Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Use self variables instead of hparams #9

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Use self variables instead of hparams #9

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
55 changes: 32 additions & 23 deletions vocos/experiment.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,15 @@ def __init__(
self.train_discriminator = False
self.base_mel_coeff = self.mel_loss_coeff = mel_loss_coeff

self.initial_learning_rate = initial_learning_rate
self.num_warmup_steps = num_warmup_steps
self.mrd_loss_coeff = mrd_loss_coeff
self.pretrain_mel_steps = pretrain_mel_steps
self.decay_mel_coeff = decay_mel_coeff
self.evaluate_utmos = evaluate_utmos
self.evaluate_pesq = evaluate_pesq
self.evaluate_periodicty = evaluate_periodicty

def configure_optimizers(self):
disc_params = [
{"params": self.multiperioddisc.parameters()},
Expand All @@ -78,15 +87,15 @@ def configure_optimizers(self):
{"params": self.head.parameters()},
]

opt_disc = torch.optim.AdamW(disc_params, lr=self.hparams.initial_learning_rate)
opt_gen = torch.optim.AdamW(gen_params, lr=self.hparams.initial_learning_rate)
opt_disc = torch.optim.AdamW(disc_params, lr=self.initial_learning_rate)
opt_gen = torch.optim.AdamW(gen_params, lr=self.initial_learning_rate)

max_steps = self.trainer.max_steps // 2 # Max steps per optimizer
scheduler_disc = transformers.get_cosine_schedule_with_warmup(
opt_disc, num_warmup_steps=self.hparams.num_warmup_steps, num_training_steps=max_steps,
opt_disc, num_warmup_steps=self.num_warmup_steps, num_training_steps=max_steps
)
scheduler_gen = transformers.get_cosine_schedule_with_warmup(
opt_gen, num_warmup_steps=self.hparams.num_warmup_steps, num_training_steps=max_steps,
opt_gen, num_warmup_steps=self.num_warmup_steps, num_training_steps=max_steps
)

return (
Expand Down Expand Up @@ -118,7 +127,7 @@ def training_step(self, batch, batch_idx, optimizer_idx, **kwargs):
)
loss_mp /= len(loss_mp_real)
loss_mrd /= len(loss_mrd_real)
loss = loss_mp + self.hparams.mrd_loss_coeff * loss_mrd
loss = loss_mp + self.mrd_loss_coeff * loss_mrd

self.log("discriminator/total", loss, prog_bar=True)
self.log("discriminator/multi_period_loss", loss_mp)
Expand Down Expand Up @@ -152,9 +161,9 @@ def training_step(self, batch, batch_idx, optimizer_idx, **kwargs):
mel_loss = self.melspec_loss(audio_hat, audio_input)
loss = (
loss_gen_mp
+ self.hparams.mrd_loss_coeff * loss_gen_mrd
+ self.mrd_loss_coeff * loss_gen_mrd
+ loss_fm_mp
+ self.hparams.mrd_loss_coeff * loss_fm_mrd
+ self.mrd_loss_coeff * loss_fm_mrd
+ self.mel_loss_coeff * mel_loss
)

Expand All @@ -164,10 +173,10 @@ def training_step(self, batch, batch_idx, optimizer_idx, **kwargs):

if self.global_step % 1000 == 0 and self.global_rank == 0:
self.logger.experiment.add_audio(
"train/audio_in", audio_input[0].data.cpu(), self.global_step, self.hparams.sample_rate
"train/audio_in", audio_input[0].data.cpu(), self.global_step, self.sample_rate
)
self.logger.experiment.add_audio(
"train/audio_pred", audio_hat[0].data.cpu(), self.global_step, self.hparams.sample_rate
"train/audio_pred", audio_hat[0].data.cpu(), self.global_step, self.sample_rate
)
with torch.no_grad():
mel = safe_log(self.melspec_loss.mel_spec(audio_input[0]))
Expand All @@ -188,7 +197,7 @@ def training_step(self, batch, batch_idx, optimizer_idx, **kwargs):
return loss

def on_validation_epoch_start(self):
if self.hparams.evaluate_utmos:
if self.evaluate_utmos:
from metrics.UTMOS import UTMOSScore

