Looking for collaborators | Building a KWS library/model on top of Speech Brain uaing Wav2Vec 2.0

This paper has open-source code GitHub - qute012/Wav2Keyword: Wav2Keyword is keyword spotting(KWS) based on Wav2Vec 2.0. This model shows state-of-the-art in Speech commands dataset V1 and V2.

The code is hard to read and hard to deploy.
I want to reproduce the results of this paper and write clean code that extends to other languages. The results of the paper is amazing. I changed some code to run a KWS system trained on Urdu and it was getting good accuracy with only a few samples! This is unprecedented.

Would love some help from the SB team!

Hi, which kind of help would you need?

I am either looking for collaborators to fast track this because I need it for Urdu - might as well do it for all languages.

If that isn’t possible - I am curious if it makes sense to do this in SB and any pointers in getting started.

Hi, if you know a bit SpeechBrain. It would be extremely simple to do it. But you have to familiarize yourself a bit with it first. Have a look at the wav2vec 2 tutorial to see how it is integrated to SB. SpeechBrain

It should be fairly simple.