-
Notifications
You must be signed in to change notification settings - Fork 0
/
hjvad.py
239 lines (221 loc) · 10.1 KB
/
hjvad.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
import pdb
import logging
import collections
import contextlib
import sys
import wave
import argparse
import torch
from torch.autograd import Variable
from torch.nn.functional import softmax
import transforms.librosa2 as lr
from torchvision.transforms import *
from transforms import *
#import webrtcvad
from datasets import CLASSES2 as _CLASS
import hjlog
DEF_SAMPLE_RATE = 10000
DEF_SAMPLE_WIDTH = 1
DEF_DURATION = 0.02
DEF_PADDING = 0.2
DEF_N_REQ = int(10000 * 2)
DEF_N_DURATION = int(DEF_N_REQ / ( DEF_DURATION * DEF_SAMPLE_RATE))
LOG = logging.getLogger(__name__)
class Nnvad(object):
def __init__(self, sample_time = 0.02, n_mels = 32, n_fft=80, hop_length=10):
self.transform = Compose([FixAudioLength(time = sample_time),
ToMelSpectrogram(n_mels = n_mels, n_fft=n_fft, hop_length=hop_length),
ToTensor('mel_spectrogram', 'input')])
self.model = torch.load('torch_vad.model')
self.model.float()
def is_speech(self, bytes, sample_rate = DEF_SAMPLE_RATE, sample_width = DEF_SAMPLE_WIDTH):
samples, _sample_rate= lr.loadfrombuff(bytes, sample_rate, sample_width)
rs = self.transform({
'samples' : samples,
'sample_rate' : _sample_rate
})
input = Variable(torch.unsqueeze(rs['input'].unsqueeze(0), 1))
output = self.model(input)
out_softmax = softmax(output, dim=1)
return _CLASS[torch.argmax(out_softmax)] == _CLASS[1]
def read_wave(path):
with contextlib.closing(wave.open(path, 'rb')) as wf:
num_channels = wf.getnchannels()
assert num_channels == 1
sample_rate = wf.getframerate()
pcm_data = wf.readframes(wf.getnframes())
return pcm_data, sample_rate
def read_wave_queue(path):
from queue import Queue
queues = [Queue()]
data, rate = read_wave(path)
l_data = len(data)
for i in range(0, l_data, 200):
if i+200 <= l_data:
frame_data = data[i: i+200];
isspeech = vad.is_speech(frame_data)
frame = Frame(frame_data, isspeech = isspeech)
queues[0].put(frame)
LOG.debug('read_wave_queue end***********')
return queues
def write_wave(path, audio, sample_width = DEF_SAMPLE_WIDTH, sample_rate = DEF_SAMPLE_RATE):
with contextlib.closing(wave.open(path, 'wb')) as wf:
wf.setnchannels(1)
wf.setsampwidth(sample_width)
wf.setframerate(sample_rate)
wf.writeframes(audio)
class Frame(object):
def __init__(self, bytes, timestamp=None, duration=None, isspeech=None):
self.bytes = bytes
self.timestamp = timestamp
self.duration = duration
self.isspeech = isspeech
def frame_generator(audio, frame_duration_ms = DEF_DURATION,
sample_rate = DEF_SAMPLE_RATE, sample_width = DEF_SAMPLE_WIDTH):
n = int(sample_rate * frame_duration_ms * sample_width)
offset = 0
timestamp = 0.0
duration = (float(n) / sample_rate)/sample_width
while offset + n < len(audio):
yield Frame(audio[offset:offset + n], timestamp, duration)
timestamp += duration
offset += n * sample_width
def socket_frame_generator(request, n_request = DEF_N_REQ, frame_duration_ms = DEF_DURATION,
sample_rate = DEF_SAMPLE_RATE, sample_width = DEF_SAMPLE_WIDTH):
n_duration_bytes = int(sample_rate * frame_duration_ms * sample_width)
timestamp = 0.0
duration = (float(n_duration_bytes) / sample_rate)/sample_width
n_remain = 0
count_ = 0
while True:
if n_remain <= 0:
count_+=1
request.sendall(b'1')
n_remain = n_request
buffer = b''
bytes_recv = request.recv(n_remain)
if bytes_recv.strip():
n_remain = n_remain - len(bytes_recv)
buffer += bytes_recv
while len(buffer) >= n_duration_bytes:
yield Frame(buffer[ : n_duration_bytes], timestamp, duration)
buffer = buffer[n_duration_bytes : ]
timestamp += duration
def queue_frame_generator(queues, frame_duration_ms = DEF_DURATION,
sample_rate = DEF_SAMPLE_RATE, sample_width = DEF_SAMPLE_WIDTH):
n_queue = len(queues)
n_duration_bytes = int(sample_rate * frame_duration_ms * sample_width)
timestamp = 0.0
duration = (float(n_duration_bytes) / sample_rate)/sample_width
i_proc = 0
#while True:
while not queues[i_proc].empty():
frame = queues[i_proc].get()
frame.timestamp, frame.duration = timestamp, duration
