-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathest_gss1.py
198 lines (164 loc) · 9.84 KB
/
est_gss1.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
#coding:utf-8
# trial estimation of glottal source spectrum condition by inverse radiation filter and anti-formant filter
# under following hypotheses.
# (1) glottal source spectrum (frequency response) characterizes simply descending rightwards without sharp peak.
# (2) resonance strength of formant is roughly same level, regardless of formant frequency.
# 声門の音源の周波数特性（スペクトル）は 険しいピークがない右肩下がりの特性である
# フォルマントの共鳴の強さは周波数によらず同じ程度である
# の仮説のもと、
# 口の放射特性の逆フィルターとフォルマント周波数で減衰するフィルターを使って
# 声門の音源のスペクトルの状態を予想する
import argparse
import numpy as np
from scipy import signal
from scipy.io.wavfile import read as wavread
from matplotlib import pyplot as plt
from get_fp4 import *
from BPF import *
from iir_LowShelving1 import *
from iir_peaking1 import *
from diff_ana import *
from fft_ana import *
from glottal2mass import *
# Check version
# Python 3.6.4 on win32 (Windows 10)
# numpy 1.14.0
# matplotlib 2.1.1
# scipy 1.0.0
class Class_estimate_gss1(object):
def __init__(self, path0): # sampling_rate=16000):
# initalize
sr, y = wavread(path0)
self.yg= y / (2 ** 15)
self.sr= sr
print ('sampling rate ', sr)
self.fp0=Class_get_fp()
# calculate lpc log-spectrum and formant, Q
spec_out, fout, pout, Qout, fout_index, Low_index, High_index = self.fp0.get_fp(path0)
self.fout= fout
self.Qout= Qout
self.gain= np.ones( self.Qout.shape)
self.NFRAME=self.fp0.NFRAME
self.NSHIFT=self.fp0.NSHIFT
print ('formant freq and Q, per frame')
print ( self.fout )
print ( self.Qout)
# apply a low shelving filter as inverse filter against high pass filter that simulates radiation from mouth
self.iir_LS1=Class_IIR_LowShelving1(sampling_rate=self.sr)
self.invhpf_wav= self.iir_LS1.filtering(self.yg)
self.invhpf_wav /= (np.amax(np.abs(self.invhpf_wav)) / np.amax(np.abs(self.yg))) # normalize
def analysis(self, frame_num=None, gain_pattern=1, figure_show = True):
# only process one frame, if frame_num is specified
for l in range(self.fout.shape[0]):
if frame_num is not None:
if frame_num >= 0 and l != frame_num:
continue
# compute start point and end point of current l-th frame
sp= self.NSHIFT * l
ep= sp + self.NFRAME
if ep > len(self.yg):
ep= len(self.yg)
print ('frame no.', l, ' start[ms]', int(sp * 1000 / self.sr))
# process BPF
self.bpf1=Class_BPF(fc=self.fout[l,0], Q=self.Qout[l,0], sampling_rate=self.sr)
self.bpf2=Class_BPF(fc=self.fout[l,1], Q=self.Qout[l,1], sampling_rate=self.sr)
self.bpf3=Class_BPF(fc=self.fout[l,2], Q=self.Qout[l,2], sampling_rate=self.sr)
self.bpf4=Class_BPF(fc=self.fout[l,3], Q=self.Qout[l,3], sampling_rate=self.sr)
self.bpf_list=[self.bpf1, self.bpf2, self.bpf3, self.bpf4]
# process BPF, filtering independently
self.f1_wav=self.bpf_list[0].iir2(self.yg)
self.f2_wav=self.bpf_list[1].iir2(self.yg)
self.f3_wav=self.bpf_list[2].iir2(self.yg)
self.f4_wav=self.bpf_list[3].iir2(self.yg)
self.filtering_list=[ self.f1_wav, self.f2_wav, self.f3_wav, self.f4_wav]
# set drop gain of iir peaking filter as anti-formant boost.
# Try several assumed gain patterns to study appropriate drop gain.
self.gain_pattern= gain_pattern
if self.gain_pattern == 1:
# （パターン１）フォルマントの強さ(ゲイン)は、周波数によらず同じ程度と仮定する。
# (Pattern 1)resonance strength of formant is roughly the same regardless of the frequency.
#
# all -20dB(=0.1) drop
self.gain= np.ones( self.Qout.shape) * 0.1
self.analysis_sub(l, sp, ep, figure_show=figure_show)
if self.gain_pattern == 2:
