Trouble on getting PWGan to work

Archives TTS (Text-to-Speech)
#1

Hello everyone!

I’m having some problems to get Parallel Wavegan to work. The steps I took until now were:

  1. Cloned the repo
  2. Checked out to the correct branch (the one of the pre-trained PWGan model)
  3. Adjusted the config.yml file to match my dataset (sample_rate from 22050 to 16000) and changed the paths
  4. Executed bin/preprocess_tts.py
  5. Executed bin/train.py

The error I get is located at parallel_wavegan.py, on assert c.size(-1) == x.size(-1) (AssertionError: 37400 vs 27200), but I still haven’t figured out what may be causing it.

I’d be glad if someone could bring a light upon this.

#2

Do you have sound files with a sample rate of 37400?

#3

@georroussos no, I don’t. Sample Rate is 16000 for my entire dataset. I reprocessed the dataset with some modified parameters and now this error has stopped happening.

#4

Different sample rate requires different upsampling factors. Right now, I am working on a native vocoder support instead of using 3rd party implementations. Therefore, consider to give it a try using the dev branch.

1 Like
#5

Thank you @erogol! Nice to hear that, I’ll give it a try.

#6

@LucasRotsen @erogol Can you please tell me, i am getting same error? Please help me how can i avoid this error. My sample rate is 16000.