Free AI Text to Speech Generator

For English, no — the model and the pronunciation both run in your browser, and once the model has loaded the tool works with no network connection at all. For the other seven languages the text you type is sent to our server, which works out how the words are pronounced and sends back a short phonetic string. We do not store it. In every language the speech itself is generated on your own device and the audio is never uploaded.

About 93 MB for English, or 102 MB for another language — the extra is the pronunciation engine those need. It downloads once and is cached by your browser, so the second visit starts immediately. That download is the trade for not sending your text to anyone else.

Yes. The model is Apache-2.0 licensed and the audio you generate is yours. There is no watermark and no attribution requirement on the output.

The model needs roughly 600 MB of memory while it runs, which many phones cannot spare. Rather than download the model and fail halfway, the tool checks first and says so. It works on most laptops and desktops.

Speed. Generation runs at roughly one and a half times the length of the audio on a desktop, and slower on weaker machines, so 500 characters is about a minute of waiting. A longer limit would mean waits that feel broken.

Forty-nine voices across eight languages: English (US and UK), Spanish, French, Italian, Portuguese, Hindi and Chinese. The English voices carry the quality grade the model’s own authors published, best first; no grades were published for the others, so none is shown rather than guessed.

The model has Japanese voices, but the pronunciation engine that runs in a browser cannot read kanji — it describes the characters instead of speaking them. Reading kanji needs a dictionary that only exists outside the browser, and offering Japanese that mispronounces most sentences would be worse than leaving it out.