coqui-ai
/
TTS
mirror of https://github.com/coqui-ai/TTS.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Update losses.py 

The block of code for use_l1_spec_loss is repeated which doubles the amount of L1 loss when enabled.
The weight for L1 loss in hifigan_ljspeech configutation will likely need to be doubled to compensate (l1_spec_loss_weight)
pull/468/head
shaun 2021-05-02 14:14:24 +02:00 committed by GitHub
parent b00b1d4680
commit 7d0ec62bf1
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 0 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
TTS/vocoder/layers/losses.py
View File

@ -350,12 +350,6 @@ class GeneratorLoss(nn.Module):
return_dict["G_l1_spec_loss"] = l1_spec_loss
gen_loss = gen_loss + self.l1_spec_loss_weight * l1_spec_loss
# L1 Spec loss
if self.use_l1_spec_loss:
l1_spec_loss = self.l1_spec_loss(y_hat, y)
return_dict["G_l1_spec_loss"] = l1_spec_loss
gen_loss = gen_loss + self.l1_spec_loss_weight * l1_spec_loss
# subband STFT Loss
if self.use_subband_stft_loss:
subband_stft_loss_mg, subband_stft_loss_sc = self.subband_stft_loss(y_hat_sub, y_sub)