Apply Code Formatter Change
This commit is contained in:
Stardust-minus
committed by
github-actions[bot]
github-actions[bot]
parent
d82ba3457a
commit
eef212fd00
425
train_ms.py
425
train_ms.py
@@ -14,37 +14,38 @@ from torch.nn.parallel import DistributedDataParallel as DDP
|
||||
from torch.cuda.amp import autocast, GradScaler
|
||||
from tqdm import tqdm
|
||||
import logging
|
||||
logging.getLogger('numba').setLevel(logging.WARNING)
|
||||
|
||||
logging.getLogger("numba").setLevel(logging.WARNING)
|
||||
import commons
|
||||
import utils
|
||||
from data_utils import (
|
||||
TextAudioSpeakerLoader,
|
||||
TextAudioSpeakerCollate,
|
||||
DistributedBucketSampler
|
||||
DistributedBucketSampler,
|
||||
)
|
||||
from models import (
|
||||
SynthesizerTrn,
|
||||
MultiPeriodDiscriminator,
|
||||
DurationDiscriminator,
|
||||
)
|
||||
from losses import (
|
||||
generator_loss,
|
||||
discriminator_loss,
|
||||
feature_loss,
|
||||
kl_loss
|
||||
)
|
||||
from losses import generator_loss, discriminator_loss, feature_loss, kl_loss
|
||||
from mel_processing import mel_spectrogram_torch, spec_to_mel_torch
|
||||
from text.symbols import symbols
|
||||
|
||||
torch.backends.cudnn.benchmark = True
|
||||
torch.backends.cuda.sdp_kernel("flash")
|
||||
torch.backends.cuda.enable_flash_sdp(True)
|
||||
torch.backends.cuda.enable_mem_efficient_sdp(True) # Not avaliable if torch version is lower than 2.0
|
||||
torch.backends.cuda.enable_mem_efficient_sdp(
|
||||
True
|
||||
) # Not avaliable if torch version is lower than 2.0
|
||||
torch.backends.cuda.enable_math_sdp(True)
|
||||
global_step = 0
|
||||
|
||||
|
||||
def run():
|
||||
dist.init_process_group(backend="nccl", init_method="env://") # Use torchrun instead of mp.spawn
|
||||
dist.init_process_group(
|
||||
backend="nccl", init_method="env://"
|
||||
) # Use torchrun instead of mp.spawn
|
||||
rank = dist.get_rank()
|
||||
n_gpus = dist.get_world_size()
|
||||
hps = utils.get_hparams()
|
||||
@@ -64,17 +65,34 @@ def run():
|
||||
[32, 300, 400, 500, 600, 700, 800, 900, 1000],
|
||||
num_replicas=n_gpus,
|
||||
rank=rank,
|
||||
shuffle=True)
|
||||
shuffle=True,
|
||||
)
|
||||
collate_fn = TextAudioSpeakerCollate()
|
||||
train_loader = DataLoader(train_dataset, num_workers=16, shuffle=False, pin_memory=True,
|
||||
collate_fn=collate_fn, batch_sampler=train_sampler,
|
||||
persistent_workers=True,prefetch_factor=4) #128G Memory suitable loader.
|
||||
train_loader = DataLoader(
|
||||
train_dataset,
|
||||
num_workers=16,
|
||||
shuffle=False,
|
||||
pin_memory=True,
|
||||
collate_fn=collate_fn,
|
||||
batch_sampler=train_sampler,
|
||||
persistent_workers=True,
|
||||
prefetch_factor=4,
|
||||
) # 128G Memory suitable loader.
