search

patrickltobing / cyclevae-vc

Non-Parallel Voice Conversion with Cyclic Variational Autoencoder
Apache License 2.0
52 stars 4 forks source link

Where could I find the bash script to extract speaker features which are in conf folder? #3

Closed v-manhlt3 closed 3 years ago

v-manhlt3 commented 3 years ago

Hi Patrick,

First of all, thank you for your brilliant work. I am working on my research paper and I am planning to use your work as a baseline, however, I would like to estimate model's performance on VCTK dataset. I could not find any python files or bash script files to generate speaker's configs as in "conf" folder. Therefore, I appreciate if you can help me to generate speaker's conf.

patrickltobing commented 3 years ago

Hi, thanks for the request.

The development related on this repo has been moved to cyclevae-vc-neuralvoco.

I could not find any python files or bash script files to generate speaker's configs as in "conf" folder.

Please checkout the directory cycvae_mwdlp_vcc20.

Follow Data preparation step and Data preprocessing steps until stage=0init.

All .f0 and .pow files will then be located in the conf folder, and will be written to conf/spkr.yml

Let me know if there are any troubles.

v-manhlt3 commented 3 years ago

Thanks for your response, now I can train your model on my dataset. However, it seems to take a very long time to finish training with only 4 speakers. So, I am curious that how long does it take to finish training on the VCC2020 dataset?

patrickltobing commented 3 years ago

However, it seems to take a very long time to finish training with only 4 speakers. how long does it take to finish training on the VCC2020 dataset?

If you are asking about the training time for the github project on the link I sent above, I have put it on the readme. It may take 2 days for first VC training of vcc20 set.

If you are asking for this github project, I kind of forgot, it might take 3 days or so.

v-manhlt3 commented 3 years ago

Thanks for your support!!!