-
Notifications
You must be signed in to change notification settings - Fork 6
/
Network_Training_MTFAA_full.py
170 lines (145 loc) · 6.91 KB
/
Network_Training_MTFAA_full.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
from typing import OrderedDict
from unicodedata import name
import torch
import torch.optim as optim
import numpy as np
from torch.optim.lr_scheduler import StepLR
from tqdm import tqdm
import soundfile as sf
from Dataloader import Dataset, collate_fn
from MTFAA_Net_full import MTFAA_Net as MTFAA
from MTFAA_Net_full_F_ASqbi import MTFAA_Net as MTFAA_ASqBi
from MTFAA_Net_full_local_atten import MTFAA_Net as MTFAA_LSA
torch.set_default_tensor_type(torch.FloatTensor)
from signal_processing import iSTFT_module_1_8
WINDOW = torch.sqrt(torch.hann_window(1536,device=device) + 1e-8)
import argparse
import librosa
from collections import OrderedDict
import os
class NoamOpt:
"Optim wrapper that implements rate."
def __init__(self, model_size, factor, warmup, optimizer):
self.optimizer = optimizer
self._step = 0
self.warmup = warmup
self.factor = factor
self.model_size = model_size
self._rate = 0
def step(self):
"Update parameters and rate"
self._step += 1
rate = self.rate()
for p in self.optimizer.param_groups:
p['lr'] = rate
self._rate = rate
self.optimizer.step()
def rate(self, step = None):
"Implement `lrate` above"
if step is None:
step = self._step
return self.factor * \
(self.model_size ** (-0.5) *
min(step ** (-0.5), step * self.warmup ** (-1.5)))
def main(args):
def Loss(y_pred, y_true, train = True, idx = -1, epoch = 0):
snr = torch.div(torch.mean(torch.square(y_pred - y_true), dim=1, keepdim=True),(torch.mean(torch.square(y_true), dim=1, keepdim=True) + 1e-7))
snr_loss = 10 * torch.log10(snr + 1e-7)
pred_stft = torch.stft(y_pred,1536,384,win_length=1536,window=WINDOW,center=True)
true_stft = torch.stft(y_true,1536,384,win_length=1536,window=WINDOW,center=True)
pred_stft_real, pred_stft_imag = pred_stft[:,:,:,0], pred_stft[:,:,:,1]
true_stft_real, true_stft_imag = true_stft[:,:,:,0], true_stft[:,:,:,1]
pred_mag = torch.sqrt(pred_stft_real**2 + pred_stft_imag**2 + 1e-12)
true_mag = torch.sqrt(true_stft_real**2 + true_stft_imag**2 + 1e-12)
pred_real_c = pred_stft_real / (pred_mag**(0.7))
pred_imag_c = pred_stft_imag / (pred_mag**(0.7))
true_real_c = true_stft_real / (true_mag**(0.7))
true_imag_c = true_stft_imag / (true_mag**(0.7))
real_loss = torch.mean((pred_real_c - true_real_c)**2)
imag_loss = torch.mean((pred_imag_c - true_imag_c)**2)
mag_loss = torch.mean((pred_mag**(0.3)-true_mag**(0.3))**2)
return 0.3*(real_loss + imag_loss) + 0.7*mag_loss, snr_loss
device = torch.device(args.device if torch.cuda.is_available() else "cpu")
'''model'''
if args.model == 'MTFAA':
model = MTFAA()
elif args.model == 'MTFAA_ASqBi'
model = MTFAA_ASqBi()
elif args.model == 'MTFAA_LSA':
model = MTFAA_LSA()
''' train from checkpoints'''
# checkpoint = torch.load('./.pth',map_location=device)
# model.load_state_dict(checkpoint['state_dict'])
model = model.to(device)
'''optimizer & lr_scheduler'''
optimizer = NoamOpt(model_size=32, factor=1., warmup=6000,
optimizer=torch.optim.Adam(model.parameters(), lr=0, betas=(0.9, 0.98), eps=1e-9))
'''load train data'''
dataset = Dataset(length_in_seconds=3.2, num_clip_per_epoch=4000, random_start_point=True, train=True)
data_loader = torch.utils.data.DataLoader(dataset, batch_size=3, shuffle=True, drop_last=True, num_workers=2)
'''start train'''
for epoch in range(args.epochs):
train_loss = []
asnr_loss = []
model.train()
dataset.sample()
dataset.train = True
dataset.random_start_point = True
idx = 0
'''train'''
print('epoch %s--training' %(epoch))
for i, data in enumerate(tqdm(data_loader)):
noisy, clean = data
noisy = noisy.to(device)
clean = clean.to(device)
optimizer.optimizer.zero_grad()
noisy_stft = torch.stft(noisy,1536,384,win_length=1536,window=WINDOW,center=True,return_complex=True)
enh_stft = model(noisy_stft)
enh_s = iSTFT_module_1_8(n_fft=1536, hop_length=384, win_length=1536,window=WINDOW,center = True,length = noisy.shape[-1])(enh_stft.permute([0, 3, 2, 1]).contiguous())
stft_loss, snr_loss = Loss(enh_s, clean, train=True, idx = idx, epoch = epoch)
loss_overall = stft_loss
loss_overall.backward()
torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=3)
optimizer.step()
train_loss.append(loss_overall.cpu().detach().numpy())
idx += 1
train_loss = np.mean(train_loss)
asnr_loss = np.mean(asnr_loss)
lr_scheduler.step()
'''eval'''
valid_loss = []
model.eval()
print('epoch %s--validating' %(epoch))
dataset.train = False
dataset.random_start_point = False
with torch.no_grad():
for i, data in enumerate(tqdm(data_loader)):
noisy, clean = data
noisy = noisy.to(device)
clean = clean.to(device)
noisy_stft = torch.stft(noisy,1536,384,win_length=1536,window=WINDOW,center=True, return_complex=True)
enh_stft = model(noisy_stft)
enh_s = iSTFT_module_1_8(n_fft=1536, hop_length=384, win_length=1536,window=WINDOW,center = True,length = noisy.shape[-1])(enh_stft.permute([0, 3, 2, 1]).contiguous())
stft_loss, snr_loss = Loss(enh_s, clean, train=True, idx = idx, epoch = epoch)
loss_overall = stft_loss
valid_loss.append(loss_overall.cpu().detach().numpy())
asnr_loss.append(snr_loss.cpu().detach().numpy())
valid_loss = np.mean(valid_loss)
asnr_loss = np.mean(asnr_loss)
print('train loss: %s, valid loss %s, snr loss: %s' %(train_loss, valid_loss, asnr_loss))
torch.save(
{'epoch': epoch,
'state_dict': model.state_dict(),
'optimizer': optimizer.optimizer.state_dict()},
args.chkpt_path + '/Epoch_%s_trainloss_%s_validloss_%s.pth' %(str(epoch), str(train_loss), str(valid_loss)))
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('-m', "--model", default='MTFAA',
help='Choose the model you wanna train: MTFAA, MTFAA_LSA or MTFAA_ASqBi')
parser.add_argument('-c', "--chkpt_path", default=None, help='Dir to save the checkpoint files')
parser.add_argument('-e', "--epochs", default='300',
help='Epochs for training')
parser.add_argument('-d', "--device", default='cuda:0',
help='Device used for training')
args = parser.parse_args()
main(args)