VAD Hangover
A timing policy that keeps a VAD segment open briefly after speech is no longer detected, reducing clipped word endings and fragmentation across short pauses.
Closing a segment immediately when speech energy drops can clip phonemes or split a phrase at a short pause. A hangover policy keeps the speech state active for several additional frames after the last positive decision.
A longer hangover reduces premature cuts but adds silence and latency. The correct value is therefore governed by endpointing goals rather than VAD accuracy alone. The system-level trade-off is examined in VAD, segmentation, and endpointing for real-time ASR.