if not hasattr(self, "utmos_model"):
Expand All @@ -198,22 +207,22 @@ def validation_step(self, batch, batch_idx, **kwargs):
audio_input = batch
audio_hat = self(audio_input, **kwargs)

audio_16_khz = torchaudio.functional.resample(audio_input, orig_freq=self.hparams.sample_rate, new_freq=16000)
audio_hat_16khz = torchaudio.functional.resample(audio_hat, orig_freq=self.hparams.sample_rate, new_freq=16000)
audio_16_khz = torchaudio.functional.resample(audio_input, orig_freq=self.sample_rate, new_freq=16000)
audio_hat_16khz = torchaudio.functional.resample(audio_hat, orig_freq=self.sample_rate, new_freq=16000)

if self.hparams.evaluate_periodicty:
if self.evaluate_periodicty:
from metrics.periodicity import calculate_periodicity_metrics

periodicity_loss, pitch_loss, f1_score = calculate_periodicity_metrics(audio_16_khz, audio_hat_16khz)
else:
periodicity_loss = pitch_loss = f1_score = 0

if self.hparams.evaluate_utmos:
if self.evaluate_utmos:
utmos_score = self.utmos_model.score(audio_hat_16khz.unsqueeze(1)).mean()
else:
utmos_score = torch.zeros(1, device=self.device)

if self.hparams.evaluate_pesq:
if self.evaluate_pesq:
from pesq import pesq

pesq_score = 0
Expand Down Expand Up @@ -243,10 +252,10 @@ def validation_epoch_end(self, outputs):
if self.global_rank == 0:
*_, audio_in, audio_pred = outputs[0].values()
self.logger.experiment.add_audio(
"val_in", audio_in.data.cpu().numpy(), self.global_step, self.hparams.sample_rate
"val_in", audio_in.data.cpu().numpy(), self.global_step, self.sample_rate
)
self.logger.experiment.add_audio(
"val_pred", audio_pred.data.cpu().numpy(), self.global_step, self.hparams.sample_rate
"val_pred", audio_pred.data.cpu().numpy(), self.global_step, self.sample_rate
)
mel_target = safe_log(self.melspec_loss.mel_spec(audio_in))
mel_hat = safe_log(self.melspec_loss.mel_spec(audio_pred))
Expand Down Expand Up @@ -286,22 +295,22 @@ def global_step(self):
return self.trainer.fit_loop.epoch_loop.total_batch_idx

def on_train_batch_start(self, *args):
if self.global_step >= self.hparams.pretrain_mel_steps:
if self.global_step >= self.pretrain_mel_steps:
self.train_discriminator = True
else:
self.train_discriminator = False

def on_train_batch_end(self, *args):
def mel_loss_coeff_decay(current_step, num_cycles=0.5):
max_steps = self.trainer.max_steps // 2
if current_step < self.hparams.num_warmup_steps:
if current_step < self.num_warmup_steps:
return 1.0
progress = float(current_step - self.hparams.num_warmup_steps) / float(
max(1, max_steps - self.hparams.num_warmup_steps)
progress = float(current_step - self.num_warmup_steps) / float(
max(1, max_steps - self.num_warmup_steps)
)
return max(0.0, 0.5 * (1.0 + math.cos(math.pi * float(num_cycles) * 2.0 * progress)))

if self.hparams.decay_mel_coeff:
if self.decay_mel_coeff:
self.mel_loss_coeff = self.base_mel_coeff * mel_loss_coeff_decay(self.global_step + 1)


Expand Down Expand Up @@ -365,7 +374,7 @@ def validation_epoch_end(self, outputs):
self.feature_extractor.encodec.set_target_bandwidth(self.feature_extractor.bandwidths[0])
encodec_audio = self.feature_extractor.encodec(audio_in[None, None, :])
self.logger.experiment.add_audio(
"encodec", encodec_audio[0, 0].data.cpu().numpy(), self.global_step, self.hparams.sample_rate,
"encodec", encodec_audio[0, 0].data.cpu().numpy(), self.global_step, self.sample_rate,
)

super().validation_epoch_end(outputs)