An experiment in voice AI that follows who is speaking, learns names as you talk, and knows when to join in—or just listen.
Set the number of people in Activity & settings, then start the conversation and allow microphone access.
Take turns saying a few words. Each voice gets an orb; names appear as Frank picks them up. Click a name to correct it.
Try “Frank, what should we do this weekend?” He can join open questions too. A name with “?” is a guess.
This is a live prototype: voices and names can be mistaken. Audio is processed by Speechmatics and conversation text by OpenAI. Each new conversation starts fresh. Sessions last five minutes; at four minutes, you can choose to continue.