On this page

Interaction Mode: Full-Duplex, Listen-Only, Walkie-Talkie

2026-08-25

Scenario Overview

When interacting with the AI agent, there are three interaction forms that can be configured through the AdvancedConfig.CommunicationMode parameter when creating an agent instance:

Interaction FormUse CaseCommon Scenarios
Full-Duplex Mode (Default)Free conversation scenarios where users can speak at any time and the AI responds instantly.AI voice call
Listen-Only ModeContinuous speech-to-text scenarios where AI listens without responding.Cloud-based speech recognition, AI interviews, user content summarization
Walkie-Talkie ModeOn-demand short-turn interaction where users press to speak and release to receive AI responses.Temporary speaking in online meetings, privacy protection in AI hardware

AI Agent Behavior Differences

DimensionFull-Duplex ModeListen-Only ModeWalkie-Talkie Mode
User Speech Start/End ConditionAutomatically detected by VAD segmentationAutomatically detected by VAD segmentationStarts with StartListening, ends with StopListening; typically within 30 seconds
AI Speech Trigger✅ AI automatically replies after user finishes speaking
✅ Proactively speak via SendAgentInstanceTTS
✅ Proactively speak via SendAgentInstanceTTS✅ AI automatically replies after listening ends
✅ Proactively speak via SendAgentInstanceTTS
ASR CallbackResult is sent after user finishes speakingEvery result matching segmentation rules is sent; a long listening session may generate multiple callbacksResult is sent after StopListening
LLM RequestRequests LLM to generate a replyDoes not request LLMRequests LLM to generate a reply
TTSBroadcasts AI reply contentDoes not automatically broadcastBroadcasts AI reply content
SubtitlesSent normallySent normallySent normally

Interaction Mode Switching

  • Full-Duplex Mode (0) and Listen-Only Mode (1) can be switched between each other. The switch takes effect from the next user speech.
  • Walkie-Talkie Mode (2) does not support switching to other modes.

Implementation Steps

Prerequisites

Implement Different Interaction Forms

Previous

Controlling AI Agent Speech Emotion

Next

Getting MetaInfo During AI Broadcast

On this page

Back to top