Kokoro: open-weight 82M text-to-speech inference library, README overview

What it supplies: The README describes Kokoro as an open-weight text-to-speech model with 82 million parameters, offered through an inference library, with Apache-licensed weights. Useful for: - Generate speech audio from text in Python and save it as WAV files. - Choose among several languages by matching the language code to the voice. - Use the lightweight model from personal projects up to production, as the README describes it. First action: - Install with pip install kokoro and soundfile, and install espeak-ng. - Create a KPipeline with the language code, call it with text and a voice, and write the audio at 24000 Hz. - On Windows, install espeak-ng from its releases page; on Apple Silicon set the MPS fallback variable for GPU acceleration. Prerequisites and limits: The README's examples need kokoro 0.9.4 or later, soundfile and espeak-ng, and its conda file pins Python 3.9. Japanese and Mandarin need extra misaki packages. - Quality and speed are the README's own claims; no test was made. - Language support is limited to the listed codes. - The README does not describe a voice list or hardware requirements. The README says the weights are Apache-licensed. The licence for the library code and the model card were not read here. Voices and text you synthesise carry their own rights, and the README makes no claim about them. Written from the README text only. Statements are as the README gives them; nothing was installed or run. Based on README text only; the model card, the samples, the code and the licence file were not read. No install or run was made. Source inspected: https://raw.githubusercontent.com/hexgrad/kokoro/dfb907a02bba8152ca444717ca5d78747ccb4bec/README.md ; 8355 bytes ; sha256 b50b84220668f8247a2405b249c34c7083acc24114279ff1429fd79f46a2c04b ; 8 exact pinned excerpts in the draft pin map

{
  "revision": {
    "ref": "commit dfb907a02bba8152ca444717ca5d78747ccb4bec",
    "pinned": false,
    "source_url": "https://raw.githubusercontent.com/hexgrad/kokoro/dfb907a02bba8152ca444717ca5d78747ccb4bec/README.md"
  },
  "checked_at": "2026-11-07T10:10:10.140Z",
  "usefulness": "- Generate speech audio from text in Python and save it as WAV files.\n- Choose among several languages by matching the language code to the voice.\n- Use the lightweight model from personal projects up to production, as the README describes it.",
  "source_type": "repository",
  "availability": "available",
  "source_owner": "hexgrad (declared repository namespace)",
  "known_limitations": [
    "The README's examples need kokoro 0.9.4 or later, soundfile and espeak-ng, and its conda file pins Python 3.9. Japanese and Mandarin need extra misaki packages.\n\n- Quality and speed are the README's own claims; no test was made.\n- Language support is limited to the listed codes.\n- The README does not describe a voice list or hardware requirements.\n\nThe README says the weights are Apache-licensed. The licence for the library code and the model card were not read here. Voices and text you synthesise carry their own rights, and the README makes no claim about them.\n\nWritten from the README text only. Statements are as the README gives them; nothing was installed or run.\nBased on README text only; the model card, the samples, the code and the licence file were not read. No install or run was made.",
    "Rendered README text of the exact archived capture at the pinned commit, read in full by the author; no media viewed or heard; nothing executed or installed.",
    "Source-documented steps; execution not performed. Internal affiliated review is not independent validation."
  ]
}