forked from G-Wang/WaveRNN-Pytorch
-
Notifications
You must be signed in to change notification settings - Fork 37
/
audio.py
159 lines (121 loc) · 5.16 KB
/
audio.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
import librosa
import librosa.filters
import math
import numpy as np
from scipy import signal
from hparams import hparams
from scipy.io import wavfile
# r9r9 preprocessing
import lws
def load_wav(path):
return librosa.load(path, sr=hparams.sample_rate)[0]
def save_wav(wav, path):
wav = wav * 32767 / max(0.01, np.max(np.abs(wav)))
wavfile.write(path, hparams.sample_rate, wav.astype(np.int16))
def preemphasis(x):
from nnmnkwii.preprocessing import preemphasis
return preemphasis(x, hparams.preemphasis)
def inv_preemphasis(x):
from nnmnkwii.preprocessing import inv_preemphasis
return inv_preemphasis(x, hparams.preemphasis)
def spectrogram(y):
D = _lws_processor().stft(preemphasis(y)).T
S = _amp_to_db(np.abs(D) ** hparams.magnitude_power) - hparams.ref_level_db
return _normalize(S)
def inv_spectrogram(spectrogram):
'''Converts spectrogram to waveform using librosa'''
S = _db_to_amp(_denormalize(spectrogram) + hparams.ref_level_db) # Convert back to linear
processor = _lws_processor()
D = processor.run_lws(S.astype(np.float64).T ** (1/hparams.magnitude_power))
y = processor.istft(D).astype(np.float32)
return inv_preemphasis(y)
def _stft(y):
if hparams.use_lws:
return _lws_processor(hparams).stft(y).T
else:
return librosa.stft(y=y, n_fft=hparams.n_fft, hop_length=hparams.hop_size, win_length=hparams.win_size, pad_mode='constant')
# def melspectrogram(y):
# D = _stft(preemphasis(y))
# S = _amp_to_db(_linear_to_mel(np.abs(D)**hparams.magnitude_power)) - hparams.ref_level_db
# if not hparams.allow_clipping_in_normalization:
# assert S.max() <= 0 and S.min() - hparams.min_level_db >= 0
# return _normalize(S)
def melspectrogram(y):
D = _stft(preemphasis(y))
S = _amp_to_db(_linear_to_mel(np.abs(D))) - hparams.ref_level_db
return _normalize(S)
def _lws_processor():
return lws.lws(hparams.win_size, hparams.hop_size, mode="speech")
# Conversions:
_mel_basis = None
def _linear_to_mel(spectrogram):
global _mel_basis
if _mel_basis is None:
_mel_basis = _build_mel_basis()
return np.dot(_mel_basis, spectrogram)
def _build_mel_basis():
if hparams.fmax is not None:
assert hparams.fmax <= hparams.sample_rate // 2
return librosa.filters.mel(hparams.sample_rate, hparams.n_fft,
fmin=hparams.fmin, fmax=hparams.fmax,
n_mels=hparams.num_mels)
def _amp_to_db(x):
min_level = np.exp(hparams.min_level_db / 20 * np.log(10))
return 20 * np.log10(np.maximum(min_level, x))
def _db_to_amp(x):
return np.power(10.0, x * 0.05)
# def _normalize(S):
# return np.clip((S - hparams.min_level_db) / -hparams.min_level_db, 0, 1)
#
#
# def _denormalize(S):
# return (np.clip(S, 0, 1) * -hparams.min_level_db) + hparams.min_level_db
#
# def _normalize(S):
# if hparams.allow_clipping_in_normalization:
# if hparams.symmetric_mels:
# return np.clip((2 * hparams.max_abs_value) * ((S - hparams.min_level_db) / (-hparams.min_level_db)) - hparams.max_abs_value,
# -hparams.max_abs_value, hparams.max_abs_value)
# else:
# return np.clip(hparams.max_abs_value * ((S - hparams.min_level_db) / (-hparams.min_level_db)), 0, hparams.max_abs_value)
#
# assert S.max() <= 0 and S.min() - hparams.min_level_db >= 0
# if hparams.symmetric_mels:
# return (2 * hparams.max_abs_value) * ((S - hparams.min_level_db) / (-hparams.min_level_db)) - hparams.max_abs_value
# else:
# return hparams.max_abs_value * ((S - hparams.min_level_db) / (-hparams.min_level_db))
#
# def _denormalize(D):
# if hparams.allow_clipping_in_normalization:
# if hparams.symmetric_mels:
# return (((np.clip(D, -hparams.max_abs_value,
# hparams.max_abs_value) + hparams.max_abs_value) * -hparams.min_level_db / (2 * hparams.max_abs_value))
# + hparams.min_level_db)
# else:
# return ((np.clip(D, 0, hparams.max_abs_value) * -hparams.min_level_db / hparams.max_abs_value) + hparams.min_level_db)
#
# if hparams.symmetric_mels:
# return (((D + hparams.max_abs_value) * -hparams.min_level_db / (2 * hparams.max_abs_value)) + hparams.min_level_db)
# else:
# return ((D * -hparams.min_level_db / hparams.max_abs_value) + hparams.min_level_db)
def _normalize(S):
# symmetric mels
return 2 * hparams.max_abs_value * ((S - hparams.min_level_db) / -hparams.min_level_db) - hparams.max_abs_value
def _denormalize(S):
# symmetric mels
return ((S + hparams.max_abs_value) * -hparams.min_level_db) / (2 * hparams.max_abs_value) + hparams.min_level_db
# Fatcord's preprocessing
def quantize(x):
"""quantize audio signal
"""
x = np.clip(x, -1., 1.)
quant = ((x + 1.)/2.) * (2**hparams.bits - 1)
return quant.astype(np.int)
# testing
def test_everything():
wav = np.random.randn(12000,)
mel = melspectrogram(wav)
spec = spectrogram(wav)
quant = quantize(wav)
print(wav.shape, mel.shape, spec.shape, quant.shape)
print(quant.max(), quant.min(), mel.max(), mel.min(), spec.max(), spec.min())