Kokoro Inno · reference tuner + Kokoro-82M

Clone a voice, then let it speak.

Drop in 3–30 seconds of one speaker. The tuner turns it into a stock Kokoro voice pack, and Kokoro-82M reads your text with it. Nothing leaves the page — the models download once and run on your CPU.

tuner model · Kokoro-82M · reference implementation

1

Reference voice

2

Speak

Ready when you are.

First run downloads ~140 MB of models: the tuner (24 MB), its speaker encoder (30 MB) and Kokoro in 8-bit (~86 MB). They are cached by the browser afterwards.