There are several SIP-specific and general settings in LiveKit you can adjust to optimize voice interruption detection and VAD responsiveness, minimizing the delay when a SIP user interrupts the bot: • VAD configuration: Adjust parameters like

minSpeechDuration

,

activationThreshold

, and

minSilenceDuration

for more sensitive or responsive detection. Lowering

activationThreshold

and

min_endpointing_delay

can help detect interruptions faster, but may increase susceptibility to background noise. • Interruption handling: Settings such as

allowInterruptions

,

interruptSpeechDuration

, and

interruptMinWords

can be fine-tuned. A lower

interruptSpeechDuration

or

min_interruption_duration

will make the system react sooner to user speech. • Turn detection modes: For SIP, using the

"vad"

mode is generally recommended, but if available,

realtime_llm

may provide better semantic responsiveness. Adjust

min_endpointing_delay

(e.g., to 300-400ms) for quicker turn transitions. • Noise cancellation: If you're using LiveKit Cloud, activating noise cancellation can improve VAD accuracy and responsiveness in SIP scenarios. Example configuration for faster interruption response:

turn_detection = "vad"
min_endpointing_delay = 0.3
allow_interruptions = true
interruptSpeechDuration = 0.05
activationThreshold = 0.4

For best results, start by lowering

min_endpointing_delay

and

activationThreshold

incrementally and enable noise cancellation if available. Monitor the system’s behavior to avoid reactiveness to background noise and adjust accordingly. Sources: Configuring turn detection and interruptions | LiveKit Docs | livekit-agents/livekit/agents/voice/agent_session.py | VPAOptions | LiveKit Agents | How does end-of-utterance detection work in conversations?