search

jaywalnut310 / vits

VITS: Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech
https://jaywalnut310.github.io/vits-demo/index.html
MIT License
6.73k stars 1.24k forks source link

[Bug]`epoch_str` should be increased after loaded checkpoint #67

Open cantabile-kwok opened 2 years ago

cantabile-kwok commented 2 years ago

VITS training automatically loads the last checkpoint. But after that, the epoch_str in train.py should be increased. In the current version of code, the training process repeats the last epoch, which in most cases does not make sense.

Compared to glow-tts official implementation, I think VITS should correct this bug, if this is not an expected feature.

Aloento commented 2 years ago

https://github.com/jaywalnut310/vits/blob/2e561ba58618d021b5b8323d3765880f7e0ecfdb/train.py#L105

I see here. how do you think to fix this?

cantabile-kwok commented 2 years ago

Compared to glow-tts, https://github.com/jaywalnut310/glow-tts/blob/13e997689d643410f5d9f1f9a73877ae85e19bc2/train.py#L78 maybe adding such a line is enough

https://github.com/jaywalnut310/vits/blob/2e561ba58618d021b5b8323d3765880f7e0ecfdb/train.py#L105

I see here. how do you think to fix this?