Hi I’ve worked with LiveKit voice agents and containerized workers at scale. The lag after 2–3 concurrent calls plus the SIGUSR1 / non-zero exit issues usually point to CPU saturation, audio pipeline blocking, or worker concurrency limits inside the container. For your main need (publishing TTS or hold music without spawning an agent session), a clean approach is to use a lightweight publisher service that joins the room as a hidden participant and streams PCM/Opus audio (TTS or music) via server SDK rather than ingress/ffmpeg, which tends to be unstable in this use case. I’ve implemented similar hold-bot and queue playout systems for SIP flows and can help you design a stable architecture (scalable workers + dedicated audio publisher microservice). Happy to take a quick look at your setup and share a working approach or sample structure. @delightful-carpenter-74154