-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathplot_enhance_wavs.py
165 lines (151 loc) · 6.5 KB
/
plot_enhance_wavs.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
from nn_se import inference
from nn_se.utils import audio
from nn_se.utils import misc_utils
from nn_se import FLAGS
from matplotlib import pyplot as plt
import os
from pathlib import Path
import numpy as np
import librosa
import scipy
import tqdm
from nn_se.utils.assess.core import calc_pesq, calc_stoi, calc_sdr, calc_SegSNR
smg = None
noisy_and_ref_list = [
# ('exp/paper_sample/0db_mix_ref/p265_015_nfree_0678.wav', 'exp/paper_sample/0db_mix_ref/p265_015.wav'),
# ('exp/paper_sample/0db_mix_ref/p265_026_nfree_0571.wav', 'exp/paper_sample/0db_mix_ref/p265_026.wav'),
# ('exp/paper_sample/0db_mix_ref/p265_038_nfree_0758.wav', 'exp/paper_sample/0db_mix_ref/p265_038.wav'),
# ('exp/paper_sample/0db_mix_ref/p267_087_nfree_0663.wav', 'exp/paper_sample/0db_mix_ref/p267_087.wav'),
# ('exp/paper_sample/voxceleb1_sample/mixed/id10001_1zcIwhmdeo4_00001.wav', None),
# ('exp/paper_sample/voxceleb1_sample/mixed/id10001_1zcIwhmdeo4_00002.wav', None),
# ('exp/paper_sample/voxceleb1_sample/mixed/id10001_1zcIwhmdeo4_00003.wav', None),
('exp/paper_sample/voxceleb1_sample/mixed/id10001_utrA-v8pPm4_00001.wav', None),
('exp/paper_sample/voxceleb1_sample/mixed/id10006_3RybHF5mX78_00001.wav', None),
# ('exp/paper_sample/voxceleb1_sample/mixed/id10006_3RybHF5mX78_00002.wav', None),
('exp/paper_sample/voxceleb1_sample/mixed/id10006_5tGaUGO_z50_00001.wav', None),
('exp/paper_sample/voxceleb1_sample/mixed/id10006_zQROl4ZsMVA_00002.wav', None),
]
save_dir = 'exp/paper_sample/voxceleb1_sample'
def magnitude_spectrum_librosa_stft(signal, NFFT, overlap):
signal = np.array(signal, dtype=np.float)
tmp = librosa.core.stft(signal,
n_fft=NFFT,
hop_length=NFFT-overlap,
window=scipy.signal.windows.hann)
tmp = np.absolute(tmp)
return tmp.T
def enhance_and_calcMetrics(noisy_and_ref):
figsize = (2.8, 1.6)
noisy_wav_dir, ref_wav_dir = noisy_and_ref
noisy_wav, sr = audio.read_audio(noisy_wav_dir)
ref_wav, sr = audio.read_audio(ref_wav_dir) if ref_wav_dir else (None, sr)
config_name = FLAGS.PARAM().config_name()
noisy_stem = Path(noisy_wav_dir).stem
global smg
if smg is None:
smg = inference.build_SMG(ckpt_name=os.path.join('exp', config_name, 'ckpt'))
enhanced_out = inference.enhance_one_wav(smg, noisy_wav)
enhanced_wav = enhanced_out.enhanced_wav
# enhanced_mag = enhanced_out.enhanced_mag
mask = enhanced_out.mask
## plot mask, enhanced_mag; save enhanced_wav; calc metrics [pesq]
name_prefix = "%s_%s" % (config_name, noisy_stem)
# calc metrics [pesq_noisy->pesq_enhanced | pesqi, stoi, sdr, ssnr]
if ref_wav is not None:
pesq_noisy = calc_pesq(ref_wav, noisy_wav, sr)
pesq_enhanced = calc_pesq(ref_wav, enhanced_wav, sr)
pesqi = pesq_enhanced - pesq_noisy
stoi_noisy = calc_stoi(ref_wav, noisy_wav, sr)
stoi_enhanced = calc_stoi(ref_wav, enhanced_wav, sr)
stoii = stoi_enhanced - stoi_noisy
sdr_noisy = calc_sdr(ref_wav, noisy_wav, sr)
sdr_enhanced = calc_sdr(ref_wav, enhanced_wav, sr)
sdri = sdr_enhanced - sdr_noisy
ssnr_noisy = calc_SegSNR(ref_wav/np.max(ref_wav), noisy_wav/np.max(noisy_wav), 256, 256)
