WaveNet with Gluon

Gluon implementation of WaveNet: A Generative Model for Raw Audio

Requirements

• Python 3.6.1
• Mxnet 1.2
• tqdm
• scipy.io

Usage

• arguments
  • batch_size : Define batch size (defualt=64)
  • epoches : Define total epoches (default=1000)
  • mu : Define mu (default=128)
  • n_residue : Define number of residue (default=24)
  • dilation_depth : Define dilation depth (default=10)
  • use_gpu : Use GPU (default=True)
  • generation : generate wav file for model (default=True)

default setting

python main.py

or

manual setting

python main.py --batch_size=32 --epoches=100 ...

Train progress

0 epoch

200 epoch

400 epoch

Results

Reference

• https://deepmind.com/blog/wavenet-generative-model-raw-audio/
• https://gist.github.com/lirnli/4282fcdfb383bb160cacf41d8c783c70#file-pytorch-wavenet-ipynb