Kokoro TTS

Free online text-to-speech in your browser

Online TTS that stays free and private: generate natural English speech locally with the 82M Kokoro ONNX model. American and British voices via WebGPU or WASM — no API key, no cloud upload. Install as a PWA for offline use after the first model load.

Engine

Loading UI

Starting the on-device TTS workspace…

How it works

Free online TTS in three steps: confirm load → write English → generate on-device. Nothing leaves your browser; install the PWA for offline use after the model is cached.

  1. Confirm load (once)

    Pick Auto, WebGPU · FP32, or WASM · Q8, optionally choose a local folder, then click Load model. Weights (~80–300MB) download into Cache Storage and — if you linked a folder — also onto disk for later visits. Nothing starts until you confirm.

  2. Write or paste English

    Use a sample chip or enter your own script (English only). Pick an American or British voice and set speaking speed between 0.5× and 2.0×. Punctuation improves natural pauses.

  3. Generate on your device

    Press Generate speech. Longer text is split by sentence with live progress. Synthesis runs in the browser via WebGPU or WASM — play the result or download a WAV. No API key and no upload. After install + first cache, the free online TTS app works offline.

Frequently asked questions

Straight answers about this free online TTS — privacy, offline/PWA use, voices, and model size.

Is this a free online TTS?

Yes. Kokoro TTS is a free online text-to-speech app that runs in your browser. There is no signup, no API key, and no usage quota — synthesis happens on your device.

What is Kokoro TTS?

Kokoro TTS is a free browser app for English text-to-speech. It runs the open Kokoro 82M ONNX model entirely on your device — no account and no API key. You can also install it as a PWA for faster return visits and offline use once the model is cached.

Does my text leave the device?

No. Text is converted to speech locally in the browser. Nothing is uploaded to a speech API. Model weights are fetched from Hugging Face (or a folder you choose) and then stay on your machine.

Can I use it offline?

Yes, after you install the PWA (or visit once online) and load the model. The app shell is cached by a service worker; model weights stay in Cache Storage or your linked local folder. A banner appears when a new app version is available so you can refresh.

Which languages and accents are supported?

English only, with American and British voice packs. Other languages are not supported in this app.

Do I need WebGPU?

No. Choose Auto (detects WebGPU, otherwise WASM), WebGPU · FP32 for speed/quality when supported, or WASM · Q8 for a smaller download that works without WebGPU.

How large is the model download?

Expect roughly 80–300MB on first load, depending on the ONNX build (Q8 is typically smaller than FP32). After that, browser Cache Storage and an optional local folder can reuse the weights — when a folder is linked, both are written.

Can I save audio?

Yes. After generation you can play the result in the page and download a WAV file.