tuna2134/sbv2-v2
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Update train_ms.py

Browse Source
This commit is contained in:
Stardust·减
2023-08-23 16:46:59 +08:00
committed by GitHub
parent 9d625f50fc
commit effec21548
1 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
train_ms.py
View File

@@ -216,7 +216,7 @@ def train_and_evaluate(rank, epoch, hps, nets, optims, schedulers, scaler, loade
with autocast(enabled=hps.train.fp16_run):
y_hat, l_length, attn, ids_slice, x_mask, z_mask, \
(z, z_p, m_p, logs_p, m_q, logs_q), (logw, logw_) = net_g(x, x_lengths, spec, spec_lengths, speakers, tone, language, bert)
(z, z_p, m_p, logs_p, m_q, logs_q), (hidden_x, logw, logw_) = net_g(x, x_lengths, spec, spec_lengths, speakers, tone, language, bert)
mel = spec_to_mel_torch(
spec,
hps.data.filter_length,
@@ -244,9 +244,9 @@ def train_and_evaluate(rank, epoch, hps, nets, optims, schedulers, scaler, loade
loss_disc, losses_disc_r, losses_disc_g = discriminator_loss(y_d_hat_r, y_d_hat_g)
loss_disc_all = loss_disc
if net_dur_disc is not None:
y_dur_hat_r, y_dur_hat_g = net_dur_disc(logw, logw_.detach())
y_dur_hat_r, y_dur_hat_g = net_dur_disc(hidden_x, x_mask, logw, logw_.detach())
with autocast(enabled=False):
# TODO: I think need to mean using the mask, but for now, just mean all
# TODO: I think need to mean using the mask, but for now, just mean all
loss_dur_disc, losses_dur_disc_r, losses_dur_disc_g = discriminator_loss(y_dur_hat_r, y_dur_hat_g)
loss_dur_disc_all = loss_dur_disc
optim_dur_disc.zero_grad()
@@ -265,7 +265,7 @@ def train_and_evaluate(rank, epoch, hps, nets, optims, schedulers, scaler, loade
# Generator
y_d_hat_r, y_d_hat_g, fmap_r, fmap_g = net_d(y, y_hat)
if net_dur_disc is not None:
y_dur_hat_r, y_dur_hat_g = net_dur_disc(x, x_mask, logw, logw_)
y_dur_hat_r, y_dur_hat_g = net_dur_disc(hidden_x, x_mask, logw, logw_)
with autocast(enabled=False):
loss_dur = torch.sum(l_length.float())
loss_mel = F.l1_loss(y_mel, y_hat_mel) * hps.train.c_mel