Text to Speech

Free text-to-speech that runs in your browser

Turn any text into natural-sounding speech without uploading a single character. This tool runs Kokoro-82M, a high-quality open source voice model, entirely inside your browser using WebGPU, with a WebAssembly fallback for devices without it. Type or paste your text, pick a voice, choose a speaking speed, and download a WAV file. No account, no API key, no per-character fee, and nothing ever leaves your device.

The same engine drives both privacy and price. Because the model lives in your browser, we have no inference bill to recover, no logs to keep, and nothing to upsell. Once the model is cached it even works offline.

How to use it

1. Type or paste up to 2,000 characters into the text box.
2. Pick a voice. American and British, male and female, are all included.
3. Set the speaking speed if you want it slower or faster than natural.
4. Click Generate speech. The first run downloads the voice model once; after that it is instant.
5. Play it back in the built-in player and download the WAV when you are happy.

What it is good for

Voiceovers for videos and slideshows, narration for articles and scripts, accessibility read-aloud, pronunciation checks, audiobook drafts, e-learning content, and proofreading by ear. Because the audio is a plain WAV file, it drops straight into any video editor, podcast tool, or presentation. The voices are expressive enough for published content, and there is no usage cap to work around.

Private and free, by design

Cloud text-to-speech services upload your text to their servers, meter you per character, and usually require an account. This tool does the opposite. The model runs on your own hardware, so your text never leaves the browser, there is nothing to log, and there is no bill to pass on to you. The model and its weights are Apache 2.0 licensed, so the audio you generate is yours to use, including commercially.

Need the reverse? Try the audio transcription tool to turn speech back into text, also entirely in your browser. And if you are weighing this against a metered cloud service, see how it stacks up as a free ElevenLabs alternative.

Frequently asked questions

Does my text get uploaded anywhere?

No. The text you type stays in your browser the entire time. The Kokoro voice model runs locally with WebGPU (or WebAssembly on devices without WebGPU), so there is no server in the loop, nothing to log, and nothing for us to retain. You can disconnect from the internet after the model has downloaded and it still works.

Why does it download a model the first time?

For the speech to be generated on your device, the neural voice model has to live on your device. The first time you click Generate, your browser downloads Kokoro and caches it. After that, every run uses the cached copy and starts instantly. You can review or delete cached models from the 'Local models' chip in the page header.

How big is the download?

About 92 MB on the WebAssembly path and around 326 MB on WebGPU. It is a one-time download cached in your browser. On a phone the smaller WebAssembly model is used automatically.

Is the voice quality good?

Kokoro-82M is one of the best-rated open text-to-speech models available and sounds natural and expressive, well beyond the robotic system voices. It is not a clone of any specific person, but for narration, drafts, accessibility, and content it is very close to commercial cloud voices, at zero cost and with full privacy.

Which languages does it support?

This tool ships natural American and British English voices. Kokoro focuses on English, so for now we keep the voice list to high-quality English speakers rather than advertising languages that sound weak. If you need spoken output in other languages, that is on our roadmap.

Can I download the audio?

Yes. Every generation produces a standard WAV file you can download with one click and use in videos, podcasts, slideshows, voiceovers, or anywhere else. There is no watermark and no per-character fee.

Is this text-to-speech tool open source?

The voice model is. Kokoro-82M and its weights are released under the Apache 2.0 license, so anyone can inspect, self-host, or fork it. This tool packages that model into a ready-to-use browser app: no install, no Python environment, no GPU driver setup. Because everything runs locally, you get the same 'nothing leaves my machine' guarantee open source software is prized for, without doing the setup yourself.

Can I use the audio commercially?

Yes. The Kokoro model and its weights are released under the Apache 2.0 license, which permits commercial use. The audio you generate is yours to use, including in monetized and commercial projects. We make no rights claim over your output.

How long can the text be?

You can synthesize up to 2,000 characters at a time, which is roughly 300 to 350 words. For longer scripts, generate in sections and stitch the WAV files together in any audio editor. The cap keeps memory use safe, especially on phones.

Can I change the speaking speed?

Yes. The Speed slider ranges from 0.5x (slower, clearer) to 2x (faster). It changes the talking pace without distorting the pitch, so the voice still sounds natural.

Does it work on iPhone?

Yes, but more slowly. iPhones do not yet have reliable browser GPU access, so the tool falls back to the WebAssembly path, which is heavier on the CPU and memory. Keep iPhone generations to a few paragraphs at a time. For long scripts, desktop Chrome, Edge, or Brave with WebGPU is much faster.

Is it really free with no signup?

Yes. Because the model runs on your hardware, we have no inference bill to pass on. There is no account, no API key, no usage cap, and no credit card. Cloud text-to-speech services typically charge per character and require an account; this does neither.

Why is WebGPU faster than WebAssembly?

WebGPU runs the model on your graphics card instead of your CPU, which is dramatically faster for neural networks. We auto-detect WebGPU and fall back to WebAssembly when it is unavailable. The badge in the status panel tells you which path you got.