WaveNet

Implementation of WaveNet: A Generative Model for Raw Audio

Dependencies

Theano
Lasagne
Keras

#####Implementation includes:

• Causal convolutional layers (implemented by masking)
• Dilated (à trous) convolutional layer blocks
• 256-class softmax
• Sample generation
• Downsampling from 48kHz to 24kHz
• Conversion of bitrate from 16 bit to 8 bit via μ-law algorithm
• Gated convolution (tanh * sigmoid) [TODO]
• Conditional distribution (speaker, text) [TODO]
• Context stacks [TODO]
• Tested on VCTK (Yamagishi, 2012) data set [In-Progress]
• Testing on music datasets [TODO]

#####DeepMind Blog Post
https://deepmind.com/blog/wavenet-generative-model-raw-audio/

#####Paper
https://arxiv.org/pdf/1609.03499.pdf

#####Parts adapted from
https://github.com/igul222/pixel_rnn and
https://github.com/kundan2510/pixelCNN