keras-wavenet

Keras implementation of deepmind's wavenet paper

Link to paper

Dataset used

I have used Librispeech corpus. I have concatenated all audio files in dev-clean to create train.wav and all files in test-clean to create validate.wav. I have resampled the audio files to 8000 Hz. Here is how you can create train.wav & validate.wav using vlc on linux:

cvlc -vvv --sout-keep --sout-all --sout "#gather:transcode{acodec=s16l,channels=1,samplerate=8000}:std{access=file,mux=wav,dst=validate.wav}" `find LibriSpeech/test-clean/ -name "*.flac"` vlc://quit
cvlc -vvv --sout-keep --sout-all --sout "#gather:transcode{acodec=s16l,channels=1,samplerate=8000}:std{access=file,mux=wav,dst=train.wav}" `find LibriSpeech/dev-clean/ -name "*.flac"` vlc://quit

Todo

• The basic generative model
• Conditioning logic (speaker)
• Conditioning logic (TTS)