wavenet vocoder using tensorflow
!!! the audio code is copied from wavenet_vocoder !!!
!!! the main tensorflow model is fixed from tensorflow-wavenet !!!
mixture is in the branch of dev, but there are some bugs in generating wavs.
for train faster, we should process the data to npy
python preprocess.py --num_workers 4 --name ljspeech --in_dir /your_path/LJSpeech-1.0 --out_dir /your_outpath/ --hparams sample_rate=22050
python train.py --num_gpus 4 --batch_size 2 --train_txt /your_train_txt/ --hparams gc_enable=False,global_channel=0,global_cardinality=0,NPY_DATAROOT=/your_npy_datadir/,sample_rate=22050 --logdir_root log_ljspeech
python train.py --batch_size 2 --num_gpus 4 --train_txt /your_train_txt/ --logdir_root log_arctic
the eval_txt is extracted from the train_txt
python mul_generate.py --eval_txt /your_eval_txt/ --wav_out_path test_ljspeech.wav /your_cheakpoint/ ---hparams gc_enable=False,global_channel=0,global_cardinality=0,NPY_DATAROOT=/your_npy_datadir/,sample_rate=22050
python mul_generate.py --eval_txt /your_eval_txt/ --wav_out_path test_arctic.wav /your_checkpoint/ --gc_id 6