wassname/TTS
1
0
Fork 0
mirror of https://github.com/wassname/TTS.git synced 2026-09-09 11:16:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

Make loss masking optional

Browse Source
This commit is contained in:
Eren Golge
2019-04-10 16:41:08 +02:00
parent 8a47b46195
commit e2cf35bb10
2 changed files with 44 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
train.py
+26 -9
View File
@@ -125,11 +125,18 @@ def train(model, criterion, criterion_st, optimizer, optimizer_st, scheduler,
# loss computation
stop_loss = criterion_st(stop_tokens, stop_targets)
decoder_loss = criterion(decoder_output, mel_input, mel_lengths)
if c.model == "Tacotron":
postnet_loss = criterion(postnet_output, linear_input, mel_lengths)
if c.loss_masking:
decoder_loss = criterion(decoder_output, mel_input, mel_lengths)
if c.model == "Tacotron":
postnet_loss = criterion(postnet_output, linear_input, mel_lengths)
else:
postnet_loss = criterion(postnet_output, mel_input, mel_lengths)
else:
postnet_loss = criterion(postnet_output, mel_input, mel_lengths)
decoder_loss = criterion(decoder_output, mel_input)
if c.model == "Tacotron":
postnet_loss = criterion(postnet_output, linear_input)
else:
postnet_loss = criterion(postnet_output, mel_input)
loss = decoder_loss + postnet_loss
# backpass and check the grad norm for spec losses
@@ -283,11 +290,18 @@ def evaluate(model, criterion, criterion_st, ap, current_step, epoch):
# loss computation
stop_loss = criterion_st(stop_tokens, stop_targets)
decoder_loss = criterion(decoder_output, mel_input, mel_lengths)
if c.model == "Tacotron":
postnet_loss = criterion(postnet_output, linear_input, mel_lengths)
if c.loss_masking:
decoder_loss = criterion(decoder_output, mel_input, mel_lengths)
if c.model == "Tacotron":
postnet_loss = criterion(postnet_output, linear_input, mel_lengths)
else:
postnet_loss = criterion(postnet_output, mel_input, mel_lengths)
else:
postnet_loss = criterion(postnet_output, mel_input, mel_lengths)
decoder_loss = criterion(decoder_output, mel_input)
if c.model == "Tacotron":
postnet_loss = criterion(postnet_output, linear_input)
else:
postnet_loss = criterion(postnet_output, mel_input)
loss = decoder_loss + postnet_loss + stop_loss
step_time = time.time() - start_time
@@ -383,7 +397,10 @@ def main(args):
optimizer_st = optim.Adam(
model.decoder.stopnet.parameters(), lr=c.lr, weight_decay=0)
criterion = L1LossMasked() if c.model == "Tacotron" else MSELossMasked()
if c.loss_masking:
criterion = L1LossMasked() if c.model == "Tacotron" else MSELossMasked()
else:
criterion = nn.L1Loss() if c.model == "Tacotron" else nn.MSELoss()
criterion_st = nn.BCEWithLogitsLoss()
if args.restore_path: