Does my text get sent to a server?

No. The TTS model is downloaded once and runs entirely in your browser with WebAssembly or WebGPU. Your text never leaves your device, and there is no API key or account.

How can a text-to-speech model run in a browser?

VoiceBox uses Kokoro, an 82-million-parameter open-weight TTS model compiled to ONNX and run through Transformers.js. The model file (about 86 MB) downloads once, is cached by your browser, and every subsequent generation is offline.

Yes. It is completely free with no sign-up, no watermark, and no limit on how much text you convert. You can download the audio as a WAV file.

What voices are available?

28 voices across American and British English, male and female, including warm, neutral, and soft styles. You can preview any of them before generating.

Free tool · neural TTS runs in your browser

Text to Speech

Type anything and hear it spoken by a real neural voice model that runs entirely on your device. 28 voices, instant WAV download, nothing uploaded.

🔒 100% private — your text never leaves your browser

0 characters

Voice

Speed

1.0×

Detecting engine…

WebGPU turbo (higher quality, ~310 MB one-time download)

A real speech model, running on your machine

VoiceBox runs Kokoro, an open-weight neural text-to-speech model with 82 million parameters. Most "free" TTS sites stream your text to a server and read it back. This one does the opposite: the model downloads to your browser once, then every word you type is turned into audio locally, with no upload and no API key. That is why it keeps working even with your network disconnected after the first load.

How does this run in a browser?

The model is compiled to ONNX and executed through Transformers.js. By default it runs on WebAssembly, which works on essentially any modern browser. If your browser supports WebGPU you can flip on Turbo mode to run the larger, higher-fidelity weights on your graphics card for faster, cleaner output. The first generation does the one-time model download (about 86 MB in the default mode); after that it is cached and instant.

What it is good for

Drafting voiceovers and narration, proofreading by ear, making audio versions of articles and notes, accessibility, learning pronunciation, or just turning a script into a downloadable WAV without signing up for anything. Longer passages are split into sentences and stitched together automatically, so you can paste a whole paragraph and get one continuous clip.

Does my text get sent anywhere?

No. The model runs inside your browser tab. Your text and the generated audio stay on your device. There is no account and no server call after the model file is fetched.

Why is the first run slow?

The first time, your browser downloads the model (about 86 MB in default mode, or roughly 310 MB in WebGPU Turbo). It is cached afterward, so later runs start instantly and work offline.

Can I download the audio?

Yes. Every result can be saved as a standard WAV file you can use anywhere, royalty-free.

Which browsers work best?

Default mode works in Chrome, Edge, Firefox, and Safari. Turbo mode needs WebGPU, which is available in recent Chrome and Edge, and increasingly elsewhere.

Is the voice licensed for my use?

Kokoro is released under the Apache 2.0 license, so the generated audio is free to use, including commercially.

I build tools like this every day.

Senior full-stack engineer, available for senior or contract work, fully remote. See the rest of the lab or get in touch.