More training?

[sbkim052]

Is this model only trained by VCTK dataset?
If so, is there a chance to improve the performance by training with more data such as Librispeech?

[jjery2243542]

This model performs not so well on libriTTS. I guess it is related to the cleaness of the data. If you have any result please share with me :)

[sbkim052]

Thanks for replying to my question:) I will try training it for other datasets and share it with you.
I have an additional question.
I have applied your model to Korean, however, it didn't perform that well.
Will it be better if I fine-tune each encoder (Speaker, Content) with Korean speech?
It will be best to experiment with it, but I don't have GPU resources yet and just want to hear your idea.

[zhu-ya]

Hello, take the liberty to disturb, seeing that you have successfully run the code, I have a question to ask you. I use LibriTTS for pre-processing, but during the pre-processing process, I can't find the pickle file, such as train.pkl, dev.pkl, etc. I wonder if you have encountered the same problem.Thanks.

[sbkim052]

I can't either. sorry

[jjery2243542]

Did you change the data_dir variable in libri.conf ?