|
||||
if rank == 0:
|
||||
eval_dataset = TextAudioSpeakerLoader(hps.data.validation_files, hps.data)
|
||||
eval_loader = DataLoader(eval_dataset, num_workers=0, shuffle=False,
|
||||
batch_size=1, pin_memory=True,
|
||||
drop_last=False, collate_fn=collate_fn)
|
||||
if "use_noise_scaled_mas" in hps.model.keys() and hps.model.use_noise_scaled_mas == True:
|
||||
eval_loader = DataLoader(
|
||||
eval_dataset,
|
||||
num_workers=0,
|
||||
shuffle=False,
|
||||
batch_size=1,
|
||||
pin_memory=True,
|
||||
drop_last=False,
|
||||
collate_fn=collate_fn,
|
||||
)
|
||||
if (
|
||||
"use_noise_scaled_mas" in hps.model.keys()
|
||||
and hps.model.use_noise_scaled_mas == True
|
||||
):
|
||||
print("Using noise scaled MAS for VITS2")
|
||||
use_noise_scaled_mas = True
|
||||
mas_noise_scale_initial = 0.01
|
||||
@@ -84,19 +102,27 @@ def run():
|
||||
use_noise_scaled_mas = False
|
||||
mas_noise_scale_initial = 0.0
|
||||
noise_scale_delta = 0.0
|
||||
if "use_duration_discriminator" in hps.model.keys() and hps.model.use_duration_discriminator == True:
|
||||
if (
|
||||
"use_duration_discriminator" in hps.model.keys()
|
||||
and hps.model.use_duration_discriminator == True
|
||||
):
|
||||
print("Using duration discriminator for VITS2")
|
||||
use_duration_discriminator = True
|
||||
net_dur_disc = DurationDiscriminator(
|
||||
hps.model.hidden_channels,
|
||||
hps.model.hidden_channels,
|
||||
3,
|
||||
0.1,
|
||||
gin_channels=hps.model.gin_channels if hps.data.n_speakers != 0 else 0,
|
||||
).cuda(rank)
|
||||
if "use_spk_conditioned_encoder" in hps.model.keys() and hps.model.use_spk_conditioned_encoder == True:
|
||||
hps.model.hidden_channels,
|
||||
hps.model.hidden_channels,
|
||||
3,
|
||||
0.1,
|
||||
gin_channels=hps.model.gin_channels if hps.data.n_speakers != 0 else 0,
|
||||
).cuda(rank)
|
||||
if (
|
||||
"use_spk_conditioned_encoder" in hps.model.keys()
|
||||
and hps.model.use_spk_conditioned_encoder == True
|
||||
):
|
||||
if hps.data.n_speakers == 0:
|
||||
raise ValueError("n_speakers must be > 0 when using spk conditioned encoder to train multi-speaker model")
|
||||
raise ValueError(
|
||||
"n_speakers must be > 0 when using spk conditioned encoder to train multi-speaker model"
|
||||
)
|
||||
use_spk_conditioned_encoder = True
|
||||
else:
|
||||
print("Using normal encoder for VITS1")
|
||||
@@ -107,27 +133,31 @@ def run():
|
||||
hps.data.filter_length // 2 + 1,
|
||||
hps.train.segment_size // hps.data.hop_length,
|
||||
n_speakers=hps.data.n_speakers,
|
||||
mas_noise_scale_initial = mas_noise_scale_initial,
|
||||
noise_scale_delta = noise_scale_delta,
|
||||
**hps.model).cuda(rank)
|
||||
mas_noise_scale_initial=mas_noise_scale_initial,
|
||||
noise_scale_delta=noise_scale_delta,
|
||||
**hps.model,
|
||||
).cuda(rank)
|
||||
|
||||
