AdversarialAutoVC

This is code for training AutoVC model with adversarial classifier for better disentanglement and WGAN for better naturalness.

Also, this model uses MelGAN vocoder instead of WaveNet.

Ideas are summed up in attached PDF.

How to

First, generate mel-spectrogram features using Audio2Mel from my_feats package.

You can run training following example in sg_script.sh

Name		Name	Last commit message	Last commit date
Latest commit History 6 Commits
my_feats		my_feats
nn_model		nn_model
Adversarial_AutoVC__Disentangling_Speech_Attributes_for_Voice_Conversion.pdf		Adversarial_AutoVC__Disentangling_Speech_Attributes_for_Voice_Conversion.pdf
README.md		README.md
config.yaml		config.yaml
main.py		main.py
sg_script.sh		sg_script.sh