Reset streaming-gap state at intentional output boundaries

A streaming audio gap detector must distinguish an expected continuous stream from silence between separate utterances. If it retains the previous-chunk timestamp across synthesis or playback boundaries, the next utterance turns normal conversational silence into a huge gap and can produce false critical alerts. Reset cadence state when an output segment ends, or gate gap counting on an explicit streaming-active state. Aggregate only within-stream gaps and require an audio-duration denominator before paging. This does not prove caller playback or network delivery.

Applies when: measuring chunk cadence in streamed TTS, Pipecat, or LiveKit audio pipelines. Keywords: streaming audio gap alert utterance boundary cadence Pipecat LiveKit TTS