Text to speech

Turn English text into natural speech with AI, in 28 American and British voices. It plays as it goes and saves as WAV. It all happens in your browser, so your text stays on your device.

Private 28 voices WAV
0 characters

Spoken on this device, never uploadedManage models

How to turn text into speech

Short answer

Paste English text, pick a voice and a speed, and press Speak. It starts talking after the first sentence is ready and keeps going while the rest is made. When it is done, save it as a WAV file.

  1. Paste your text

    Up to 50,000 characters of English: an article, a chapter, a script. Numbers, money, times and years are read the way people say them.

  2. Choose a voice

    Eleven American women, nine American men, four British women and four British men. Heart and Bella are the most natural.

  3. Speak, then save

    The first time, the AI model (Kokoro, 92 MB) and its runtime download once and stay on your device. Each sentence is made on your processor and played as it is ready.

Voices

Kokoro's authors grade their voices by how well they turned out; the grade is shown next to each name. A and B voices sound close to a real person; C and D voices are clear but flatter. Only English is offered for now: Kokoro has voices for other languages, and they will be added once they are checked.

Where does the text go?

It stays on your device. The voice model downloads from this site once and runs right in your browser.

Models and licences

Voice modelKokoro-82M v1.0 by hexgrad, Apache-2.0, in onnx-community's 8-bit export (92 MB)
PronunciationeSpeak NG, GPL-3.0, built for the browser. Its licence and complete source code are published on this site, and you may use, share and change it under the GPL
Runtimeonnxruntime-web (MIT)

Specs and browser support

LanguageEnglish (American and British)
OutputWAV, 24 kHz, 16-bit mono
Speed0.5× to 2×. Making the speech takes roughly as long as playing it on one processor core, and less with several
BrowsersCurrent Chrome, Edge, Firefox and Safari on a computer
PriceFree, no watermark, no account

Questions

No. The voice model is downloaded to your browser and runs there. Your text stays on your device.

Kokoro's weights are Apache-2.0, which allows commercial use. The GPL of eSpeak NG covers the program, not the sound it helps make. Check the terms that apply to your own text.

Kokoro's English voices are by far its best. Its other languages have fewer, weaker voices; each will be added after it has been listened to and checked.