The question about additional adversarial loss ?

nipponjo / tts-arabic-pytorch

TTS models for Arabic (Tacotron2, FastPitch)

84 stars 18 forks source link

The question about additional adversarial loss ? #3

Closed WhiteFu closed 1 year ago

WhiteFu commented 1 year ago

Good Job! Looking forward to your repo! I also think that the (adv) version often sounds a bit clearer. but i have some question?

Is it a joint training for tacotron2 and hifigan, or are there other details？ I'm curious about the fastpitch_ar_adv.pth model.
When it is convenient to open source more code details, I would like to further verify the quality in other languages.

I'm ooking forward to your reply

nipponjo commented 1 year ago

Hello, thanks for your interest. I have pushed the files for training the FastPitch model with and without the adversarial loss to the scripts/ folder.

Some notes:

More information about training the FastPitch model can be found at NVIDIA/DeepLearningExamples
I used a DataLoader ('DynBatchDataset') that uses different batch sizes depending on the sequence lengths in the batch. The sizes are adapted to the graphics card I use (11 GB).

WhiteFu commented 1 year ago

Thank you very much for your reply and the kindly notes. I will try to do the experiment on the new dataset as soon as possible.