# （パターン２）鼻音効果で 2kHz以上のフォルマントの強さ（ゲイン）は弱まっている仮定とする。
# 2KHzは可変値。
# (Pattern 2)due to nose effect, resonance strength of the formant over 2kHz (adjustable) become weak.
# When formant frequency > 2KHz -10dB(=0.3162), other -20dB(=0.1) drop
self.gain= np.ones( self.Qout.shape) * 0.1
highside_gain=-10
self.gain[np.where(self.fout > 2000)]= np.power(10.0, ( highside_gain /20))
self.analysis_sub(l, sp, ep, figure_show=figure_show)
def analysis_sub(self, l, sp, ep, figure_show):
# instance peaking drop filter
self.pk1=Class_IIR_Peaking1(fpeak=self.fout[l,0], gain=self.gain[l,0], Q=self.Qout[l,0] , sampling_rate=self.sr)
self.pk2=Class_IIR_Peaking1(fpeak=self.fout[l,1], gain=self.gain[l,1], Q=self.Qout[l,1] , sampling_rate=self.sr)
self.pk3=Class_IIR_Peaking1(fpeak=self.fout[l,2], gain=self.gain[l,2], Q=self.Qout[l,2] , sampling_rate=self.sr)
self.pk4=Class_IIR_Peaking1(fpeak=self.fout[l,3], gain=self.gain[l,3], Q=self.Qout[l,3] , sampling_rate=self.sr)
self.pk_list=[ self.pk1, self.pk2, self.pk3, self.pk4]
# process filtering in series
self.pk1_wav=self.pk_list[0].filtering(self.invhpf_wav)
self.pk2_wav=self.pk_list[1].filtering(self.pk1_wav)
self.pk3_wav=self.pk_list[2].filtering(self.pk2_wav)
self.pk4_wav=self.pk_list[3].filtering(self.pk3_wav)
self.pk_filtering_list=[ self.pk1_wav, self.pk2_wav, self.pk3_wav, self.pk4_wav]
# get a pitch duration in the frame, in order to avoid fundamental frequency F0 influence of frequency response
# 基本周波数 F0の影響を除くため1ピッチ分の信号を取り出す
sub_sp, sub_ep = diff_ana( self.pk_filtering_list[-1] [sp:ep] , self.sr)
# show waveform
if figure_show:
self.plot_waveform2( l, sp, ep, sub_sp, sub_ep)
# generate same length pseudo glottal waveform as a reference
# 同じ長さのリファレンス波形（基準とする波形）として、疑似的な声門の波形を生成する
glo0=Class_Glottal2(length_points=(sub_ep-sub_sp), sampling_rate=self.sr)
#glo0=Class_Glottal2(length_points=(sub_ep-sub_sp),tclosed=5.0, trise=5.0, tfall=0.8, tdiff=1.0, gain0=0.9, sampling_rate=self.sr)
# get frequency response as glottal source spectrum
comment0=': gain pattern ' + str(self.gain_pattern)
fft_ana( self.pk_filtering_list[-1][sp+sub_sp: sp+sub_ep], glo0.yg, self.sr, comment0, show=True)
def plot_waveform2(self,loop, sp, ep, sub_sp, sub_ep):
# plot every waveform per frame
# set draw number
max_display=2 # no display of BPF output
fig = plt.figure()
"""
max_display=6 # display BPF output
fig = plt.figure(figsize=(6, 7)) # adjust draw display area size
"""
plt.subplot(max_display,1,1)
plt.xlabel('mSec')
plt.ylabel('level')
plt.title( 'frame no. ' + str(loop) + ': blue original: red inverse radiation filter' )
plt.plot( (np.arange(len(self.yg[sp:ep])) * 1000.0 / self.sr) , self.yg[sp:ep])
plt.plot( (np.arange(len(self.invhpf_wav[sp:ep])) * 1000.0 / self.sr) , self.invhpf_wav[sp:ep], color='r')
plt.subplot(max_display,1,2)
plt.xlabel('mSec')
plt.ylabel('level')
plt.title( 'frame no. ' + str(loop) + ': anti-formant filter: red cirles, selected pitch portion' )
plt.plot( (np.arange(len(self.pk_filtering_list[-1][sp:ep])) * 1000.0 / self.sr) , self.pk_filtering_list[-1][sp:ep], color='g')
if sub_sp != 0 and sub_ep != 0 :
indices1=np.array([sub_sp, sub_ep])
infections1 = self.pk_filtering_list[-1][sp:ep][indices1]
plt.plot( ( indices1 * 1000.0 / self.sr) , infections1, 'ro', ms=5)
for i in range (max_display-2):
plt.subplot(max_display,1,i+3)
plt.xlabel('mSec')
plt.ylabel('level')
plt.title( 'f' + str(i+1) + ': '+ str(self.fout[loop,i]) + '[Hz] bpf output')
plt.plot( (np.arange(len( self.filtering_list[i][sp:ep])) * 1000.0 / self.sr) , self.filtering_list[i][sp:ep])
plt.grid()
fig.tight_layout()
plt.show()
if __name__ == '__main__':
#
parser = argparse.ArgumentParser(description='estimation glottal source spectrum condition')
parser.add_argument('--wav_file', '-w', default='a_1-16k.wav', help='wav-file-name(mono,16bit)')
parser.add_argument('--frame', '-f', type=int, default=-1, help='specify the frame number, set negative value if ignore')
parser.add_argument('--gain', '-g', type=int, default=1, help='specify anti-formant drop gain pattern, set 1 if equal ')
args = parser.parse_args()
# examples:
# vowel /a/ : python3 est_gss1.py -w a_1-16k.wav -f 3 -g 1
# nasal voice /na/ : python3 est_gss1.py -w na_1-16k.wav -f 8 -g 2
# instance
ana=Class_estimate_gss1(args.wav_file )
ana.analysis( frame_num=args.frame, gain_pattern=args.gain )