Small bug in train_finetune

[Karesto]

In train_finetune, line 580 :

mask = length_to_mask(mel_input_length // (2 ** n_down)).to("cuda")

I assume this should be .to(device) in case of specific devices.

In addition to that, when running the FineTune notebook, multiple erros linked to iters_test = 0 pop up.

[Faolain]

Can you open a PR?

[Karesto]

Hi, there's already a PR opened for this specific issue : https://github.com/yl4579/StyleTTS2/pull/229.