net_d = MultiPeriodDiscriminator(hps.model.use_spectral_norm).cuda(rank)
|
||||
optim_g = torch.optim.AdamW(
|
||||
filter(lambda p: p.requires_grad, net_g.parameters()),
|
||||
hps.train.learning_rate,
|
||||
betas=hps.train.betas,
|
||||
eps=hps.train.eps)
|
||||
eps=hps.train.eps,
|
||||
)
|
||||
optim_d = torch.optim.AdamW(
|
||||
net_d.parameters(),
|
||||
hps.train.learning_rate,
|
||||
betas=hps.train.betas,
|
||||
eps=hps.train.eps)
|
||||
eps=hps.train.eps,
|
||||
)
|
||||
if net_dur_disc is not None:
|
||||
optim_dur_disc = torch.optim.AdamW(
|
||||
net_dur_disc.parameters(),
|
||||
hps.train.learning_rate,
|
||||
betas=hps.train.betas,
|
||||
eps=hps.train.eps)
|
||||
net_dur_disc.parameters(),
|
||||
hps.train.learning_rate,
|
||||
betas=hps.train.betas,
|
||||
eps=hps.train.eps,
|
||||
)
|
||||
else:
|
||||
optim_dur_disc = None
|
||||
net_g = DDP(net_g, device_ids=[rank], find_unused_parameters=True)
|
||||
@@ -136,40 +166,82 @@ def run():
|
||||
net_dur_disc = DDP(net_dur_disc, device_ids=[rank], find_unused_parameters=True)
|
||||
try:
|
||||
if net_dur_disc is not None:
|
||||
_, _, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "DUR_*.pth"), net_dur_disc, optim_dur_disc, skip_optimizer=True)
|
||||
_, optim_g, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "G_*.pth"), net_g,
|
||||
optim_g, skip_optimizer=True)
|
||||
_, optim_d, _, epoch_str = utils.load_checkpoint(utils.latest_checkpoint_path(hps.model_dir, "D_*.pth"), net_d,
|
||||
optim_d, skip_optimizer=True)
|
||||
|
||||
_, _, _, epoch_str = utils.load_checkpoint(
|
||||
utils.latest_checkpoint_path(hps.model_dir, "DUR_*.pth"),
|
||||
net_dur_disc,
|
||||
optim_dur_disc,
|
||||
skip_optimizer=True,
|
||||
)
|
||||
_, optim_g, _, epoch_str = utils.load_checkpoint(
|
||||
utils.latest_checkpoint_path(hps.model_dir, "G_*.pth"),
|
||||
net_g,
|
||||
optim_g,
|
||||
skip_optimizer=True,
|
||||
)
|
||||
_, optim_d, _, epoch_str = utils.load_checkpoint(
|
||||
utils.latest_checkpoint_path(hps.model_dir, "D_*.pth"),
|
||||
net_d,
|
||||
optim_d,
|
||||
skip_optimizer=True,
|
||||
)
|
||||
|
||||
epoch_str = max(epoch_str, 1)
|
||||
global_step = (epoch_str - 1) * len(train_loader)
|
||||
except Exception as e:
|
||||
print(e)
|
||||
epoch_str = 1
|
||||
global_step = 0
|
||||
print(e)
|
||||
epoch_str = 1
|
||||
global_step = 0
|
||||
|
||||
|
||||
scheduler_g = torch.optim.lr_scheduler.ExponentialLR(optim_g, gamma=hps.train.lr_decay, last_epoch=epoch_str - 2)
|
||||
scheduler_d = torch.optim.lr_scheduler.ExponentialLR(optim_d, gamma=hps.train.lr_decay, last_epoch=epoch_str - 2)
|
||||
scheduler_g = torch.optim.lr_scheduler.ExponentialLR(
|
||||
optim_g, gamma=hps.train.lr_decay, last_epoch=epoch_str - 2
|
||||
)
|
||||
scheduler_d = torch.optim.lr_scheduler.ExponentialLR(
|
||||
