still.

A QUIETER WAY TO LISTEN

Read with your ears.

Still is a free reading studio that turns text into speech on your own device. Use it to listen to a passage, hear the rhythm of a draft, or take a break from reading on screen. You can save the result as a WAV recording.

Your first reading

  1. Paste text into the editor, or choose the file import tab and open a saved HTML or TXT file.
  2. Choose a voice model. Supertonic is the default, with ten English voices. Download it once and wait for the model to become ready.
  3. Choose a voice and reading pace, then select Create audio. Playback begins as chunks finish.
  4. Pause, seek, or change playback speed while listening. Download WAV to keep the finished recording.

Choosing a model

Supertonic downloads approximately 263 MB of model weights. Kokoro Compact uses approximately 92 MB and offers 28 English voices. Kokoro Full precision uses approximately 326 MB. Runtime files and voices add to these totals. Start with Compact if your device has limited memory.

What stays on your device?

Your pasted text, imported files and generated recordings are processed in this browser. Models download from Hugging Face and are cached separately for each browser, device and website address. No model downloads until you choose to prepare one.

Reading a webpage

For this static website, open the article in your browser and copy its text here, or save it as HTML and import the saved file. Automatic URL fetching requires a separate server and is not included in the Cloudflare Pages deployment.

Offline listening

After the app and your chosen model have downloaded and worked once, cached speech generation can work offline. Keep any recording you want by downloading it before closing the page. Browser storage may be cleared or evicted, so offline availability depends on what remains cached.

If a model will not load

Open Device & storage and try CPU mode or a smaller model. Close other memory-intensive tabs. If a download was interrupted, try loading it again. Removing downloaded models and voices clears the cache so you can start fresh, but requires another download.

Limits and model terms

A reading can contain up to 40,000 characters. Saved HTML and TXT files can be up to 3 MB. Device memory and browser support affect performance. Each model has its own usage terms; see the credits and model licenses before using recordings commercially.