Integrating reltime lowletency voice with langgraph graph but the letency is more than 6-7 seconds,it should be less than 2-3 seconds

OSS Product Help LangGraph
,
1

Currently i am trying to integrate a realtime interview agent using livekit ,which is a webrtc based platform,for low letency.I used the func-tool approch (tool calling) for that.still getting the letency of 5-8secs.
I tried the langgraph direct integration part.It didnot work.Also the other integrations as well which didnot work as well.

Good part is, with integration with livekit gives it a feel of the output adding extra lines of (thats a good answer,okh lets move on to the next question). In my case i am developing a reltime ai agent for interview, which actually make the roleplay reverse,rather than me quering the graph ,the graph should start asking me questions here.

But the major problem is i still want to stay with livekit for these perks.Can you help me with how to integrate this in a way so that it would be helpful for me to dispath the product quick for student to help them selves.A help from you and code refreence would be helpful for me.thank you.

2

Hey @Racasekumar ! Sending through a reference application by our teammate here: https://github.com/dqbd/langgraph-livekit-agents

3

Thanks for sharing this,i have tried the similar code ,but it is not working as expected.hope you help in some other way,there we need to only change the url and paths from json,i know that part,thats not functional for this if you could help in other way it would be helpful,if any other integration maybe useful if they satisfies all the criteria,webrtc and untractivity.