working on parallel wavenet
- code base on https://github.com/r9y9/wavenet_vocoder/
- there are many bugs in the code, i am still working on it,hope people can help to build.
- you should modify the train_student.py to meet your condition.
- some ideas maybe wrong(just feel free to point out and i'm very thankful for that) and any discussion are welcomed.
- there are already some discussions you can find from kensun0/Parallel-Wavenet#1 and r9y9/wavenet_vocoder#7
- you can download the teacher model from https://www.dropbox.com/s/8qgcbd1mm2xsqgq/20180127_mixture_lj_checkpoint_step000410000_ema.pth?dl=0 from discuss r9y9/wavenet_vocoder#19 ,and change some parameters in hparameter.py and reprocess the ljspeech dataset for training student wavenet.
- a pretrained model can be download from https://www.dropbox.com/s/i42do45t4ep9pir/checkpoint_step000680000.pth?dl=0, not very good quality, I'm still training it. use the synthesis_student.py to generate wave,check synthesis_student.py for more details.
- join us to discuss:QQ群 548019638