optim_d, gamma=hps.train.lr_decay, last_epoch=epoch_str - 2
|
||||
)
|
||||
if net_dur_disc is not None:
|
||||
scheduler_dur_disc = torch.optim.lr_scheduler.ExponentialLR(optim_dur_disc, gamma=hps.train.lr_decay, last_epoch=epoch_str-2)
|
||||
scheduler_dur_disc = torch.optim.lr_scheduler.ExponentialLR(
|
||||
optim_dur_disc, gamma=hps.train.lr_decay, last_epoch=epoch_str - 2
|
||||
)
|
||||
else:
|
||||
scheduler_dur_disc = None
|
||||
scaler = GradScaler(enabled=hps.train.fp16_run)
|
||||
|
||||
for epoch in range(epoch_str, hps.train.epochs + 1):
|
||||
if rank == 0:
|
||||
train_and_evaluate(rank, epoch, hps, [net_g, net_d, net_dur_disc], [optim_g, optim_d, optim_dur_disc], [scheduler_g, scheduler_d, scheduler_dur_disc], scaler, [train_loader, eval_loader], logger, [writer, writer_eval])
|
||||
train_and_evaluate(
|
||||
rank,
|
||||
epoch,
|
||||
hps,
|
||||
[net_g, net_d, net_dur_disc],
|
||||
[optim_g, optim_d, optim_dur_disc],
|
||||
[scheduler_g, scheduler_d, scheduler_dur_disc],
|
||||
scaler,
|
||||
[train_loader, eval_loader],
|
||||
logger,
|
||||
[writer, writer_eval],
|
||||
)
|
||||
else:
|
||||
train_and_evaluate(rank, epoch, hps, [net_g, net_d, net_dur_disc], [optim_g, optim_d, optim_dur_disc], [scheduler_g, scheduler_d, scheduler_dur_disc], scaler, [train_loader, None], None, None)
|
||||
train_and_evaluate(
|
||||
rank,
|
||||
epoch,
|
||||
hps,
|
||||
[net_g, net_d, net_dur_disc],
|
||||
[optim_g, optim_d, optim_dur_disc],
|
||||
[scheduler_g, scheduler_d, scheduler_dur_disc],
|
||||
scaler,
|
||||
[train_loader, None],
|
||||
None,
|
||||
None,
|
||||
)
|
||||
scheduler_g.step()
|
||||
scheduler_d.step()
|
||||
if net_dur_disc is not None:
|
||||
scheduler_dur_disc.step()
|
||||
|
||||
|
||||
def train_and_evaluate(rank, epoch, hps, nets, optims, schedulers, scaler, loaders, logger, writers):
|
||||
def train_and_evaluate(
|
||||
rank, epoch, hps, nets, optims, schedulers, scaler, loaders, logger, writers
|
||||
):
|
||||
net_g, net_d, net_dur_disc = nets
|
||||
optim_g, optim_d, optim_dur_disc = optims
|
||||
scheduler_g, scheduler_d, scheduler_dur_disc = schedulers
|
||||
@@ -184,13 +256,34 @@ def train_and_evaluate(rank, epoch, hps, nets, optims, schedulers, scaler, loade
|
||||
net_d.train()
|
||||
if net_dur_disc is not None:
|
||||
net_dur_disc.train()
|
||||
for batch_idx, (x, x_lengths, spec, spec_lengths, y, y_lengths, speakers, tone, language, bert, ja_bert) in tqdm(enumerate(train_loader)):
|
||||
for batch_idx, (
|
||||
x,
|
||||
x_lengths,
|
||||
spec,
|
||||
spec_lengths,
|
||||
y,
|
||||
y_lengths,
|
||||
speakers,
|
||||
tone,
|
||||
language,
|
||||
bert,
|
||||
ja_bert,
|
||||
) in tqdm(enumerate(train_loader)):
|
||||
if net_g.module.use_noise_scaled_mas:
|
||||
