How to ignore some samples in torch.nn.CTCLoss?

1

I want to ignore some noisy training samples. There is ignore_index in torch.nn. CrossEntropyLoss but how to do this in torch.nn.CTCLoss?