seorim0/DCCRN-with-various-loss-functions

Format of .npy files

Closed this issue · 4 comments

cherif54 commented

Hi, the datasets in the code are loaded with .npy files. What type of information do they contain? Is it just the path to each individual file?

seorim0 commented

Hi~ It is a bundle of clean speeches used as target (label) and noisy speeches used as input. Not the path, but the data itself.
So, the shape of .npy files is [total data number, 2 (input & target), wav length (chunk size)]

You can also load and use only the path like the code below.
https://github.com/seorim0/NUNet-TLS/blob/main/dataloader.py

cherif54 commented

Thanks. I'm running through this error "No module names estimate" in train.py. I think that file is missing from the repo.
image

cherif54 commented

I resolved this with the tools.py. However, there is a "write_on_tensorboard" module that is missing.

seorim0 commented

I uploaded it! Sorry, I forgot.
The output is written to the tensorboard during training, and you can edit write_on_tensorboard.py if you have additional outputs you would like to add. Or you can delete that part.