current_mas_noise_scale = net_g.module.mas_noise_scale_initial - net_g.module.noise_scale_delta * global_step
|
||||
current_mas_noise_scale = (
|
||||
net_g.module.mas_noise_scale_initial
|
||||
- net_g.module.noise_scale_delta * global_step
|
||||
)
|
||||
net_g.module.current_mas_noise_scale = max(current_mas_noise_scale, 0.0)
|
||||
x, x_lengths = x.cuda(rank, non_blocking=True), x_lengths.cuda(rank, non_blocking=True)
|
||||
spec, spec_lengths = spec.cuda(rank, non_blocking=True), spec_lengths.cuda(rank, non_blocking=True)
|
||||
y, y_lengths = y.cuda(rank, non_blocking=True), y_lengths.cuda(rank, non_blocking=True)
|
||||
x, x_lengths = x.cuda(rank, non_blocking=True), x_lengths.cuda(
|
||||
rank, non_blocking=True
|
||||
)
|
||||
spec, spec_lengths = spec.cuda(rank, non_blocking=True), spec_lengths.cuda(
|
||||
rank, non_blocking=True
|
||||
)
|
||||
y, y_lengths = y.cuda(rank, non_blocking=True), y_lengths.cuda(
|
||||
rank, non_blocking=True
|
||||
)
|
||||
speakers = speakers.cuda(rank, non_blocking=True)
|
||||
tone = tone.cuda(rank, non_blocking=True)
|
||||
language = language.cuda(rank, non_blocking=True)
|
||||
@@ -198,16 +291,37 @@ def train_and_evaluate(rank, epoch, hps, nets, optims, schedulers, scaler, loade
|
||||
ja_bert = ja_bert.cuda(rank, non_blocking=True)
|
||||
|
||||
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), (hidden_x, logw, logw_) = net_g(x, x_lengths, spec, spec_lengths, speakers, tone, language, bert, ja_bert)
|
||||
(
|
||||
y_hat,
|
||||
l_length,
|
||||
attn,
|
||||
ids_slice,
|
||||
x_mask,
|
||||
z_mask,
|
||||
(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,
|
||||
ja_bert,
|
||||
)
|
||||
mel = spec_to_mel_torch(
|
||||
spec,
|
||||
hps.data.filter_length,
|
||||
hps.data.n_mel_channels,
|
||||
hps.data.sampling_rate,
|
||||
hps.data.mel_fmin,
|
||||
hps.data.mel_fmax)
|
||||
y_mel = commons.slice_segments(mel, ids_slice, hps.train.segment_size // hps.data.hop_length)
|
||||
hps.data.mel_fmax,
|
||||
)
|
||||
y_mel = commons.slice_segments(
|
||||
mel, ids_slice, hps.train.segment_size // hps.data.hop_length
|
||||
)
|
||||
y_hat_mel = mel_spectrogram_torch(
|
||||
y_hat.squeeze(1),
|
||||
hps.data.filter_length,
|
||||
@@ -216,26 +330,38 @@ def train_and_evaluate(rank, epoch, hps, nets, optims, schedulers, scaler, loade
|
||||
hps.data.hop_length,
|
||||
hps.data.win_length,
|
||||
hps.data.mel_fmin,
|
||||
hps.data.mel_fmax
|
||||
hps.data.mel_fmax,
|
||||
)
|
||||
|
||||
y = commons.slice_segments(y, ids_slice * hps.data.hop_length, hps.train.segment_size) # slice
|
||||
y = commons.slice_segments(
|
||||
y, ids_slice * hps.data.hop_length, hps.train.segment_size
|
||||
) # slice
|
||||
|
||||
# Discriminator
|
||||
y_d_hat_r, y_d_hat_g, _, _ = net_d(y, y_hat.detach())
|
||||
with autocast(enabled=False):
|
||||
