Training on a different dataset

[DWhettam]

What's the best way to go about training this model on a different dataset (EPIC-KICTCHENS audio)? Is there a particular format I need to store the data in, or can I just insert replace your dataloader with my own custom dataloader? There's a lot of references to the config file making it difficult to determine exactly how I should go about this. Any suggestions would be really appreciated, thanks!

[kkoutini]

Hi Dan, if this is still relevant, I've added multiple datasets in the new repo https://github.com/kkoutini/cpjku_dcase20/tree/master/configs/datasets