fuankarion / maas Goto Github PK

I have successfully processed ava-activespeaker Dataset and trained MAAS with forward embedding. The following results are obtained:

This result surprised me very much. The effect of the model on the training set is good, but the effect on the validation set is too poor. Why does this happen?

The training parameters I use are as follows:
python maas_train.py 13 1 3 64 0 3

Hello, it seems that the Talkies dataset is not available now, could you share it again?

The parameters for running the maas_train.py file are not given in the paper. Can you share the configuration of these parameters?

Hi,
I'm trying to run the code for a different dataset. By any chance can you please tell the format of the data (toy example) passed from the dataloader to the model.

Thanks

Thank you very much for contributing the dataset. Could I kindly request if it would be possible to provide the preprocessing code for the Talkies dataset? Thank you.