loss_disc, losses_disc_r, losses_disc_g = discriminator_loss(y_d_hat_r, y_d_hat_g)
|
||||
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(hidden_x.detach(), x_mask.detach(), logw.detach(), logw_.detach())
|
||||
y_dur_hat_r, y_dur_hat_g = net_dur_disc(
|
||||
hidden_x.detach(), x_mask.detach(), logw.detach(), logw_.detach()
|
||||
)
|
||||
with autocast(enabled=False):
|
||||
# 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)
|
||||
# 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()
|
||||
scaler.scale(loss_dur_disc_all).backward()
|
||||
scaler.unscale_(optim_dur_disc)
|
||||
grad_norm_dur_disc = commons.clip_grad_value_(net_dur_disc.parameters(), None)
|
||||
grad_norm_dur_disc = commons.clip_grad_value_(
|
||||
net_dur_disc.parameters(), None
|
||||
)
|
||||
scaler.step(optim_dur_disc)
|
||||
|
||||
optim_d.zero_grad()
|
||||
@@ -269,51 +395,97 @@ def train_and_evaluate(rank, epoch, hps, nets, optims, schedulers, scaler, loade
|
||||
|
||||
if rank == 0:
|
||||
if global_step % hps.train.log_interval == 0:
|
||||
lr = optim_g.param_groups[0]['lr']
|
||||
lr = optim_g.param_groups[0]["lr"]
|
||||
losses = [loss_disc, loss_gen, loss_fm, loss_mel, loss_dur, loss_kl]
|
||||
logger.info('Train Epoch: {} [{:.0f}%]'.format(
|
||||
epoch,
|
||||
100. * batch_idx / len(train_loader)))
|
||||
logger.info(
|
||||
"Train Epoch: {} [{:.0f}%]".format(
|
||||
epoch, 100.0 * batch_idx / len(train_loader)
|
||||
)
|
||||
)
|
||||
logger.info([x.item() for x in losses] + [global_step, lr])
|
||||
|
||||
scalar_dict = {"loss/g/total": loss_gen_all, "loss/d/total": loss_disc_all, "learning_rate": lr,
|
||||
"grad_norm_d": grad_norm_d, "grad_norm_g": grad_norm_g}
|
||||
scalar_dict = {
|
||||
"loss/g/total": loss_gen_all,
|
||||
"loss/d/total": loss_disc_all,
|
||||
"learning_rate": lr,
|
||||
"grad_norm_d": grad_norm_d,
|
||||
"grad_norm_g": grad_norm_g,
|
||||
}
|
||||
scalar_dict.update(
|
||||
{"loss/g/fm": loss_fm, "loss/g/mel": loss_mel, "loss/g/dur": loss_dur, "loss/g/kl": loss_kl})
|
||||
scalar_dict.update({"loss/g/{}".format(i): v for i, v in enumerate(losses_gen)})
|
||||
scalar_dict.update({"loss/d_r/{}".format(i): v for i, v in enumerate(losses_disc_r)})
|
||||
scalar_dict.update({"loss/d_g/{}".format(i): v for i, v in enumerate(losses_disc_g)})
|
||||
|
||||
{
|
||||
"loss/g/fm": loss_fm,
|
||||
"loss/g/mel": loss_mel,
|
||||
"loss/g/dur": loss_dur,
|
||||
"loss/g/kl": loss_kl,
|
||||
}
|
||||
)
|
||||
scalar_dict.update(
|
||||
{"loss/g/{}".format(i): v for i, v in enumerate(losses_gen)}
|
||||
)
|
||||
scalar_dict.update(
|
||||
{"loss/d_r/{}".format(i): v for i, v in enumerate(losses_disc_r)}
|
||||
)
|
||||
scalar_dict.update(
|
||||
{"loss/d_g/{}".format(i): v for i, v in enumerate(losses_disc_g)}
|
||||
)
|
||||
|
||||
image_dict = {
|
||||
