hexgrad

/

Kokoro-82M

like 159

Hello, there isn't training code for Kokoro right now, and voice cloning is not yet supported (see Philosophy).

However, Kokoro uses a StyleTTS2 architecture, which is an MIT licensed project, w/ training code if you would like to roll your own model: https://github.com/yl4579/StyleTTS2

There are other open source TTS projects as well, with various licenses and tradeoffs, such as:

• E2/F5 TTS
• XTTS v2
• GPT SoVITS
• MaskGCT
• PiperTTS (if you want speed over quality)