Expand description
Barge-in (interruption) detection and handling.
Coordinates client-side VAD with jitter buffer flush and server signaling to achieve atomic barge-in with minimal latency.
When tentative mode is enabled (the default), the detector follows a
three-step duck-confirm-flush sequence:
- Duck — On the first speech frame during
ModelSpeaking, reduce playback volume instead of immediately silencing. This avoids jarring silence from false-positive VAD triggers (e.g., background noise). - Interrupt — Once speech has been sustained for
min_speech_frames, flush the jitter buffer and signal the server. - Restore — If speech stops before reaching the confirmation threshold, restore the original playback volume (false positive resolved).
Structs§
- Barge
InConfig - Configuration for barge-in behavior.
- Barge
InDetector - Barge-in detector — checks whether user speech should interrupt model output.
Enums§
- Barge
InAction - Result of a barge-in check.