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

Printing fix with flush and spaceing

This commit is contained in:
Eren G 2018-07-30 13:52:39 +02:00
parent 8864252941
commit 8bc4fe8aac
2 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
train.py
View File

@ -182,7 +182,7 @@ def train(model, criterion, criterion_st, data_loader, optimizer, optimizer_st,
# print epoch stats # print epoch stats
print(" | | > EPOCH END -- GlobalStep:{} AvgTotalLoss:{:.5f} "\ print(" | | > EPOCH END -- GlobalStep:{} AvgTotalLoss:{:.5f} "\
"AvgLinearLoss:{:.5f} AvgMelLoss:{:.5f} "\ "AvgLinearLoss:{:.5f} AvgMelLoss:{:.5f} "\
"AvgStopLoss:{:.5f} EpochTime:{:.2f}"\ "AvgStopLoss:{:.5f} EpochTime:{:.2f} "\
"AvgStepTime:{:.2f}".format(current_step, "AvgStepTime:{:.2f}".format(current_step,
avg_total_loss, avg_total_loss,
avg_linear_loss, avg_linear_loss,
@ -260,7 +260,7 @@ def evaluate(model, criterion, criterion_st, data_loader, ap, current_step):
"StopLoss: {:.5f} ".format(loss.item(), "StopLoss: {:.5f} ".format(loss.item(),
linear_loss.item(), linear_loss.item(),
mel_loss.item(), mel_loss.item(),
stop_loss.item())) stop_loss.item()), flush=True)
avg_linear_loss += linear_loss.item() avg_linear_loss += linear_loss.item()
avg_mel_loss += mel_loss.item() avg_mel_loss += mel_loss.item()
@ -373,7 +373,7 @@ def main(args):
ap.num_freq, ap.num_freq,
c.num_mels, c.num_mels,
c.r) c.r)
print(" | > Num output units : {}".format(ap.num_freq)) print(" | > Num output units : {}".format(ap.num_freq), flush=True)
optimizer = optim.Adam(model.parameters(), lr=c.lr) optimizer = optim.Adam(model.parameters(), lr=c.lr)
optimizer_st = optim.Adam(model.decoder.stopnet.parameters(), lr=c.lr) optimizer_st = optim.Adam(model.decoder.stopnet.parameters(), lr=c.lr)
@ -394,20 +394,20 @@ def main(args):
for k, v in state.items(): for k, v in state.items():
if torch.is_tensor(v): if torch.is_tensor(v):
state[k] = v.cuda() state[k] = v.cuda()
print(" > Model restored from step %d" % checkpoint['step']) print(" > Model restored from step %d" % checkpoint['step'], flush=True)
start_epoch = checkpoint['step'] // len(train_loader) start_epoch = checkpoint['step'] // len(train_loader)
best_loss = checkpoint['linear_loss'] best_loss = checkpoint['linear_loss']
args.restore_step = checkpoint['step'] args.restore_step = checkpoint['step']
else: else:
args.restore_step = 0 args.restore_step = 0
print("\n > Starting a new training") print("\n > Starting a new training", flush=True)
if use_cuda: if use_cuda:
model = nn.DataParallel(model.cuda()) model = nn.DataParallel(model.cuda())
criterion.cuda() criterion.cuda()
criterion_st.cuda() criterion_st.cuda()
num_params = count_parameters(model) num_params = count_parameters(model)
print(" | > Model has {} parameters".format(num_params)) print(" | > Model has {} parameters".format(num_params), flush=True)
if not os.path.exists(CHECKPOINT_PATH): if not os.path.exists(CHECKPOINT_PATH):
os.mkdir(CHECKPOINT_PATH) os.mkdir(CHECKPOINT_PATH)
@ -418,7 +418,7 @@ def main(args):
for epoch in range(0, c.epochs): for epoch in range(0, c.epochs):
train_loss, current_step = train(model, criterion, criterion_st, train_loader, optimizer, optimizer_st, ap, epoch) train_loss, current_step = train(model, criterion, criterion_st, train_loader, optimizer, optimizer_st, ap, epoch)
val_loss = evaluate(model, criterion, criterion_st, val_loader, ap, current_step) val_loss = evaluate(model, criterion, criterion_st, val_loader, ap, current_step)
print(" | > Train Loss: {:.5f} Validation Loss: {:.5f}".format(train_loss, val_loss)) print(" | > Train Loss: {:.5f} Validation Loss: {:.5f}".format(train_loss, val_loss), flush=True)
best_loss = save_best_model(model, optimizer, train_loss, best_loss = save_best_model(model, optimizer, train_loss,
best_loss, OUT_PATH, best_loss, OUT_PATH,
current_step, epoch) current_step, epoch)

2
utils/audio_lws.py
View File

@ -105,4 +105,4 @@ class AudioProcessor(object):
else: else:
D = self._lws_processor().stft(y).T D = self._lws_processor().stft(y).T
S = self._amp_to_db(self._linear_to_mel(np.abs(D))) - self.ref_level_db S = self._amp_to_db(self._linear_to_mel(np.abs(D))) - self.ref_level_db
return self._normalize(S) return self._normalize(S)