I am using Livekit self hosted with gpt-4o-realtime to build conversation search assistant.
We are facing issues with voice isolation. Tried semantic and normal noice cancellation configurations. The LLM is picking up lot of back grounds text unless i try it in a isolated room.
Is Anyone facing similar issues ? and is there any best practices we could do to handle this.
Tried Krisp in the android app (inbound), tried Livekit BVC but this is not supported in self hosting and it is not that effective too.
it works correctly with a near-field microphone? The issue is that it picks up background and the model responds to unintended text?