I am interested in your code. And here I noticed that DurationDiscriminatorV2 uses in_channels and filter_channels parameters, but in the train.py file instead of taking the hidden channel hyperparameters. I would like to validate it from you @p0p4k
this is train.py

this is model.py
