LiveKit coordinates function execution with participant speech using an advanced turn detection system and agent state management. Before executing a function, the system waits until speech completion is detected, based on configurable methods such as Voice Activity Detection (VAD), Speech-to-Text endpointing, real-time LLM, or custom models. • Configure the system to use your preferred turn detection method, e.g., VAD or STT, in your agent/session settings. • Set the

minEndpointingDelay

(JavaScript) or

min_endpointing_delay

(Python) to determine how long the system should wait after it detects silence before considering speech completed. When a participant is speaking, function execution will be automatically held until the system confirms the user has finished, ensuring proper synchronization between speech input and function calls. The process is event-driven and configurable to suit your application's needs. Citations: Sources: How does end-of-utterance detection work in conversations? | VPAOptions | LiveKit Agents | Configuring turn detection and interruptions | LiveKit Docs | agents/src/pipeline/pipeline_agent.ts | livekit-agents/livekit/agents/voice/agent_activity.py