-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
171 lines (157 loc) · 8.22 KB
/
train.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
from config import ConfigArgs as args
import os, sys, glob
import random
import torch
import torch.nn as nn
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.optim
from torch.optim.lr_scheduler import MultiStepLR, LambdaLR
from torch.utils.data import DataLoader
from tqdm import tqdm, trange
from tensorboardX import SummaryWriter
import numpy as np
import pandas as pd
from collections import deque
from models import DCTTS
from data import SpeechDataset, t2m_ga_collate_fn, load_vocab_tool
from utils import att2img, plot_att, lr_policy
def main():
model = DCTTS(args).to(DEVICE)
print('Model {} is working...'.format(args.model_name))
print('{} threads are used...'.format(torch.get_num_threads()))
ckpt_dir = os.path.join(args.logdir, args.model_name)
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
# scheduler = MultiStepLR(optimizer, milestones=[50000, 150000, 300000], gamma=0.5) #
scheduler = LambdaLR(optimizer, lr_policy)
if not os.path.exists(ckpt_dir):
os.makedirs(os.path.join(ckpt_dir, 'A', 'train'))
if args.pretrained_path is not None:
print('Train with pretrained model {}'.format(args.pretrained_path))
state = torch.load(args.pretrained_path)
model.custom_load_state_dict(state['model'])
else:
print('Already exists. Retrain the model.')
ckpt = sorted(glob.glob(os.path.join(ckpt_dir, '*k.pth.tar')))[-1]
state = torch.load(ckpt)
model.load_state_dict(state['model'])
args.global_step = state['global_step']
optimizer.load_state_dict(state['optimizer'])
# scheduler.load_state_dict(state['scheduler'])
# model = torch.nn.DataParallel(model, device_ids=list(range(args.no_gpu))).to(DEVICE)
dataset = SpeechDataset(args.data_path, args.meta_train, mem_mode=args.mem_mode)
validset = SpeechDataset(args.data_path, args.meta_eval, mem_mode=args.mem_mode)
data_loader = DataLoader(dataset=dataset, batch_size=args.batch_size,
shuffle=True, collate_fn=t2m_ga_collate_fn,
drop_last=True, pin_memory=True)
valid_loader = DataLoader(dataset=validset, batch_size=args.test_batch,
shuffle=False, collate_fn=t2m_ga_collate_fn, pin_memory=True)
writer = SummaryWriter(ckpt_dir)
train(model, data_loader, valid_loader, optimizer, scheduler,
batch_size=args.batch_size, ckpt_dir=ckpt_dir, writer=writer)
return None
def train(model, data_loader, valid_loader, optimizer, scheduler, batch_size=32, ckpt_dir=None, writer=None, mode='1'):
epochs = 0
global_step = args.global_step
l1_criterion = nn.L1Loss() # default average
bd_criterion = nn.BCELoss()
GO_frames = torch.zeros([batch_size, 1, args.n_mels]).to(DEVICE) # (N, Ty/r, n_mels)
idx2char = load_vocab_tool(args.lang)[-1]
while global_step < args.max_step:
epoch_loss = 0
for step, (texts, mels, pmels, gas, f0) in tqdm(enumerate(data_loader), total=len(data_loader), unit='B', ncols=70, leave=False):
optimizer.zero_grad()
texts, mels, pmels, gas, f0 = texts.to(DEVICE), mels.to(DEVICE), pmels.to(DEVICE), gas.to(DEVICE), f0.to(DEVICE)
prev_mels = torch.cat((GO_frames, mels[:, :-1, :]), 1)
mels_hat, pmels_hat, A = model(texts, prev_mels, f0) # mels_hat: (N, Ty/r, n_mels), A: (N, Tx, Ty/r)
mel_loss = l1_criterion(mels_hat, mels)
bd_loss = bd_criterion(mels_hat, mels)
pmel_loss = l1_criterion(pmels_hat, pmels)
att_loss = torch.mean(A*gas)
loss = mel_loss + bd_loss + att_loss + pmel_loss
loss.backward()
# nn.utils.clip_grad_norm_(model.parameters(), 2.0)
optimizer.step()
if args.lr_decay:
scheduler.step()
if global_step % args.save_term == 0:
model.eval()
val_loss = evaluate(model, valid_loader, writer, global_step, args.test_batch)
