Press Start and talk. What's heard shows up here.
The translation shows up here, line by line.
Speech is turned into text by your browser's speech service (Google, in Chrome), so the audio is sent there. Recognising who's talking happens on this device only.
Saved only in this browser on this device, never in the page itself. Get a key at console.anthropic.com. If Claude fails, that line falls back to Google.
Auto learns each person's voice as they talk and matches every new line to it, adding people as it hears them. Voice matching runs on this device; the first time the page opens it downloads about 40 MB, which is then kept on this device. Short replies are matched to someone already heard (under half a second goes to whoever spoke last). Tap a line's label to fix a mistake; it learns from that. If the mic stops working on your phone, switch to Manual. Voice model: WeSpeaker ResNet34-LM (CC BY 4.0).