-
Notifications
You must be signed in to change notification settings - Fork 29
/
utils.py
253 lines (206 loc) · 8.69 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
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
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
# -*- coding: utf-8 -*-
#/usr/bin/python2
'''
By kyubyong park. [email protected].
https://www.github.com/kyubyong/kss
'''
from __future__ import print_function, division
import numpy as np
import librosa
import os, copy
import matplotlib
matplotlib.use('pdf')
import matplotlib.pyplot as plt
from scipy import signal
from hyperparams import Hyperparams as hp
import tensorflow as tf
def get_spectrograms(fpath):
'''Parse the wave file in `fpath` and
Returns normalized melspectrogram and linear spectrogram.
Args:
fpath: A string. The full path of a sound file.
Returns:
mel: A 2d array of shape (T, n_mels) and dtype of float32.
mag: A 2d array of shape (T, 1+n_fft/2) and dtype of float32.
'''
# Loading sound file
y, sr = librosa.load(fpath, sr=hp.sr)
# Trimming
y, _ = librosa.effects.trim(y, top_db=40)
# Preemphasis
y = np.append(y[0], y[1:] - hp.preemphasis * y[:-1])
# stft
linear = librosa.stft(y=y,
n_fft=hp.n_fft,
hop_length=hp.hop_length,
win_length=hp.win_length)
# magnitude spectrogram
mag = np.abs(linear) # (1+n_fft//2, T)
# mel spectrogram
mel_basis = librosa.filters.mel(hp.sr, hp.n_fft, hp.n_mels) # (n_mels, 1+n_fft//2)
mel = np.dot(mel_basis, mag) # (n_mels, t)
# to decibel
mel = 20 * np.log10(np.maximum(1e-5, mel))
mag = 20 * np.log10(np.maximum(1e-5, mag))
# normalize
mel = np.clip((mel - hp.ref_db + hp.max_db) / hp.max_db, 1e-8, 1)
mag = np.clip((mag - hp.ref_db + hp.max_db) / hp.max_db, 1e-8, 1)
# Transpose
mel = mel.T.astype(np.float32) # (T, n_mels)
mag = mag.T.astype(np.float32) # (T, 1+n_fft//2)
return mel, mag
def spectrogram2wav(mag):
'''# Generate wave file from linear magnitude spectrogram
Args:
mag: A numpy array of (T, 1+n_fft//2)
Returns:
wav: A 1-D numpy array.
'''
# transpose
mag = mag.T
# de-noramlize
mag = (np.clip(mag, 0, 1) * hp.max_db) - hp.max_db + hp.ref_db
# to amplitude
mag = np.power(10.0, mag * 0.05)
# wav reconstruction
wav = griffin_lim(mag**hp.power)
# de-preemphasis
wav = signal.lfilter([1], [1, -hp.preemphasis], wav)
# trim
wav = trim(wav)
return wav.astype(np.float32)
def griffin_lim(spectrogram):
'''Applies Griffin-Lim's raw.'''
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):
'''Applies inverse fft.
Args:
spectrogram: [1+n_fft//2, t]
'''
return librosa.istft(spectrogram, hp.hop_length, win_length=hp.win_length, window="hann")
def plot_alignment(alignment, gs, dir=hp.logdir):
"""Plots the alignment.
Args:
alignment: A numpy array with shape of (encoder_steps, decoder_steps)
gs: (int) global step.
dir: Output path.
"""
if not os.path.exists(dir): os.mkdir(dir)
fig, ax = plt.subplots()
im = ax.imshow(alignment)
fig.colorbar(im)
plt.title('{} Steps'.format(gs))
plt.savefig('{}/alignment_{}.png'.format(dir, gs), format='png')
def guided_attention(g=0.2):
'''Guided attention. Refer to page 3 on the paper.'''
