231
Fixture
{
  "file": "kokoro-82m-v1.json",
  "url": "https://f005.backblazeb2.com/file/hubters-webai-public/assets/benchmarks/kokoro-82m-v1.json",
  "size": 338,
  "sha1": "9b670566697f3b67757bbbb2f15a20512540258a",
  "license": "CC0-1.0",
  "reference": {
    "file": "kokoro-82m-v1-reference.wav",
    "url": "https://f005.backblazeb2.com/file/hubters-webai-public/assets/benchmarks/kokoro-82m-v1-reference.wav",
    "size": 158444,
    "sha1": "fd763f81add8aa7fe892dea39117072a35a31fe0",
    "samplingRate": 24000
  }
}
Rerun when
artifact revision change; runtime version change; fixture or reference waveform change; quality-gate change; browser/backend behavior change
Per-case quality & failure details
[
  {
    "precision": "fp32",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "envelopeCorrelation": 1,
      "durationRatio": 1
    },
    "runtime": {
      "totalMs": 47807.299999952316,
      "loadMs": 41754.4,
      "inferenceMs": 4349.5,
      "secondsPerAudioSecond": 1.318,
      "grade": "slow"
    }
  },
  {
    "precision": "fp32",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "envelopeCorrelation": 1,
      "durationRatio": 1
    },
    "runtime": {
      "totalMs": 42450.299999952316,
      "loadMs": 40312.9,
      "inferenceMs": 464.9,
      "secondsPerAudioSecond": 0.141,
      "grade": "excellent"
    }
  },
  {
    "precision": "fp16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "envelopeCorrelation": 0.9994,
      "durationRatio": 1
    },
    "runtime": {
      "totalMs": 30141.700000047684,
      "loadMs": 23716.8,
      "inferenceMs": 4484.7,
      "secondsPerAudioSecond": 1.359,
      "grade": "slow"
    }
  },
  {
    "precision": "fp16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "envelopeCorrelation": 0.9703,
      "durationRatio": 1
    },
    "runtime": {
      "totalMs": 25218.600000023842,
      "loadMs": 22822.5,
      "inferenceMs": 473.2,
      "secondsPerAudioSecond": 0.143,
      "grade": "excellent"
    }
  },
  {
    "precision": "q4",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "envelopeCorrelation": 0.9069,
      "durationRatio": 1
    },
    "runtime": {
      "totalMs": 44154.299999952316,
      "loadMs": 37887.5,
      "inferenceMs": 4511.3,
      "secondsPerAudioSecond": 1.367,
      "grade": "slow"
    }
  },
  {
    "precision": "q4",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "envelopeCorrelation": 0.895,
      "durationRatio": 1
    },
    "runtime": {
      "totalMs": 39691.799999952316,
      "loadMs": 37563.2,
      "inferenceMs": 470.7,
      "secondsPerAudioSecond": 0.143,
      "grade": "excellent"
    }
  },
  {
    "precision": "q4f16",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "envelopeCorrelation": 0.9089,
      "durationRatio": 1
    },
    "runtime": {
      "totalMs": 31990.400000095367,
      "loadMs": 25810.3,
      "inferenceMs": 4515.3,
      "secondsPerAudioSecond": 1.368,
      "grade": "slow"
    }
  },
  {
    "precision": "q4f16",
    "device": "webgpu",
    "status": "pass",
    "quality": {
      "envelopeCorrelation": 0.8906,
      "durationRatio": 1
    },
    "runtime": {
      "totalMs": 24406.899999976158,
      "loadMs": 22052.7,
      "inferenceMs": 472.9,
      "secondsPerAudioSecond": 0.143,
      "grade": "excellent"
    }
  },
  {
    "precision": "q8f16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "kokoro-js rejects the q8f16 dtype"
  },
  {
    "precision": "q8f16",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "kokoro-js rejects the q8f16 dtype"
  },
  {
    "precision": "q8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "envelopeCorrelation": 0.688,
      "durationRatio": 1.0303
    },
    "runtime": {
      "totalMs": 22718,
      "loadMs": 14929.6,
      "inferenceMs": 5986,
      "secondsPerAudioSecond": 1.76,
      "grade": "slow"
    }
  },
  {
    "precision": "q8",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "generated audio failed the reference-envelope quality threshold"
  },
  {
    "precision": "uint8",
    "device": "wasm",
    "status": "pass",
    "quality": {
      "envelopeCorrelation": 0.8494,
      "durationRatio": 1.0152
    },
    "runtime": {
      "totalMs": 33196.200000047684,
      "loadMs": 26865.8,
      "inferenceMs": 4562,
      "secondsPerAudioSecond": 1.362,
      "grade": "slow"
    }
  },
  {
    "precision": "uint8",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "generated audio contained invalid samples"
  },
  {
    "precision": "uint8f16",
    "device": "wasm",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "kokoro-js rejects the uint8f16 dtype"
  },
  {
    "precision": "uint8f16",
    "device": "webgpu",
    "status": "fail",
    "quality": "Not embedded",
    "runtime": "Not embedded",
    "failure": "kokoro-js rejects the uint8f16 dtype"
  }
]

Something not quite right?

Tell us what happened with Kokoro 1. 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.