forked from riverphoenix/tacotron2
-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
143 lines (110 loc) · 3.7 KB
/
utils.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
# -*- coding: utf-8 -*-
#/usr/bin/python2
from __future__ import print_function
import numpy as np
import librosa
import copy
import matplotlib
import matplotlib.pyplot as plt
from scipy import signal
from matplotlib.pyplot import step, show
import librosa.display
from hyperparams import Hyperparams as hp
def spectrogram2wav(mag):
'''# Generate wave file from spectrogram'''
# transpose
mag = mag.T
# de-noramlize
mag = (np.clip(mag, 0, 1) * hp.max_db) - hp.max_db + hp.ref_db
# to amplitude
mag = librosa.db_to_power(mag)
# wav reconstruction
wav = griffin_lim(mag)
# de-preemphasis
wav = signal.lfilter([1], [1, -hp.preemphasis], wav)
# trim
wav, _ = librosa.effects.trim(wav)
return wav
def griffin_lim(spectrogram):
X_best = copy.deepcopy(spectrogram)
for i in range(hp.n_iter):
X_t = invert_spectrogram(X_best)
est = librosa.stft(X_t, hp.n_fft, hp.hop_length, win_length=hp.win_length)
phase = est / np.maximum(1e-8, np.abs(est))
X_best = spectrogram * phase
X_t = invert_spectrogram(X_best)
y = np.real(X_t)
return y
def invert_spectrogram(spectrogram):
return librosa.istft(spectrogram, hp.hop_length, win_length=hp.win_length, window="hann")
def plot_losses(config,Kmel_out,Ky1,Kdone_out,Ky2,Kmag_out,Ky3,gs):
plt.figure(figsize=(10, 10))
if hp.train_form == 'Both':
if Kdone_out is not None:
sizeP = 3
else:
sizeP = 2
elif hp.train_form == 'Encoder':
if Kdone_out is not None:
sizeP = 2
else:
sizeP = 1
else:
sizeP = 1
fig_cnt = 1
if Kmel_out is not None:
ax1 = plt.subplot(sizeP, 2, fig_cnt)
librosa.display.specshow(Kmel_out[0,:,:].T,y_axis='linear')
plt.title('Predicted mel')
plt.colorbar()
plt.tight_layout()
ax2 = plt.subplot(sizeP, 2, fig_cnt+1,sharey=ax1)
librosa.display.specshow(Ky1[0,:,:].T,y_axis='linear')
plt.title('Original mel')
plt.colorbar()
plt.tight_layout()
fig_cnt = fig_cnt + 2
if Kmag_out is not None:
ax3 = plt.subplot(sizeP, 2, fig_cnt)
librosa.display.specshow(Kmag_out[0,:,:].T,y_axis='linear')
plt.title('Predicted mag')
plt.colorbar()
plt.tight_layout()
ax4 = plt.subplot(sizeP, 2, fig_cnt+1,sharey=ax3)
librosa.display.specshow(Ky3[0,:,:].T,y_axis='linear')
plt.title('Original mag')
plt.colorbar()
plt.tight_layout()
fig_cnt = fig_cnt + 2
if Kdone_out is not None:
KDone = Kdone_out[0,:,:]
Kd = []
for i in range(KDone.shape[0]):
if KDone[i,0] > KDone[i,1]:
Kd.append(0)
else:
Kd.append(1)
ind = np.arange(len(Kd))
width = 1.0
ax5 = plt.subplot(sizeP, 2, fig_cnt)
ax5.bar(ind, Kd, width, color='r')
plt.title('Predicted Dones')
plt.tight_layout()
ax6 = plt.subplot(sizeP, 2, fig_cnt+1)
ax6.bar(ind, Ky2[0,:], width, color='r')
plt.title('Original Dones')
plt.tight_layout()
plt.savefig('{}/losses_{}.png'.format(config.log_dir, gs), format='png')
plt.close('all')
def plot_wavs(config,wavs,gs):
if len(wavs)!=0:
plt.figure(figsize=(10, 10))
for i in range(len(wavs)):
wav = wavs[i]
txt = str(wav[2])+':'+str(wav[0])
wv = wav[1]
plt.subplot(len(wavs),1, i+1)
librosa.display.waveplot(wv, sr=hp.sr)
plt.title(txt)
plt.savefig('{}/wavs_{}.png'.format(config.log_dir, gs), format='png')
plt.close('all')