Hold to talk.
Deepgram listens, Sage writes, then Deepgram speaks each completed sentence. Audio is not stored.