Yeah, This happens when the voice pipeline hits the LLM session limit / no fallback model, so after 3–4 mins the call drops because there’s no available model to continue. I can help fix this