Text, out loud.

Yes. The voices run on your own machine, so there is no per-word cost for us to recover and nothing to gate. You do not need an account and we do not see the text you type.

Yes. The Supertonic model behind these voices is licensed under the BigScience Open RAIL-M licence, which permits commercial use but attaches conditions on how the output may be used. In short: do not impersonate a real person without their consent, and do not pass generated speech off as a recording of someone. The full list is in our Terms.

Only for quality and patience. The model gets less reliable over very long passages and a single long run gives you no feedback while it works. The desktop app handles long-form reading properly, a paragraph at a time.

Same engine and the same weights. The app offers the full set of ten voice styles where this page offers four, adds an expressive engine with voice cloning, and can read text from any app rather than only a text box.

All 31 supported languages use the same model. Supertonic infers pronunciation from spelling rather than needing a pronunciation dictionary per language, so Spanish, German, Korean and the rest work out of the box with any of the voices.

The voice engine is about 145MB and runs entirely on your machine, so it has to arrive before it can speak. Your browser caches it, so every later visit starts immediately and works with no connection at all.