Closed machineko closed 4 years ago
I am currently in the process of training. A pretrained model would be helpful for closer comparison.
I have trained a FastPitch model using the official pipeline. But I do not know if the audios are competitive with Tacotron2.
Here are the link and password. link password to extract file: g3fn
@Yablon Thanks for this but it's in Chinese so I'm struggling to navigate - I think it's asking me to download and install some software? I'm not comfortable doing that. Could you host the file somewhere else, for example Google Drive, Dropbox, Apple Drive, basically anywhere else?
I (and others) really appreciate it.
@Yablon Also, I've done a lot of research in the field - if you could share your email I could let you know how does the model you trained compare to tacotron2.
@NinoSkopac Yeah, here is a new link on Google Drive. I have never trained a tacotron2 model using LJSpeech data, so it's hard for me to compare them. If FastPitch is better, I think a lot of things can be done.
I would also love an English pre-trained model. @NinoSkopac did you ever make any progress?
Hey @gad2103, what do you mean? Yablon posted a link to a pretrained model above (I haven't verified it myself yet).
Related to FastPitch/Pytorch
Are you guys planning on releasing pretrained FastPitch model? I would love to try few things with it but I don't have computer power at the moment to train on LJ dataset from scratch :P