search

r9y9 / wavenet_vocoder

WaveNet vocoder
https://r9y9.github.io/wavenet_vocoder/
Other
2.33k stars 500 forks source link

pretrained ljspeech model are all 13100 files used in training? #144

Closed soumimaiti closed 5 years ago

soumimaiti commented 5 years ago

Hi,

I wanted to know if all LJspeech files are used for training the pretrained ljspeech model?

I want to fine-tune the model, so I wanted to use the same training files for the fine-tuning.

r9y9 commented 5 years ago

https://github.com/r9y9/wavenet_vocoder/blob/44e0e361bc02b3e20eb1db4fc5c1c8c04ba01da2/presets/ljspeech_mixture.json#L42

https://github.com/r9y9/wavenet_vocoder/blob/44e0e361bc02b3e20eb1db4fc5c1c8c04ba01da2/train.py#L113-L121

Roughly 4.4% of data was excluded for training.