Should I train lossnet_train.py with my datasets ?

[ucasiggcas]

Dear,

If I use my own data,should I run lossnet_train.py ?

Thx.

[ucasiggcas]

If I directly run python senet_train.py -d folder_name -l loss_folder -o out_folder it ups bug, `Traceback (most recent call last): File "senet_train.py", line 89, in loss_saver.restore(sess, "./%s/loss_model.ckpt" % modfolder) File "D:\python\lib\site-packages\tensorflow\python\training\saver.py", line 1538, in restore

• compat.as_text(save_path)) ValueError: The passed save_path is not a valid checkpoint: ./loss_folder/loss_model.ckpt`

[ucasiggcas]

and if I run python lossnet_train.py -o out_folder it also tells me Traceback (most recent call last): File "lossnet_train.py", line 50, in <module> load_asc_data("dataset/asc/") File "D:\python\SpeechDenoisingWithDeepFeatureLosses-master\data_import.py", line 126, in load_asc_data with open('%s/meta.txt' % foldername, 'rb') as csvfile: FileNotFoundError: [Errno 2] No such file or directory: 'dataset/asc//trainset/meta.txt'

so how to solve the problem ? Thx

[ucasiggcas]

When I run the code python lossnet_train.py on ubuntu 18.04 it also have the same problem Traceback (most recent call last): File "lossnet_train.py", line 50, in <module> load_asc_data("dataset/asc/") File "/root/SpeechDenoisingWithDeepFeatureLosses/data_import.py", line 126, in load_asc_data with open('%s/meta.txt' % foldername, 'rb') as csvfile: IOError: [Errno 2] No such file or directory: 'dataset/asc//trainset/meta.txt'

so what's wrong ?

[zeynepgulhanuslu]

I've changed the code like this : "open('%s/meta.txt' % foldername, 'r') " end it worked for me:) And also you should have a meta.txt file in that folder.

[ucasiggcas]

I've changed the code like this : "open('%s/meta.txt' % foldername, 'r') " end it worked for me:) And also you should have a meta.txt file in that folder.

where is the meta.txt file ? Or how to write the file ? how to set ? No details.

[zeynepgulhanuslu]

I'm no expert but i created meta.txt file myself. I think you should create it for your dataset. The file contains wav file path and its label. For example; audio/003/012.wav car noise audio/923/42.wav white noise audio/234/11.wav white noise .. etc. It could contain more than one file for a noise. I've put meta.txt files in trainset and valset folders. At least i could train my dataset this way. I don't know for sure this is the exact solution:) You can try :)

[TeerathChandani]

I have noisy and clean wave files. Each has two channel. I want to train on it. Is there any modification I have to do?

like 32-float, Sample rate changing or model input?

Thanks,