CANTONESE · OUR CURRENT FOCUS

Cantonese, in your own words.

For people who think in Cantonese and want to write short messages and everyday notes the way they speak. We are building towards that experience.

WHAT YOU CAN EXPLORE

Two ways to get words down.

Windows preview

Type with Jyutping

The Windows preview includes offline Cantonese Jyutping without tone numbers. Choose candidates as you type.

See the Windows page for the test release, installation notes and current limitations.

View the Windows preview
Local development prototype

Speak a short sentence

A local prototype accepts short recordings or PCM16 WAV files, up to 60 seconds. Recognition starts after recording stops; you can then edit and copy the draft.

Cantonese speech recognition is our current development priority.

How the local trial works

THE LOCAL VOICE WORKFLOW

Say it. Review it. Make it yours.

  1. Record a short thought

    Use the local trial page to record, or choose a PCM16 WAV file. Each clip can be up to 60 seconds.

  2. Stop and recognise

    The model runs on the configured computer. Audio is handled in memory by default.

  3. Edit and copy

    Check the draft, make your changes, then copy it into the app where you want to write.

WHERE DEVELOPMENT STANDS

A working local foundation.

The prototype runs the published SenseVoice-small-Yue model locally. We have not trained a new model of our own yet.

Voice is not yet integrated into the system input method, and there is no standalone voice installer. The full microphone workflow still needs hands-on testing.

Next: test real short messages and everyday notes, and use the corrections to guide development.

FOR AN ALREADY CONFIGURED COMPUTER

Open your local trial.

This address works only on a computer where the voice environment and model are already configured.

  1. Start the configured local voice service on that computer.
  2. Copy the address below and paste it directly into your browser’s address bar.

Opening this website does not start the local service, record audio or upload recordings. If your computer is not configured, follow the development updates here.