Agora’s short demo uses a voice agent called Ruby to explain context-aware turn-taking. Jev reads the transcript rather than the audio: the conversation illustrates waiting while someone asks for more time, staying quiet during side-talk, and continuing through brief acknowledgements such as ‘okay’.
Original by AgoraIntegrationsIntermediate1 min 21 secPublished
Use the live transcript and conversation context to judge whether a thought is complete; speech recognition still turns audio into text before Jev can assess it.
Distinguish a request for more time from a finished turn, and words meant for someone else from words addressed to the agent; the full demo includes both decisions.
A brief ‘right’ or ‘okay’ while the agent speaks can mean ‘keep going’ rather than ‘stop’; judge the meaning of the words instead of interrupting on every sound.
Worth knowing
This is an illustrative conversation, not a reliability test or setup tutorial. The companion Agora recipe keeps only end-of-turn and interruption decisions; side-talk detection and holding-the-floor patience belong to the fuller repository demo. Jev sees text and timing context, not audio tone, and transcription or judgment errors can still affect turn-taking.