ericwudayi / skipvqvc Goto Github PK

Thanks for preparing the code. I think there are some minor issues for beginners to execute the code:

1. The required packages and their specific versions are not listed in readme file (and I couldn't find it somewhere else in repositories).
2. It would be better to prepare a link to the dataset in readme file. In this way we will make sure we are using the exact version of the dataset which was used in the paper.
3. I downloaded the VCTK dataset and the files have .wav format in under /wav48 folder. So, I think it the preprocessing.py file, the extension variable (line 113) should be '.wav' instead of '.mp3'.
4. After preprocessing, I found out that the test and train sets are not separated in tow different folders. Should we separate them manually by taking 20 speakers' as the test set and preprocess them individually?
   Please let me know if I was incorrect in each of the above three points. Thanks!

Hi, thank you for the great work.

I have two questions.

python train.py -train_dir your-path-to-npy-dir -m vqvc+ -n 64 -ch 512 -t train_simple_normalize

with the above command, can i recover the results you have shown on the demo website?

And also with the same command, only changing the channel 512 to 80, will the result may similar or worsen the quality?