Yes, I can help with this. What’s happening is that audio.set_audio_enabled() usually just mutes output but doesn’t fully stop the underlying STT/TTS processing loop or stream subscriptions. So the recognizer keeps receiving frames, which leads to backlog and eventual crash. I’ve worked with real-time LLM + STT + TTS pipelines and handled similar toggle/control issues before. I can review your current implementation and help you achieve reliable programmatic on/off control without crashes. Happy to take a look at your setup and assist. @sparse-teacher-281