"slice/mel_org": utils.plot_spectrogram_to_numpy(y_mel[0].data.cpu().numpy()),
|
||||
"slice/mel_gen": utils.plot_spectrogram_to_numpy(y_hat_mel[0].data.cpu().numpy()),
|
||||
"all/mel": utils.plot_spectrogram_to_numpy(mel[0].data.cpu().numpy()),
|
||||
"all/attn": utils.plot_alignment_to_numpy(attn[0, 0].data.cpu().numpy())
|
||||
"slice/mel_org": utils.plot_spectrogram_to_numpy(
|
||||
y_mel[0].data.cpu().numpy()
|
||||
),
|
||||
"slice/mel_gen": utils.plot_spectrogram_to_numpy(
|
||||
y_hat_mel[0].data.cpu().numpy()
|
||||
),
|
||||
"all/mel": utils.plot_spectrogram_to_numpy(
|
||||
mel[0].data.cpu().numpy()
|
||||
),
|
||||
"all/attn": utils.plot_alignment_to_numpy(
|
||||
attn[0, 0].data.cpu().numpy()
|
||||
),
|
||||
}
|
||||
utils.summarize(
|
||||
writer=writer,
|
||||
global_step=global_step,
|
||||
images=image_dict,
|
||||
scalars=scalar_dict)
|
||||
scalars=scalar_dict,
|
||||
)
|
||||
|
||||
if global_step % hps.train.eval_interval == 0:
|
||||
evaluate(hps, net_g, eval_loader, writer_eval)
|
||||
utils.save_checkpoint(net_g, optim_g, hps.train.learning_rate, epoch,
|
||||
os.path.join(hps.model_dir, "G_{}.pth".format(global_step)))
|
||||
utils.save_checkpoint(net_d, optim_d, hps.train.learning_rate, epoch,
|
||||
os.path.join(hps.model_dir, "D_{}.pth".format(global_step)))
|
||||
utils.save_checkpoint(
|
||||
net_g,
|
||||
optim_g,
|
||||
hps.train.learning_rate,
|
||||
epoch,
|
||||
os.path.join(hps.model_dir, "G_{}.pth".format(global_step)),
|
||||
)
|
||||
utils.save_checkpoint(
|
||||
net_d,
|
||||
optim_d,
|
||||
hps.train.learning_rate,
|
||||
epoch,
|
||||
os.path.join(hps.model_dir, "D_{}.pth".format(global_step)),
|
||||
)
|
||||
if net_dur_disc is not None:
|
||||
utils.save_checkpoint(net_dur_disc, optim_dur_disc, hps.train.learning_rate, epoch, os.path.join(hps.model_dir, "DUR_{}.pth".format(global_step)))
|
||||
keep_ckpts = getattr(hps.train, 'keep_ckpts', 5)
|
||||
utils.save_checkpoint(
|
||||
net_dur_disc,
|
||||
optim_dur_disc,
|
||||
hps.train.learning_rate,
|
||||
epoch,
|
||||
os.path.join(hps.model_dir, "DUR_{}.pth".format(global_step)),
|
||||
)
|
||||
keep_ckpts = getattr(hps.train, "keep_ckpts", 5)
|
||||
if keep_ckpts > 0:
|
||||
utils.clean_checkpoints(path_to_models=hps.model_dir, n_ckpts_to_keep=keep_ckpts, sort_by_time=True)
|
||||
|
||||
utils.clean_checkpoints(
|
||||
path_to_models=hps.model_dir,
|
||||
n_ckpts_to_keep=keep_ckpts,
|
||||
sort_by_time=True,
|
||||
)
|
||||
|
||||
global_step += 1
|
||||
|
||||
if rank == 0:
|
||||
logger.info('====> Epoch: {}'.format(epoch))
|
||||
|
||||
logger.info("====> Epoch: {}".format(epoch))
|
||||
|
||||
|
||||
def evaluate(hps, generator, eval_loader, writer_eval):
|
||||
@@ -322,7 +494,19 @@ def evaluate(hps, generator, eval_loader, writer_eval):
|
||||
audio_dict = {}
|
||||
print("Evaluating ...")
