Hey! Our team switched over from google cloud over to livekit cloud for our agents. What I noticed after switching is that the silero/vad inference becomes slower than realtime. Looking at the delay, the inference can be delayed up to 15 seconds at the worst conditions. Is there any way we can optimize this? I tried prewarming the model but it still has some noticeable delay. Here is our configuration and some trace logs. def prewarm(proc: JobProcess): """Prewarm models at worker startup for faster job handling.""" proc.userdata["vad"] = silero.VAD.load( min_silence_duration=0.5, activation_threshold=0.4 ) Trace error logs we see: Inspect log Dec 19, 2025, 12817.958 PM Severity Logger livekit.plugins.silero Message inference is slower than realtime Attributes

{delay: 13.753095978001308

code.file.path: "/home/appuser/.local/lib/python3.11/site-packages/livekit/plugins/silero/vad.py"

code.function.name: "_main_task"

code.line.number: 397n

room_id: "RM_5oNDScCsHvX8"

job_id: "AJ_zbYsRjndG4zK"

logger.name: "livekit.plugins.silero"

}

Usually this kind of report is caused by insufficient hardware resources, but you say this was working previously and only fails when you switch to cloud agents, so that can't be the case. I don't see any similar reports, and it looks like your agent is successfully deployed to our cloud. The only thing I can think to suggest is that you are running the latest version of livekit and agents, 1.3.9. There's also a similar report related to the STT, https://github.com/livekit/agents/issues/3739, but again, that wouldn't explain why it previously worked.