-
Notifications
You must be signed in to change notification settings - Fork 41
Expand file tree
/
Copy pathtrain_dcunet.py
More file actions
159 lines (134 loc) · 7 KB
/
train_dcunet.py
File metadata and controls
159 lines (134 loc) · 7 KB
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
import torch
import torch.nn as nn
import torch.optim as optim
import glob, os
import numpy as np
from torch.utils.data import DataLoader
from torchcontrib.optim import SWA
import PinkBlack.trainer
from DCUNet.constant import *
from DCUNet.noisedataset import NoiseDataset
from DCUNet.sedataset import SEDataset
from DCUNet.source_separator import SourceSeparator
from DCUNet.criterion import WeightedSDR
from DCUNet.metric import PESQ
args = PinkBlack.io.setup(default_args=dict(gpu="0",
batch_size=12,
train_signal="/data/jongho/data/2019challenge/ss/reference/16k/clean_trainset_28spk_wav/",
train_noise="/data/jongho/data/2019challenge/ss/reference/16k/noisy_trainset_28spk_wav/",
test_signal="/data/jongho/data/2019challenge/ss/reference/16k/clean_testset_wav/",
test_noise="/data/jongho/data/2019challenge/ss/reference/16k/noisy_testset_wav/",
sequence_length=16384,
num_step=100000,
validation_interval=500,
num_workers=0,
ckpt="unet/ckpt.pth",
model_complexity=45,
lr=0.01,
num_signal=0,
num_noise=0,
optimizer="adam",
lr_decay=0.5,
momentum=0,
multi_gpu=False,
complex=False,
model_depth=20,
swa=False,
loss="wsdr",
log_amp=False,
metric="pesq",
train_dataset="se",
valid_dataset="se",
preload=False, # Whether to load datasets on memory
padding_mode="reflect", # conv2d's padding mode
))
def get_dataset(args):
def get_wav(dir):
wavs = []
wavs.extend(glob.glob(os.path.join(dir, "**/*.wav"), recursive=True))
wavs.extend(glob.glob(os.path.join(dir, "**/*.flac"), recursive=True))
wavs.extend(glob.glob(os.path.join(dir, "**/*.pcm"), recursive=True))
return wavs
if args.train_dataset == "mix":
train_signals = get_wav(args.train_signal)
train_noises = get_wav(args.train_noise)
else:
train_signals = get_wav(args.train_signal)
train_noises = [signal.replace("clean", "noisy") for signal in train_signals]
if args.valid_dataset == "mix":
test_signals = get_wav(args.test_signal)
test_noises = get_wav(args.test_noise)
else:
test_signals = get_wav(args.test_signal)
test_noises = [signal.replace("clean", "noisy") for signal in test_signals]
if args.num_signal > 0:
train_signals = train_signals[:args.num_signal]
test_signals = test_signals[:args.num_signal]
if args.num_noise > 0:
train_noises = train_noises[:args.num_noise]
test_noises = test_noises[:args.num_noise]
if args.train_dataset == "mix":
train_dset = NoiseDataset(train_signals, train_noises, sequence_length=args.sequence_length, is_validation=False, preload=args.preload)
else:
train_noises = [signal.replace("clean", "noisy") for signal in train_signals]
train_dset = SEDataset(train_signals, train_noises, sequence_length=args.sequence_length, is_validation=False)
if args.valid_dataset == "mix":
rand = np.random.RandomState(0)
rand.shuffle(test_signals)
test_signals = test_signals[:1000]
valid_dset = NoiseDataset(test_signals, test_noises, sequence_length=args.sequence_length, is_validation=True, preload=args.preload)
else:
test_noises = [signal.replace("clean", "noisy") for signal in test_signals]
valid_dset = SEDataset(test_signals, test_noises, sequence_length=args.sequence_length, is_validation=True)
return dict(train_dset=train_dset,
valid_dset=valid_dset)
dset = get_dataset(args)
train_dset, valid_dset = dset['train_dset'], dset['valid_dset']
train_dl = DataLoader(train_dset, batch_size=args.batch_size, shuffle=True, num_workers=args.num_workers,
pin_memory=False)
valid_dl = DataLoader(valid_dset, batch_size=args.batch_size, shuffle=False, num_workers=args.num_workers,
pin_memory=False)
if args.loss == "wsdr":
loss = WeightedSDR()
else:
raise NotImplementedError(f"unknown loss ({args.loss})")
if args.metric == "pesq":
metric = PESQ()
else:
def metric(output, bd):
with torch.no_grad():
return -loss(output, bd)
net = SourceSeparator(complex=args.complex, model_complexity=args.model_complexity, model_depth=args.model_depth, log_amp=args.log_amp, padding_mode=args.padding_mode).cuda()
print(net)
if args.multi_gpu:
net = nn.DataParallel(net).cuda()
if args.optimizer == "adam":
optimizer = optim.Adam(filter(lambda x: x.requires_grad, net.parameters()), lr=args.lr)
elif args.optimizer == "sgd":
optimizer = optim.SGD(filter(lambda x: x.requires_grad, net.parameters()), lr=args.lr, momentum=args.momentum)
else:
raise ValueError(f"Unknown optimizer - {args.optimizer}")
if args.swa:
steps_per_epoch = args.validation_interval
optimizer = SWA(optimizer, swa_start=int(20) * steps_per_epoch, swa_freq=1 * steps_per_epoch)
if args.lr_decay >= 1 or args.lr_decay <= 0:
scheduler = None
else:
if args.optimizer == "swa":
scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer.optimizer, mode="max", patience=5, factor=args.lr_decay)
else:
scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode="max", patience=5, factor=args.lr_decay)
trainer = PinkBlack.trainer.Trainer(net,
criterion=loss,
metric=metric,
train_dataloader=train_dl,
val_dataloader=valid_dl,
ckpt=args.ckpt,
optimizer=optimizer,
lr_scheduler=scheduler,
is_data_dict=True,
logdir="log_se")
trainer.train(step=args.num_step, validation_interval=args.validation_interval)
if args.swa:
trainer.swa_apply(bn_update=True)
trainer.train(1, phases=['val'])