Turn detection and adaptive interruption

I have self-hosted the livekit server.
However, I am finding the solution for the noise cancellation for the self-hosted server. Livekit cloud servers are using BVC which gives very good results. for self-hosted, VAD is available. but that does not give much more accurate results.
I need any solution for this if there is anything available

in my LiveKit dashboard I’ve added webrtc (free), DTLN (free) and the best is AI-acoustics. (paid)

WebRTC is sometimes better and sometimes makes things worse as it cuts off audio.
DTNL is free and works but nothing special. But ai-coustics if you’re happy to pay, is worlds better.

This is based on my experience.

Troy

Hey @Rushi_Patel have you considered using ai-coustics standalone VAD 2.0 or VAD Voice Focus 2.0? It’s a good timing as we have just released a temporary solution before the official update, where you can use our VAD models as well as noise cancellation and voice enhancement models within self-hosted environment. I have updated people in several topics Today, as there was a growing demand for that lately. You can check out more details here: Quail Voice Focus 2.2 L - #4 by Pawel_Lach and I am also happy to assist you if you need any help with that.

So, do I have to pay for this? and if yes, then what will be charges?
Also at what level does this provide noise cancelation?

Let me tell you my need and expectations.
I am facing a problem- when user is talking with the agent in a crowded place or someone near the user is speaking, the agent gets distracted easily and it starts listening to them and does not speak at all, as it is focusing to the background voices.

will these models help me to solve this problem??

You can get the SDK key for free from our website https://ai-coustics.com/ and use it for a month with no limitations - we also do not ask for credit card details etc. If you like it, after a month you can either get a subscription or we can setup a custom plan for you.

Answering your questions - yes, this is the exact case we are working on, such as focusing on the foreground speaker, cancelling the background noise, and ignoring the interfering speech.

I would suggest trying out Quail VF 2.2 for the audio enhancement and VAD 2.0 or even VAD Voice Focus 2.0 on the VAG gateway. Feel free to play around with the sample files either directly on our website, or you can also bring your own audio in the Dev platform playground: https://developers.ai-coustics.com/dashboard/sdk/playground