This is a Pytorch implementation of extended VQVAE mentioned in our paper.

We introduce an important extension to VQ-VAE for learning F0-related suprasegmental information simultaneously along with traditional phone features. The proposed framework uses two encoders such that the F0 trajectory and speech waveform are both input to the system, therefore two separate codebooks are learned.

We reconstructed the speech using both original VQVAE and extended VQVAE with F0 encoder.

In brief, we have done:

1. Extended the original VQ-VAE with an F0 encoder.
2. Extended the global condition to speaker code, gender code, and emotion code.
3. Trained a model with multi-speaker & multi-emotional Japanese corpus.
4. Trained a model with a public Chinese corpus.
5. A parallel training script on multiple gpus.

To do:

1. Testing on voice conversion
2. Testing on emotions' conversion

Authors of the paper: Yi Zhao, Haoyu Li, Cheng-I Lai, Jennifer Williams, Erica Cooper, Junichi Yamagishi

For any question related to the paper or the scripts, please contact zhaoyi[email mark]nii.ac.jp.

Please find our samples here.

Please install packages in requirement.txt before using the scripts.

1. extract F0. (We used crepe to extract F0. )
2. F0 and Wavefrom Alignment
3. converting F0 and waveform into *.npy format.

Please use ./run.sh when train an extended vavae model.

Or you can use python3 train.py -m [model type]. The -m option can be used to tell the the script to train a different model.

[model type] can be:

• 'vqvae': Train original VQVAE
• 'wavernn': train an WaveRNN model
• 'vcf0': extended VQVAE with F0 encoder

Please modify sampling rate and other parameters in config.py before training.

We have Japanese or Chinese trained models for both original VQVAE and extended VQVAE. If you’re interested in using our pre-trained models for research purpose, please contact the zhaoyi[email mark]nii.ac.jp.

Please see multi_gpu_wavernn.py

The code is based on mkotha/WaveRNN

Cheng-I is supported by the Merrill Lynch Fel- lowship, MIT. This work was partially supported by a JST CREST Grant (JPMJCR18A6, VoicePersonae project), Japan, and by MEXT KAKENHI Grants (16H06302, 18H04112, 18KT0051, 19K24373 Japan. The numerical calculations were carried out on the TSUBAME 3.0 supercomputer at the Tokyo Institute of Technology.

MIT License

• Copyright (c) 2019 fatchord (https://github.com/fatchord)
• Copyright (c) 2019 mkotha (https://github.com/mkotha)
• Copyright (c) 2020, Yamagishi Laboratory, National Institute of Informatics.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.