Hey there! If you're noticing audio chopping because the dispatch rates are a bit slow, a great way to smooth things out is by adding some buffering, either on the agent or client side. Here are a few ideas you can try: • Built-in Audio Buffering: You can use the InputAudioBuffer class to gather audio before sending it off to the model. Just make sure you have at least 100ms of audio buffered before you send it. • AudioSource Queue Configuration: Try increasing the internal buffer size by adjusting the

queue_size_ms

parameter (like setting it to 2000ms) when you create your AudioSource. This gives you more breathing room and helps avoid frame capture hiccups. • Custom Buffering: Another approach is to build your own buffer to hold audio frames. For example, you could add frames to a list and manage its size so you keep the most recent ones ready for processing. • System Buffering: LiveKit’s audio publishing system already has a 50ms internal buffer that queues audio, which can help even out those dispatch rates. • VAD-Based Buffering: You might also consider Voice Activity Detection (VAD) systems that come with configurable buffering options, like setting the max buffered speech duration and prefix padding. Would you like me to walk you through how to set up any of these? Just let me know! Sources: plugins/silero/src/vad.ts | livekit-plugins/livekit-plugins-silero/livekit/plugins/silero/vad.py | Resolving "InvalidState - failed to capture frame" errors in AudioSource | basics/echo_transcriber_agent.py | Publishing tracks | LiveKit Docs