WaveNet

A Chainer implementation of mel-spectrogram vocoder using WaveNet.

Usage

1. Install requirements.
   • pip3 install -r requirements.txt
2. Download dataset.
   • wget http://homepages.inf.ed.ac.uk/jyamagis/release/VCTK-Corpus.tar.gz
   • tar -xf VCTK-Corpus.tar.gz
3. Start training.
   • python train.py -g <gpu id> --dataset <directory of dataset e.g. ./VCTK-Corpus/>
   • You can change other parameters. Please see args.
4. Generate audio with trained model.
   • python generate.py -i <input file> -m <trained model e.g. snapshot_iter_500000>

Details

• Default parameters of WaveNet are same as nv-wavenet.
• Mel-spectrograms are calculated with librosa.
• If you want to get more audible results, use MoL Wavenet with exponential moving average.