It translates the conversation in front of you as it happens. And when you're done talking, you already have a record with both languages side by side.
Preparing to submit to the App Store.
It works out who spoke on its own. No button to press back and forth each time someone talks.
Two recognizers run side by side. Which one spoke is decided from the shape of the script and how sure each one is.
Opening the AirPods microphone drops playback to phone-call quality, so listening is done with the iPhone itself.
The screen splits in two, with the top half turned around to face the other person — for places where speaking aloud isn't easy.
The conversation becomes a record with the original and its translation, line by line. Nothing to transcribe afterwards.
It uses the on-device speech recognition and translation built into iOS. What's said is never sent to our servers.