I see that you're encountering the warning "inference is slower than realtime" in the

livekit.plugins.silero

logs. This usually means that the Silero Voice Activity Detection (VAD) model is taking longer to process each audio segment than the length of that segment, which can cause delays in real-time audio processing. There are a few common reasons this might happen: • High CPU usage on your machine • ONNX runtime not running as efficiently as it could • Hardware limitations, like CPU throttling on burstable cloud instances (for example, AWS t3/t4) To help improve this, you might consider: • Switching to more powerful, stable compute instances without burst limits • Optimizing your runtime environment to reduce latency If you'd like, I’m here to help you explore these options or troubleshoot further—just let me know! Sources: livekit-plugins/livekit-plugins-silero/livekit/plugins/silero/onnx_model.py | README.md | Troubleshooting Latency and Timeout Errors with Turn Detection on AWS