ForwardTTSE2E implementations and related API changes #1510
Conversation
|
Do I understand correctly that these changes were made for the only VITS model (YourTTS) from the repository? Since it is the only End2End you have? |
| @@ -191,6 +191,9 @@ def __init__( | |||
| ): | |||
| super().__init__() | |||
|
|
|||
| if c_in_channels and c_in_channels != 0: | |||
There was a problem hiding this comment.
c_in_channels != 0 is redundant since the first clause of the and would be evaluated to False if c_in_channels == 0
| @@ -225,6 +228,9 @@ def forward(self, x, x_mask, g=None): # pylint: disable=unused-argument | |||
| x_mask: [B, 1, T] | |||
| g: [B, C_g, 1] | |||
| """ | |||
| # TODO: implement multi-speaker | |||
| o = self.decoder(x, x_mask, g) | |||
| # multi-speaker conditioning | |||
There was a problem hiding this comment.
won't that break the vctk/fast_pitch released model?
There was a problem hiding this comment.
Not sure. Even if it does, I think this is the right way to go, similar to the VITS model.
| return outputs, loss_dict | ||
|
|
||
| if optimizer_idx == 1: | ||
| mel = batch["mel_input"].transpose(1, 2) |
There was a problem hiding this comment.
Maybe implementing steps_to_start_discriminator would allow for faster training
| import torch | ||
| from torch import nn | ||
|
|
||
|
|
There was a problem hiding this comment.
I don't know how useful it would be, but maybe it would be nice to have a differentiable equivalent of functions in numpy_transform here when possible.
|
Thx for the review @WeberJulian |
Update CI badges
erogol
force-pushed
the
unified_api_forwardtts2
branch
from
b6073d1 to
8adcd1d
Compare
|
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. You might also look our discussion channels. |
No description provided.