The Problem about Data Preprocessing

[ConstantineWayne]

It is an excellent work. But I have questions about the preprocessing of the datasets. As you mentioned, I followed the way you process the data in OGM_GE. And It turn out to be that, the accuracy performance of your proposed model dropped to around 69% on CREMA-D dataset. I wonder whether it is the problem of data preprocessing, because I notice that the preprocessing used in this code is different from the one you used in the OGM_GE, for example, in OGM_GE, you save the audio file as ".wav", but in this code, you load the audio file in the form of ".pt". If it is the problem of data preprocessing, could you provide the pipeline of preprocessing and the corresponding datasets file? Or, if it is not the problem of data preprocessing, could you give the more detailed hyperparameters settings? I'm looking forward to your response, and I'm eager to follow your track to do some improvement!