You're asking how to disable turn detection in the Google Realtime (Gemini Live) model — here's how to do that and what to consider. 1) Disable built-in turn detection • Set

automatic_activity_detection.disabled = true

in

realtime_input_config

. 2) Provide an alternative turn detector (recommended) • Disabling built-in VAD removes automatic end-of-speech detection. Add a separate turn-detection method such as LiveKit’s

MultilingualModel

or another custom detector. • If using LiveKit’s turn detector, you must also supply a streaming STT plugin (for example

assemblyai/universal-streaming

) because the detector operates on STT output. 3) Examples Python:

from livekit.agents import AgentSession
from livekit.plugins.turn_detector.multilingual import MultilingualModel
from livekit.agents.models import google

session = AgentSession(
    turn_detection=MultilingualModel(),  # optional alternative
    llm=google.realtime.RealtimeModel(
        realtime_input_config=google.types.RealtimeInputConfig(
            automatic_activity_detection=google.types.AutomaticActivityDetection(
                disabled=True,
            ),
        ),
        input_audio_transcription=None,
    ),
    stt="assemblyai/universal-streaming",  # required if using multilingual detector
)

TypeScript:

import * as google from '@livekit/agents-plugin-google';
import * as livekit from '@livekit/agents-plugin-livekit';

const session = new voice.AgentSession({
  turnDetection: new livekit.turnDetector.MultilingualModel(),
  llm: new google.realtime.RealtimeModel({
    model: "gemini-2.0-flash-exp",
    realtimeInputConfig: {
      automaticActivityDetection: { disabled: true },
    },
  }),
  stt: "assemblyai/universal-streaming",
});

Would you like a minimal example without any alternate detector (noting that the agent then won't detect end-of-speech), or help wiring the

MultilingualModel

+ AssemblyAI STT into your existing session? Sources: Gemini Live API plugin | LiveKit docs | Realtime model plugins | LiveKit Docs