About LSSED wav2vec2 features

[ndisci]

I asked the LSSED data but the given data is only feature. You shared a training code with me. When I used your code for training, because of class imbalance, overfitting occured. Did you realize it before?
Avoid overfitting, I tried to create a subset from the LSSED Wav2Vec2 features to have all the classes equally distributed. But it did not work.