Speech to text, online and free

Press start, allow the microphone, and speak naturally — your words are transcribed as you talk and refreshed every couple of seconds. The first time, a small speech model downloads to your browser; after that it's cached and even works offline, and your audio never leaves your device. Pick from a dozen languages, keep dictating across pauses, and copy the finished text into an email, document, or message. It's a fast way to draft hands-free — or just to see how good in-browser speech recognition has become.

This tool runs an open-source Whispermodel locally in your browser — your audio never leaves your device. That's great for a private, quick demo, but it has real limits:

• It only types into this page — not into your editor, chat, or terminal.
• The first run downloads a model, and transcription speed depends on your hardware — it's fastest on devices with WebGPU.
• To stay lightweight it uses a small model, so it won't match a full cloud model on tricky audio, names, or accents.

Openbroca takes voice typing system-wide on your Mac: speak, and your words become polished writinginserted wherever your cursor is — editors, browsers, chat, the terminal — with a copy fallback when an app won't take direct insertion, and talk-to-translate into 28 languages along the way. It's a native macOS app, so it isn't limited to one browser tab. Learn what Openbroca does or see how it compares to Apple Dictation.

• Use a decent microphone in a quiet room.
• Speak at a steady, natural pace — no need to over-enunciate.
• Pick the language that matches your speech.
• Whisper adds punctuation and capitalisation for you — just talk normally.
• On the first run, give the model a moment to download; it's cached for next time.