wavenet

A deep generative model for producing realistic speech. It creates waveforms sample-by-sample using autoregressive connections and dilated convolutions.