ssnr_enhanced = calc_SegSNR(ref_wav/np.max(ref_wav), enhanced_wav/np.max(enhanced_wav), 256, 256)
ssnri = ssnr_enhanced - ssnr_noisy
metrics_eval_ans_f = "%s_metrics_eval_ans.log" % config_name
with open(os.path.join(save_dir, metrics_eval_ans_f), 'a') as f:
f.write(name_prefix+":\n")
f.write(" pesq: %.3f->%.3f, pesqi: %.3f.\n" % (pesq_noisy, pesq_enhanced, pesqi))
f.write(" stoi: %.3f->%.3f, stoii: %.3f.\n" % (stoi_noisy, stoi_enhanced, stoii))
f.write(" sdr : %.3f->%.3f, sdri : %.3f.\n" % (sdr_noisy, sdr_enhanced, sdri))
f.write(" ssnr: %.3f->%.3f, ssnri: %.3f.\n" % (ssnr_noisy, ssnr_enhanced, ssnri))
# get x_ticks
n_frame = np.shape(mask)[0]
n=0
i=0
x1 = []
x2 = []
while(n<n_frame):
x1.append(n)
x2.append(i)
n = n+125
i = i+1
# mask
plt.figure(figsize=figsize)
plt.pcolormesh(mask.T, vmin=-0.2, vmax=1.5)
plt.subplots_adjust(top=0.97, right=0.96, left=0.17, bottom=0.27)
plt.xlabel('Time(S)')
plt.ylabel('Frequency(Hz)')
plt.xticks(x1, x2)
plt.yticks((0,33,64,97,129), ("0","1k","2k","3k","4k"))
plt.colorbar()
plt.savefig(os.path.join(save_dir,"%s_%s" % (name_prefix, "mask.jpg")))
# plt.show()
plt.close()
# enhanced_mag
enhanced_mag = magnitude_spectrum_librosa_stft(enhanced_wav, 256, 64*3)
enhanced_mag = np.log(enhanced_mag*3+1e-2)
plt.figure(figsize=figsize)
# print(np.max(enhanced_mag), np.min(enhanced_mag))
plt.pcolormesh(enhanced_mag.T, cmap='hot', vmin=-4.5, vmax=2.5)
plt.subplots_adjust(top=0.97, right=0.96, left=0.17, bottom=0.27)
# plt.title('STFT Magnitude')
plt.xlabel('Time(S)')
plt.ylabel('Frequency(Hz)')
plt.xticks(x1, x2)
plt.yticks((0,33,64,97,129), ("0","1k","2k","3k","4k"))
plt.colorbar()
plt.savefig(os.path.join(save_dir,"%s_%s" % (name_prefix, "enhanced_mag.jpg")))
# plt.show()
plt.close()
# enhanced_wav
audio.write_audio(os.path.join(save_dir, "%s_%s" % (name_prefix, "enhanced.wav")), enhanced_wav, sr)
# noisy_mag
noisy_mag_file = os.path.join(save_dir, "%s_%s" % (noisy_stem, "noisy_mag.jpg"))
noisy_mag = magnitude_spectrum_librosa_stft(noisy_wav, 256, 64*3)
noisy_mag = np.log(noisy_mag*3+1e-2)
plt.figure(figsize=figsize)
plt.pcolormesh(noisy_mag.T, cmap='hot', vmin=-4.5, vmax=2.5)
plt.subplots_adjust(top=0.97, right=0.96, left=0.17, bottom=0.27)
# plt.title('STFT Magnitude')
plt.xlabel('Time(S)')
plt.ylabel('Frequency(Hz)')
plt.xticks(x1, x2)
plt.yticks((0,33,64,97,129), ("0","1k","2k","3k","4k"))
plt.colorbar()
plt.savefig(noisy_mag_file)
# plt.show()
plt.close()
# clean_mag
clean_mag_file = os.path.join(save_dir, "%s_%s" % (noisy_stem, "clean_mag.jpg"))
if ref_wav is not None:
clean_mag = magnitude_spectrum_librosa_stft(ref_wav, 256, 64*3)
clean_mag = np.log(clean_mag*3+1e-2)
plt.figure(figsize=figsize)
plt.pcolormesh(clean_mag.T, cmap='hot', vmin=-4.5, vmax=2.5)
plt.subplots_adjust(top=0.97, right=0.96, left=0.17, bottom=0.27)
# plt.title('STFT Magnitude')
plt.xlabel('Time(S)')
plt.ylabel('Frequency(Hz)')
plt.xticks(x1, x2)
plt.yticks((0,33,64,97,129), ("0","1k","2k","3k","4k"))
plt.colorbar()
plt.savefig(clean_mag_file)
# plt.show()
plt.close()
if __name__ == "__main__":
# enhance_and_calcMetrics(noisy_and_ref_list[0])
for noisy_and_ref in tqdm.tqdm(noisy_and_ref_list, ncols=100):
enhance_and_calcMetrics(noisy_and_ref)