It is necessary to update WebRTC VAD to a new version

[AlexeyBoiler]

Hello. It is necessary to update WebRTC VAD to a new version since its developers have greatly increased the quality. They already started using 4-layer biLSTM.
look at the history of commits everything has changed a lot.
Take the latest version of WebRTC with changes in RNN VAD
https://chromium.googlesource.com/external/webrtc/+/3a99ebefbb5e3faf33055a63e11eadb69c6ddc6f
Compile and extract VAD separately, it is located here https://chromium.googlesource.com/external/webrtc/+/3a99ebefbb5e3faf33055a63e11eadb69c6ddc6f/common_audio/vad/
precision | recall for the better is evident on RNN VAD using biLSTM model.

Add multichannel there. There is already a concept of batch inference

[atoultaro]

I did not find any RNN-based VAD on the links you offered. The latest version (Feb 25, 2022) still has the same old GMM-based VAD.

[atoultaro]

rnn_vad should be here instead:
https://chromium.googlesource.com/external/webrtc/+/3a99ebefbb5e3faf33055a63e11eadb69c6ddc6f/modules/audio_processing/agc2/rnn_vad/