mesm-tts
Copyright 2026 mesmer.tools

This product is licensed under the Apache License, Version 2.0 (see LICENSE).

The models in this repository are new models (students) trained on audio rendered by the following Apache-2.0 models,
and the Heart decoder descends from Paradee-8M v1.1's weights. Please keep these credits when you redistribute them:

- Kokoro-82M by hexgrad (Apache-2.0), voice af_heart: teacher of Heart.
  https://huggingface.co/hexgrad/Kokoro-82M
- Paradee-8M v1.1 by Sahil Mahendrakar (Apache-2.0): the starting point for Heart, and its training code.
  https://huggingface.co/sahilmahendrakar/Paradee-8M-v1.1  https://github.com/sahilmahendrakar/paradee
- Phonon-31m-en by Gradium (Apache-2.0), with its built-in voice: teacher of Wren.
  https://huggingface.co/gradium/phonon-31m-en
  Phonon builds on Pocket TTS and the Mimi codec by Kyutai. No Phonon, Pocket TTS or Mimi weights are included here.
- gradium/phonon-voices (Apache-2.0, revision 0ee17205): the voices Everly, Hadley, Bryony, Maeve, Miles, Trevor, Edwin and
  Hugh, rendered by Phonon-31m-en: teachers of the students of the same names.
  https://huggingface.co/gradium/phonon-voices
- misaki by hexgrad (Apache-2.0): the grapheme-to-phoneme front end. web/*/frontend/ contains a JavaScript port of its
  English G2P and its lexicons (us_gold / us_silver), re-encoded.
  https://github.com/hexgrad/misaki
- kokoro.js phonemize.js by hexgrad (Apache-2.0): web/v2/frontend/kokoro-phonemize.mjs and web/espeak.mjs.

Not included: eSpeak NG (GPL-3.0-or-later). The optional fallback for words outside misaki's lexicon loads it at run time
(the `phonemizer` npm package in a browser or Node; espeakng-loader through misaki in Python). Review its licence if you
ship that fallback.

The voices are synthetic. Do not use them to impersonate a real person or to mislead listeners about who is speaking.