matijama / field-recording-segmentation Goto Github PK

Hi!

We got the task to use your code for an assignment at university, but there are some areas where we do not know exactly how to deal with.

Our initial aim was using your tensorflow code, together with the pretrained model, to label audio excerpts using the file label_file.py.
When trying to do that we got three errors:

• Unexpected keyword arguments: Function calc_features() is called by get_input_spectrogram() with a keyword argument do_mel - but in calc_features() there is no such keyword argument and it's not clear what this argument would change.
  Similarly, function residual_model() is called with a keyword argument use_avg_pool which it does not expect.
  Now we can of course just remove these parameters but maybe these errors are an indication of some mixup in the files which have been put into this repository?
• Can't load save_path: In file label_file.py (line 54) an error is thrown: ValueError: Can't load save_path when it is None.. Most probably we are doing something wrong when calling it?

As said before, we can of course try to make it run somehow, but I was curious if maybe you have some insight, like a wrong file version which might have been put by accident on the GitHub repository?
As a side note, would you accept pull requests from our attempts to make it run?

Thanks,
Helmuth