Audio stream stutter/lag when talking over SIP

[vykintazo]

I am not sure if this is the right place to ask about this as I am currently using Cloud LiveKit, but I noticed that Agent using default OpenAI TTS lags very noticeably when calling over phone via SIP. For me it sounds as if audio chunks got re-encoded and because of that contain artifacts at places where chunk ended, but speech didn't. I of course don't know for sure. I've tried 2 different SIP trunk providers (Vonage and Twilio) and the results where exactly the same. Is there anything I can do to debug and resolve this? Thanks

[davidzhao]

there wouldn't be any differences if you are running over telephony or not.. the main thing that will influence latency is where you are running your agents. try to ensure the agent is close to inference servers to minimize latency

[alibabaintelligence]

Same thing here, and it's weirder because using the exact same agent and backend but on a LiveKit Sandbox Web Demo, works flawlessly

it's just on telephony SIP that generates that choppiness no matter if you're using Realtime API or any TTS

[CesarCloneOps]

Have you found a workaround? We notice it now on outbound calls to cell phones. (inbound calls and calls to google voice numbers work fine)

[multib0t]

same issue which I reported here