You can integrate Silero VAD with the multimodal OpenAI agent in LiveKit, and both can be used concurrently with the right configuration. Key steps for integration: • Silero VAD can be used for turn detection while the OpenAI agent handles language understanding and generation. • To avoid conflicts, disable OpenAI’s built-in server-side VAD by setting

turn_detection=None

in the

RealtimeModel

configuration. • Both components can process audio in parallel—Silero VAD for voice activity detection, and OpenAI for conversational AI. This approach is supported by LiveKit’s plugin architecture, which allows you to combine different providers for different tasks within the same agent session. Example configuration:

session = AgentSession(
    llm=openai.realtime.RealtimeModel(
        turn_detection=None,  # Disable OpenAI VAD
    ),
    vad=silero.VAD.load(),
)

This setup enables both Silero VAD and the multimodal OpenAI agent to operate concurrently without interference. Sources: mcp/agent.py | examples/voice_agents/realtime_turn_detector.py | AI voice agents | LiveKit Docs | Working with plugins | LiveKit Docs