I'm having problems with the Conversational Agent, whenever the human in line is on a noisy place, | ElevenLabs | Page 1

AI-generated Answer for justino1902

Here are the levers ElevenLabs documents for handling noisy environments that falsely trigger “listening”:

Use the WebRTC transport for your web/widget clients. WebRTC provides built‑in echo cancellation and background noise removal, improving VAD robustness without extra code. (elevenlabs.io)
Expose a mute control or implement push‑to‑talk in your client:
- Widget: enable the microphone mute button in the Widget UI (added March 24, 2025). (elevenlabs.io)
- SDKs: toggle the mic programmatically (e.g., JavaScript SDK conversation.setMicMuted(true/false); similar APIs exist in Kotlin/React Native). This lets you default to muted and only unmute while the user holds a key/button. (elevenlabs.io)
If background noise is likely, offer a text fallback:
- Enable Voice + Text in the widget so users can switch to typing mid‑conversation. (elevenlabs.io)
- For sessions that should be text‑only (e.g., a subway), enable Chat Mode. You can set textOnly via overrides or toggle “Chat only” in the agent settings. (elevenlabs.io)
Tune conversation flow to reduce noise‑caused interruptions:
- Disable “Interruptions” when you don’t want ambient sounds to stop the agent while it’s speaking. Enable them only where barge‑in is important. (elevenlabs.io)
- Adjust Turn Timeout so the agent re‑prompts after silence instead of waiting indefinitely during noisy dead air. (elevenlabs.io)
Optional: signal user activity to pause agent speech (useful if you fall back to chat in a noisy setting). The SDK’s sendUserActivity temporarily prevents the agent from speaking. (elevenlabs.io)

Notes and limits:

The WebSocket API emits VAD score events, but the documentation does not expose a server‑side parameter to change VAD sensitivity/thresholds. Mitigations today are at the transport (WebRTC), UX (mute/push‑to‑talk, chat fallback), and conversation‑flow levels above. (elevenlabs.io)

If you can share whether this is happening on web/widget vs. PSTN (Twilio/SIP) calls, I can point you to the exact setup steps for

#I'm having problems with the Conversational Agent, whenever the human in line is on a noisy place,