|
||||
with torch.no_grad():
|
||||
for batch_idx, (x, x_lengths, spec, spec_lengths, y, y_lengths, speakers, tone, language, bert, ja_bert) in enumerate(eval_loader):
|
||||
for batch_idx, (
|
||||
x,
|
||||
x_lengths,
|
||||
spec,
|
||||
spec_lengths,
|
||||
y,
|
||||
y_lengths,
|
||||
speakers,
|
||||
tone,
|
||||
language,
|
||||
bert,
|
||||
ja_bert,
|
||||
) in enumerate(eval_loader):
|
||||
x, x_lengths = x.cuda(), x_lengths.cuda()
|
||||
spec, spec_lengths = spec.cuda(), spec_lengths.cuda()
|
||||
y, y_lengths = y.cuda(), y_lengths.cuda()
|
||||
@@ -332,7 +516,18 @@ def evaluate(hps, generator, eval_loader, writer_eval):
|
||||
tone = tone.cuda()
|
||||
language = language.cuda()
|
||||
for use_sdp in [True, False]:
|
||||
y_hat, attn, mask, *_ = generator.module.infer(x, x_lengths, speakers, tone, language, bert, ja_bert, y=spec, max_len=1000, sdp_ratio=0.0 if not use_sdp else 1.0)
|
||||
y_hat, attn, mask, *_ = generator.module.infer(
|
||||
x,
|
||||
x_lengths,
|
||||
speakers,
|
||||
tone,
|
||||
language,
|
||||
bert,
|
||||
ja_bert,
|
||||
y=spec,
|
||||
max_len=1000,
|
||||
sdp_ratio=0.0 if not use_sdp else 1.0,
|
||||
)
|
||||
y_hat_lengths = mask.sum([1, 2]).long() * hps.data.hop_length
|
||||
|
||||
mel = spec_to_mel_torch(
|
||||
@@ -341,7 +536,8 @@ def evaluate(hps, generator, eval_loader, writer_eval):
|
||||
hps.data.n_mel_channels,
|
||||
hps.data.sampling_rate,
|
||||
hps.data.mel_fmin,
|
||||
hps.data.mel_fmax)
|
||||
hps.data.mel_fmax,
|
||||
)
|
||||
y_hat_mel = mel_spectrogram_torch(
|
||||
y_hat.squeeze(1).float(),
|
||||
hps.data.filter_length,
|
||||
@@ -350,25 +546,40 @@ def evaluate(hps, generator, eval_loader, writer_eval):
|
||||
hps.data.hop_length,
|
||||
hps.data.win_length,
|
||||
hps.data.mel_fmin,
|
||||
hps.data.mel_fmax
|
||||
hps.data.mel_fmax,
|
||||
)
|
||||
image_dict.update({
|
||||
f"gen/mel_{batch_idx}": utils.plot_spectrogram_to_numpy(y_hat_mel[0].cpu().numpy())
|
||||
})
|
||||
audio_dict.update({
|
||||
f"gen/audio_{batch_idx}_{use_sdp}": y_hat[0, :, :y_hat_lengths[0]]
|
||||
})
|
||||
image_dict.update({f"gt/mel_{batch_idx}": utils.plot_spectrogram_to_numpy(mel[0].cpu().numpy())})
|
||||
audio_dict.update({f"gt/audio_{batch_idx}": y[0, :, :y_lengths[0]]})
|
||||
image_dict.update(
|
||||
{
|
||||
f"gen/mel_{batch_idx}": utils.plot_spectrogram_to_numpy(
|
||||
y_hat_mel[0].cpu().numpy()
|
||||
)
|
||||
}
|
||||
)
|
||||
audio_dict.update(
|
||||
{
|
||||
f"gen/audio_{batch_idx}_{use_sdp}": y_hat[
|
||||
0, :, : y_hat_lengths[0]
|
||||
]
|
||||
}
|
||||
)
|
||||
image_dict.update(
|
||||
{
|
||||
f"gt/mel_{batch_idx}": utils.plot_spectrogram_to_numpy(
|
||||
mel[0].cpu().numpy()
|
||||
)
|
||||
}
|
||||
)
|
||||
audio_dict.update({f"gt/audio_{batch_idx}": y[0, :, : y_lengths[0]]})
|
||||
|
||||
utils.summarize(
|
||||
writer=writer_eval,
|
||||
global_step=global_step,
|
||||
images=image_dict,
|
||||
audios=audio_dict,
|
||||
audio_sampling_rate=hps.data.sampling_rate
|
||||
audio_sampling_rate=hps.data.sampling_rate,
|
||||
)
|
||||
generator.train()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
run()
|
||||
|
||||
Reference in New Issue
Block a user