W = np.zeros((hp.max_N, hp.max_T), dtype=np.float32)
for n_pos in range(W.shape[0]):
for t_pos in range(W.shape[1]):
W[n_pos, t_pos] = 1 - np.exp(-(t_pos / float(hp.max_T) - n_pos / float(hp.max_N)) ** 2 / (2 * g * g))
return W
def learning_rate_decay(init_lr, global_step, warmup_steps = 4000.0):
'''Noam scheme from tensor2tensor'''
step = tf.to_float(global_step + 1)
return init_lr * warmup_steps**0.5 * tf.minimum(step * warmup_steps**-1.5, step**-0.5)
def load_spectrograms(fpath):
'''Read the wave file in `fpath`
and extracts spectrograms'''
fname = os.path.basename(fpath)
mel, mag = get_spectrograms(fpath)
t = mel.shape[0]
# Marginal padding for reduction shape sync.
num_paddings = hp.r - (t % hp.r) if t % hp.r != 0 else 0
mel = np.pad(mel, [[0, num_paddings], [0, 0]], mode="constant")
mag = np.pad(mag, [[0, num_paddings], [0, 0]], mode="constant")
# Reduction
mel = mel[::hp.r, :]
return fname, mel, mag
#This is adapted by
# https://github.com/keithito/tacotron/blob/master/util/audio.py#L55-62
def trim(wav, top_db=40, min_silence_sec=0.8):
frame_length = int(hp.sr * min_silence_sec)
hop_length = int(frame_length / 4)
endpoint = librosa.effects.split(wav, frame_length=frame_length,
hop_length=hop_length,
top_db=top_db)[0, 1]
return wav[:endpoint]
def load_j2hcj():
'''
Arg:
jamo: A Hangul Jamo character(0x01100-0x011FF)
Returns:
A dictionary that converts jamo into Hangul Compatibility Jamo(0x03130 - 0x0318F) Character
'''
jamo = u'''␀␃ !,.?ᄀᄁᄂᄃᄄᄅᄆᄇᄈᄉᄊᄋᄌᄍᄎᄏᄐᄑ하ᅢᅣᅤᅥᅦᅧᅨᅩᅪᅫᅬᅭᅮᅯᅰᅱᅲᅳᅴᅵᆨᆩᆪᆫᆬᆭᆮᆯᆰᆱᆲᆴᆶᆷᆸᆹᆺᆻᆼᆽᆾᆿᇀᇁᇂ'''
hcj = u'''␀␃ !,.?ㄱㄲㄴㄷㄸㄹㅁㅂㅃㅅㅆㅇㅈㅉㅊㅋㅌㅍㅎㅏㅐㅑㅒㅓㅔㅕㅖㅗㅘㅙㅚㅛㅜㅝㅞㅟㅠㅡㅢㅣㄱㄲㄳㄴㄵㄶㄷㄹㄺㄻㄼㄾㅀㅁㅂㅄㅅㅆㅇㅈㅊㅋㅌㅍㅎ'''
assert len(jamo) == len(hcj)
j2hcj = {j: h for j, h in zip(jamo, hcj)}
return j2hcj
def load_j2sj():
'''
Arg:
jamo: A Hangul Jamo character(0x01100-0x011FF)
Returns:
A dictionary that decomposes double consonants into two single consonants.
'''
jamo = u'''␀␃ !,.?ᄀᄁᄂᄃᄄᄅᄆᄇᄈᄉᄊᄋᄌᄍᄎᄏᄐᄑ하ᅢᅣᅤᅥᅦᅧᅨᅩᅪᅫᅬᅭᅮᅯᅰᅱᅲᅳᅴᅵᆨᆩᆪᆫᆬᆭᆮᆯᆰᆱᆲᆴᆶᆷᆸᆹᆺᆻᆼᆽᆾᆿᇀᇁᇂ'''
sj = u'''␀|␃| |!|,|.|?|ᄀ|ᄀᄀ|ᄂ|ᄃ|ᄃᄃ|ᄅ|ᄆ|ᄇ|ᄇᄇ|ᄉ|ᄉᄉ|ᄋ|ᄌ|ᄌᄌ|ᄎ|ᄏ|ᄐ|ᄑ|ᄒ|ᅡ|ᅢ|ᅣ|ᅤ|ᅥ|ᅦ|ᅧ|ᅨ|ᅩ|ᅪ|ᅫ|ᅬ|ᅭ|ᅮ|ᅯ|ᅰ|ᅱ|ᅲ|ᅳ|ᅴ|ᅵ|ᆨ|ᆨᆨ|ᆨᆺ|ᆫ|ᆫᆽ|ᆫᇂ|ᆮ|ᆯ|ᆯᆨ|ᆯᆷ|ᆯᆸ|ᆯᇀ|ᆯᇂ|ᆷ|ᆸ|ᆸᆺ|ᆺ|ᆺᆺ|ᆼ|ᆽ|ᆾ|ᆿ|ᇀ|ᇁ|ᇂ'''
assert len(jamo)==len(sj.split("|"))
j2sj = {j: s for j, s in zip(jamo, sj.split("|"))}
return j2sj
def load_j2shcj():
'''
Arg:
jamo: A Hangul Jamo character(0x01100-0x011FF)
Returns:
A dictionary that converts jamo into Hangul Compatibility Jamo(0x03130 - 0x0318F) Character.
