wavernn-16bit

The (unofficial) vanilla version of WaveRNN. Read the WaveRNN paper at here.

Note:

• We use the upsample network from Lyra paper.

To install:

git clone https://github.com/NTT123/wavernn-16bit.git
cd wavernn-16bit
pip3 install -e .

To train WaveRNN on dataset:

python3 -m wavernn.trainer --wav-dir=path/to/wav/directory --ckpt-dir=path/to/checkpoint/directory

To synthesize speech from melspectrogram:

python3 -m wavernn.mel2wave -m path/to/mel_file.npy -c path/to/training/checkpoint.pickle -o path/to/output.wav