Chainer implementation of Deepmind's WaveNet
Python
Latest commit 39ce4f7 Feb 17, 2017 @musyoku committed on GitHub Merge pull request #3 from nakosung/patch-2
Filter out non .wav files from scanned file list
Permalink
Failed to load latest commit information.
_tests_ j Sep 21, 2016
train_audio Filter out non .wav files from scanned file list Feb 15, 2017
README.md j Sep 21, 2016
data.py Convert to mono only if necessary Feb 15, 2017
faster_wavenet.py j Sep 21, 2016
wavenet.py j Sep 24, 2016

README.md

WaveNet: A Generative Model for Raw Audio

This is the Chainer implementation of WaveNet

この記事で実装したコードです。

まだ完成していませんが音声の生成はできます。

Todo:

  • Generating audio
  • Local conditioning
  • Global conditioning
  • Training on CSTR VCTK Corpus

Training the network

Requirements

  • Chainer 1.12
  • scipy.io.wavfile

Preprocessing

Donwsample your .wav to 16KHz / 8KHz to speed up convergence.

Create data directory

Add all .wav files to /train_audio/wav

Hyperparameters

You can edit the hyperparameters of the network in model.py before running train.py, or edit /params/params.json after training starts.

Training

run train.py

Generating audio

run generate.py

Passing --use_faster_wavenet will generate audio faster than original WaveNet.

Listen to a sample generated by WaveNet

🎶 music

Implementation

figure

figure

figure