search

r9y9 / wavenet_vocoder

WaveNet vocoder
https://r9y9.github.io/wavenet_vocoder/
Other
2.31k stars 499 forks source link

Training directory error #196

Closed ghost closed 4 years ago

ghost commented 4 years ago

Nice work @r9y9, your work is helping me in my master degree project, thanks so much. However, i was stuck on training after i have preprocess my data to this directory ./sounded/audio. Using this command for training: python train.py --dump-root=./sounded/audio --preset=params. i even followed a similar pattern with someone like this: python train.py --data-root=./data/BZNSYP/ --speaker-id=0 --hparams="cin_channels=80,gin_channels=-1" and also tried to use different version of nnmnkwii modules which is not still working keeps on getting this error from the nnmnkwii modules that: "No files are collected. You might have specified wrong data source." which i pointed to the audio folder containing the output from the preprocess run Thank you in advance

r9y9 commented 4 years ago

https://github.com/r9y9/wavenet_vocoder/blob/63d3fe0a85a69aa25d80700974fe282c96fb6b65/train.py#L982-L995

Please make sure your have at leaset two directories in dump root: train_no_dev and dev. Also please first check working recipes, rather than directly invoking CLI scripts.

ghost commented 4 years ago

i'm grateful and happy for your response @r9y9 , i have included train_no_dev and dev folders in my dump root directory containing all my preprocessed output, but still giving me same error, please correct me if i'm mistaken, i have gone through the nnmnnkwii file to check the function that was called in collecting the files from the dump root directory and i notice that the collectfiles() didn't do anything other than raising errors

again i have install pytorch for cpu users as i want to test if it will work here before shifting my codebase to digital ocean to perform the training, but i don't think that should be an issue?.

Thank you for your time

Screenshot (1497) Screenshot (1498)

r9y9 commented 4 years ago

I am sorry if I am wrong but you must be the same person as https://github.com/r9y9/wavenet_vocoder/issues/194. I can see exactly the same directory, asking for the same question. I will close this issue as this is a problem on your side. The error should not happen if you run one of the provided recipes.