Double consonants are further decomposed into single consonants.
'''
jamo = u'''␀␃ !,.?ᄀᄁᄂᄃᄄᄅᄆᄇᄈᄉᄊᄋᄌᄍᄎᄏᄐᄑ하ᅢᅣᅤᅥᅦᅧᅨᅩᅪᅫᅬᅭᅮᅯᅰᅱᅲᅳᅴᅵᆨᆩᆪᆫᆬᆭᆮᆯᆰᆱᆲᆴᆶᆷᆸᆹᆺᆻᆼᆽᆾᆿᇀᇁᇂ'''
shcj = u'''␀|␃| |!|,|.|?|ㄱ|ㄱㄱ|ㄴ|ㄷ|ㄷㄷ|ㄹ|ㅁ|ㅂ|ㅂㅂ|ㅅ|ㅅㅅ|ㅇ|ㅈ|ㅈㅈ|ㅊ|ㅋ|ㅌ|ㅍ|ㅎ|ㅏ|ㅐ|ㅑ|ㅒ|ㅓ|ㅔ|ㅕ|ㅖ|ㅗ|ㅘ|ㅙ|ㅚ|ㅛ|ㅜ|ㅝ|ㅞ|ㅟ|ㅠ|ㅡ|ㅢ|ㅣ|ㄱ|ㄱㄱ|ㄱㅅ|ㄴ|ㄴㅈ|ㄴㅎ|ㄷ|ㄹ|ㄹㄱ|ㄹㅁ|ㄹㅂ|ㄹㅌ|ㄹㅎ|ㅁ|ㅂ|ㅂㅅ|ㅅ|ㅅㅅ|ㅇ|ㅈ|ㅊ|ㅋ|ㅌ|ㅍ|ㅎ'''
assert len(jamo)==len(shcj.split("|"))
j2shcj = {j: s for j, s in zip(jamo, shcj.split("|"))}
return j2shcj
# def jamo2syl(jamoset):
# """Inspired by Jamo.
# Return the Hangul character for the given jamo characters.
# """
# _JAMO_LEAD_OFFSET = 0x10ff
# _JAMO_VOWEL_OFFSET = 0x1160
# _JAMO_TAIL_OFFSET = 0x11a7
# _JAMO_OFFSET = 44032
# assert len(jamoset) in (2, 3), "CANNOT compose a syllable!"
#
# onset = ord(jamoset[0]) - _JAMO_LEAD_OFFSET
# vowel = ord(jamoset[1]) - _JAMO_VOWEL_OFFSET
# coda = ord(jamoset[2]) - _JAMO_TAIL_OFFSET if len(jamoset)==3 else 0
# return unichr(coda + (vowel - 1) * 28 + (onset - 1) * 588 + _JAMO_OFFSET)
#
# def compose(jamotext):
# converted, jamoset = "", ""
# for char in jamotext:
# codepoint = ord(char)
# if 4352 <= codepoint <= 4447: # Hangul Onset
# if jamoset:
# converted += jamo2syl(jamoset)
# jamoset = char
# elif 4448 <= codepoint <= 4519: # Hangul Nuclues
# jamoset += char
# elif 4520 <= codepoint <= 4607: # Hangul Coda
# jamoset += char
# converted += jamo2syl(jamoset)
# jamoset = ""
# else: # Non Hangul
# if jamoset:
# converted += jamo2syl(jamoset)
# jamoset = ""
# converted += char
# return converted