How to verify if masking is appropriately applied? LSTM performance drop with short sequences

Question

I am trying to understand how to apply masking and make sure that my output is masked. I use the following model to mask the inputs:

def tdcnn2d_bilstm_mask()