UserState
Lifecycle state of the human participant, as reported during a live call.AgentState
Lifecycle state of the agent during a live call.SpeechEventType
Stage of a speech-to-text result, reported on STTResponse.VADEventType
Voice-activity-detection event, reported on VADResponse.EOUConfig
Configure end-of-utterance (turn-end) detection. See EOUConfigOptions for defaults.Constructor
End-of-utterance detection strategy.
'ADAPTIVE' tunes wait time to speech; 'DEFAULT' uses fixed bounds. Defaults to 'DEFAULT'.[min, max] seconds to wait for the user to resume speaking before treating the turn as ended. Defaults to [0.5, 0.8].InterruptConfig
Configure how the user can interrupt (barge in on) the agent while it speaks. If InterruptConfigOptions.interruptFadeDuration is set but the ms field is0, the ms field is derived from it. See InterruptConfigOptions for defaults.
Constructor
Signal used to detect a barge-in: voice activity, transcription, or both. Defaults to
'HYBRID'.Minimum seconds of user speech required to count as an interruption. Defaults to
0.5.Minimum number of recognized words required to interrupt. Defaults to
2.Minimum STT confidence (0-1) required to interrupt. Defaults to
0.0 (no threshold).Seconds to pause the agent after a suspected false interruption. Defaults to
2.0.Whether to resume the agent’s speech if an interruption turns out to be false. Defaults to
true.Same as InterruptConfigOptions.falseInterruptPauseDuration but in milliseconds. Defaults to
2000.Seconds to fade out agent audio when interrupted. Defaults to
0.0.Fade-out in milliseconds; derived from InterruptConfigOptions.interruptFadeDuration when that is set. Defaults to
400.RealtimeConfig
Configure a realtime (low-latency) model session. See RealtimeConfigOptions.Constructor
Realtime processing mode. See RealtimeConfigMode. Defaults to
null (provider default).Output modalities the model should produce (e.g.
['text', 'audio']). Defaults to null (provider default).