Terms of use · v2 · read before you start

Whose voice is it?

This page can copy a voice from about ten seconds of speech. That is a genuinely useful tool and a genuinely abusable one. The rules below are the deal for using it.

  1. Everything runs on this device.

    Your microphone audio, your reference clip and the speech you generate stay in this browser tab. They are never uploaded. The model files are downloaded from Hugging Face and cached locally; after that the page works offline.

  2. Clone your own voice.

    The only voice you may clone here is your own — or one where you hold explicit, documented permission from the person it belongs to, for this specific use. There is no third category.

  3. Do not use it to deceive.

    No impersonating real people. No putting words in someone's mouth. No fraud, no harassment, no defeating voice-based identity checks at a bank, a helpdesk or anywhere else.

  4. Label what you publish.

    If synthetic speech from this page leaves your machine, say that it is AI-generated. Downloads from here are named accordingly.

  5. The law on this is real, and it varies.

    Voice likeness is protected by right-of-publicity and anti-impersonation statutes in many US states, by biometric-privacy laws such as Illinois BIPA, and by transparency duties under the EU AI Act. Working out what applies to you is your job. This is a technical demo, not legal advice, and its author is not your lawyer.

  6. Responsibility sits with you.

    If you clone somebody else's voice without their permission, that is on you — not on whoever deployed this demo, not on Kyutai who trained the model, not on the authors of the open-source libraries it runs on. The page is provided as is, with no warranty of any kind, and its operator accepts no liability for anything you generate with it.

Your answer is stored in this browser's localStorage. You can withdraw it at any time from the footer.

Consent withdrawn

The lab is closed.

Nothing here runs without the terms accepted. Reload the page if you change your mind.

Advanced mode · back to the simple version

Voice Lab.

Kyutai Pocket TTS, running in your browser on WebAssembly. Give it ten seconds of your voice; it will read anything back to you.

Engine cold
Threads: checking
IN —
OUT 24 kHz
1

Voice

none loaded

! Record yourself. If the voice isn't yours, you need that person's permission — you agreed to this.

reference · 0.0 s

6–10 s of clear, continuous speech works best. Longer is trimmed to 10 s.

Kyutai's eight stock speakers. The voice bank is a 52 MB one-time download.

Saved in this browser

·

Engine

not loaded

…

Real-time factor
—
First audio
—
Generated
—
2

Script

0/600
Load the model first
3

Takes

kept in IndexedDB, on this device

No takes yet. Generate something and hit Save to takes — it stays in this browser.

·

Browser storage

—
Cache Storagepocket-tts-js-v1model weightsempty
IndexedDBvoicelabtakes + saved voicesempty
localStoragevoicelab.*consent + settings—

Terms accepted — · you agreed to clone only your own voice

Model: Pocket TTS © Kyutai, CC-BY-4.0 · ONNX export: vlapky/pocket-tts-onnx · Browser runtime: pocket-tts-js (MIT) on onnxruntime-web · Native Rust/Candle port: babybirdprd/pocket-tts

Provided as is. No warranty. Not legal advice. Synthetic speech generated here is your responsibility.