save_model(model, optimizer, scheduler, val_loss, global_step, ckpt_dir) # save best 5 models
model.train()
global_step += 1
if args.log_mode:
# Summary
writer.add_scalar('train/mel_loss', mel_loss.item(), global_step)
writer.add_scalar('train/pmel_loss', pmel_loss.item(), global_step)
writer.add_scalar('train/lr', scheduler.get_lr()[0], global_step)
alignment = A[0:1].clone().cpu().detach().numpy()
guided_att = gas[0:1].clone().cpu().detach().numpy()
writer.add_image('train/alignments', att2img(alignment), global_step) # (Tx, Ty)
writer.add_image('train/guided_att', att2img(guided_att), global_step) # (Tx, Ty)
writer.add_scalar('train/ga_loss', att_loss, global_step)
# text = texts[0].cpu().detach().numpy()
# text = [idx2char[ch] for ch in text]
# plot_att(alignment[0], text, global_step, path=os.path.join(args.logdir, type(model).__name__, 'A', 'train'))
mel_hat = mels_hat[0:1].transpose(1,2)
mel = mels[0:1].transpose(1, 2)
writer.add_image('train/mel_hat', mel_hat, global_step)
writer.add_image('train/mel', mel, global_step)
# print('Training Loss: {}'.format(avg_loss))
epochs += 1
print('Training complete')
def evaluate(model, data_loader, writer, global_step, batch_size=100):
valid_loss = 0.
A = None
l1_loss = nn.L1Loss()
with torch.no_grad():
mel_sum_loss = 0.
pmel_sum_loss = 0.
for step, (texts, mels, pmels, gas, f0) in enumerate(data_loader):
texts, mels, pmels, gas, f0 = texts.to(DEVICE), mels.to(DEVICE), pmels.to(DEVICE), gas.to(DEVICE), f0.to(DEVICE)
GO_frames = torch.zeros([mels.shape[0], 1, args.n_mels]).to(DEVICE) # (N, Ty/r, n_mels)
prev_mels = torch.cat((GO_frames, mels[:, :-1, :]), 1)
mels_hat, pmels_hat, A = model(texts, prev_mels, f0) # mels_hat: (N, Ty/r, n_mels), A: (N, Tx, Ty/r)
mel_loss = l1_loss(mels_hat, mels)
pmel_loss = l1_loss(pmels_hat, pmels)
att_loss = torch.mean(A*gas)
mel_sum_loss += mel_loss.item()
pmel_sum_loss += pmel_loss.item()
mel_avg_loss = mel_sum_loss / (len(data_loader))
pmel_avg_loss = pmel_sum_loss / (len(data_loader))
writer.add_scalar('eval/mel_loss', mel_avg_loss, global_step)
writer.add_scalar('eval/pmel_loss', pmel_avg_loss, global_step)
writer.add_scalar('eval/ga_loss', att_loss, global_step)
alignment = A[0:1].clone().cpu().detach().numpy()
guided_att = gas[0:1].clone().cpu().detach().numpy()
writer.add_image('eval/alignments', att2img(alignment), global_step) # (Tx, Ty)
writer.add_image('eval/guided_att', att2img(guided_att), global_step) # (Tx, Ty)
# text = texts[0].cpu().detach().numpy()
# text = [load_vocab_tool(args.lang)[-1][ch] for ch in text]
# plot_att(alignment[0], text, global_step, path=os.path.join(args.logdir, args.model_name, 'A'))
writer.add_image('eval/mel_hat', mels_hat[0:1].transpose(1,2), global_step)
writer.add_image('eval/mel', mels[0:1].transpose(1,2), global_step)
writer.add_image('eval/pmel_hat', pmels_hat[0:1].transpose(1,2), global_step)
writer.add_image('eval/pmel', pmels[0:1].transpose(1,2), global_step)
return mel_avg_loss
def save_model(model, optimizer, scheduler, val_loss, global_step, ckpt_dir):
fname = '{}-{:03d}k.pth'.format(args.model_name, global_step//1000)
state = {
'global_step': global_step,
'model': model.state_dict(),
'loss': val_loss,
'optimizer': optimizer.state_dict(),
# 'scheduler': scheduler.state_dict(),
}
torch.save(state, os.path.join(ckpt_dir, fname))
if __name__ == '__main__':
gpu_id = int(sys.argv[1])
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "{}".format(gpu_id)
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Set random seem for reproducibility
seed = 999
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
main()