yl4579/pitchextractor

Deep Neural Pitch Extractor for Voice Conversion and TTS Training

pitchextractor's Issues

network codes seems not same with codes in StarGanv2VC

A possible bug

When loading a pretrained model checkpoint, the model cannot correctly read the correct epoch and always starts from epoch zero.

Num_class must be 1?

It seems that the num_class must be one, otherwise I need to change the code of L1_loss, so if I have two classes, should I use class=1 or change the loss code?
Thank you for your help!

sil loss problem

I trained 3 times with different data, the data includes talking and singing, but the eval sil loss goes up each time. Can you help me with this problem? thank you very much. @yl4579

About checkpoint file format

Using the codebase, the checkpoint is saved as "epoch_00100.pth", while the StarGANv2-VC's default checkpoint file is "bst.t7". They have different extensions.
When loading in the "epoch_00100.pth" checkpoint in StarGANv2-VC, it throws me an error
File "/opt/conda/envs/star/lib/python3.9/site-packages/click/core.py", line 1130, in __call__ return self.main(*args, **kwargs) File "/opt/conda/envs/star/lib/python3.9/site-packages/click/core.py", line 1055, in main rv = self.invoke(ctx) File "/opt/conda/envs/star/lib/python3.9/site-packages/click/core.py", line 1404, in invoke return ctx.invoke(self.callback, **ctx.params) File "/opt/conda/envs/star/lib/python3.9/site-packages/click/core.py", line 760, in invoke return __callback(*args, **kwargs) File "/home/charlie_babelcast_com/github/StarGANv2-VC/train.py", line 90, in main params = torch.load(F0_path, map_location='cpu')['net'] KeyError: 'net
How can I save the correct .t7 file?
Thank you!

About the training/eval loss

Hi. Thank you for your great work. Could you also share your loss curve? Thank you very much~

mvpe

Have you considered using rmvpe to produce ground truth f0 labels for this model? RVC project uses it and from my understanding it is considered to be more accurate than pyworld dio/harvest