#LOG.debug('queue_frame_generate %s yield %s*************************' % (i_proc, len(frame.bytes)))
yield frame
timestamp += duration
i_proc = (i_proc + 1) % n_queue
def vad_collector(vad, frames,
sample_rate = DEF_SAMPLE_RATE, sample_width = DEF_SAMPLE_WIDTH,
frame_duration_ms = DEF_DURATION,
padding_duration_ms = DEF_PADDING):
num_padding_frames = int(padding_duration_ms / frame_duration_ms)
ring_buffer = collections.deque(maxlen=num_padding_frames)
ring_buffer_vad_bool = collections.deque(maxlen=num_padding_frames)
triggered = False
voiced_frames = []
for frame in iter(frames):
isspeech='1' if vad.is_speech(frame.bytes, sample_rate, sample_width) else '0'
ring_buffer_vad_bool.append(isspeech)
#sys.stdout.write(isspeech)
if not triggered:
ring_buffer.append(frame)
num_voiced = len([f for f in ring_buffer_vad_bool if f == '1'])
if num_voiced > 0.9 * ring_buffer.maxlen:
#sys.stdout.write('+(%s)' % (ring_buffer[0].timestamp,))
triggered = True
voiced_frames.extend(ring_buffer)
ring_buffer.clear()
ring_buffer_vad_bool.clear()
else:
voiced_frames.append(frame)
ring_buffer.append(frame)
num_unvoiced = len([f for f in ring_buffer_vad_bool if f == '0'])
if num_unvoiced > 0.9 * ring_buffer.maxlen:
#sys.stdout.write('-(%s)' % (frame.timestamp + frame.duration))
triggered = False
LOG.debug('queue_vad_collect yield***********************')
yield b''.join([f.bytes for f in voiced_frames])
ring_buffer.clear()
ring_buffer_vad_bool.clear()
voiced_frames = []
#if triggered:
LOG.debug('queue_vad_collect end yield***********************')
#sys.stdout.write('-(%s)' % (frame.timestamp + frame.duration))
#sys.stdout.write('\n')
if voiced_frames:
yield b''.join([f.bytes for f in voiced_frames])
def queue_vad_collector(frames,
sample_rate = DEF_SAMPLE_RATE, sample_width = DEF_SAMPLE_WIDTH,
frame_duration_ms = DEF_DURATION,
padding_duration_ms = DEF_PADDING,
n_duration = DEF_N_DURATION, sense = 0.9):
num_padding_frames = int(padding_duration_ms / frame_duration_ms)
ring_buffer = collections.deque(maxlen = num_padding_frames)
ring_buffer_vad_bool = collections.deque(maxlen = num_padding_frames)
voiced_frames = collections.deque(maxlen = n_duration)
triggered = False
for frame in iter(frames):
ring_buffer_vad_bool.append(frame.isspeech)
LOG.debug('Is speech : %s' % frame.isspeech)
if not triggered:
ring_buffer.append(frame)
num_voiced = len([f for f in ring_buffer_vad_bool if f])
#LOG.debug('hejie num_voiced=%s' % num_voiced)
if num_voiced > sense * ring_buffer.maxlen:
#sys.stdout.write('+(%s)' % (ring_buffer[0].timestamp,))
triggered = True
voiced_frames.extend(ring_buffer)
ring_buffer.clear()
ring_buffer_vad_bool.clear()
else:
voiced_frames.append(frame)
ring_buffer.append(frame)
num_unvoiced = len([f for f in ring_buffer_vad_bool if not f])
#LOG.debug('hejie num_unvoiced=%s' % num_unvoiced)
if num_unvoiced > sense * ring_buffer.maxlen:
#sys.stdout.write('-(%s)' % (frame.timestamp + frame.duration))
triggered = False
yield b''.join([f.bytes for f in voiced_frames])
ring_buffer.clear()
ring_buffer_vad_bool.clear()
voiced_frames.clear()
#if triggered:
#sys.stdout.write('-(%s)' % (frame.timestamp + frame.duration))
#sys.stdout.write('\n')
if voiced_frames:
yield b''.join([f.bytes for f in voiced_frames])
vad = Nnvad()
def vad_split(queues):
frames = queue_frame_generator(queues)
segments = queue_vad_collector(frames)
for segment in iter(segments):
yield segment
if __name__ == '__main__':
#queues = read_wave_queue('datasets/speech_commands_esp/guandeng/20181209192400.wav')
#queues = read_wave_queue('datasets/speech_commands_esp/kaideng/20181209192108.wav')
#queues = read_wave_queue('datasets/speech_commands_esp/_background_noise_/20181209190233.wav')
#queues = read_wave_queue('datasets/speech_commands_esp/_background_noise_/20181209190241.wav')
#queues = read_wave_queue('datasets/speech_commands_esp/guankongtiao/20190106152103.wav')
#queues = read_wave_queue('datasets/speech_commands_esp/_background_noise_/20190106152632.wav')
queues = read_wave_queue('datasets/speech_commands_esp/_background_noise_/20190106144805.wav')
for segment in iter(vad_split(queues)):
print('--end')
#write_wave('chunck02.wav', segment)
if __name__ == '__main__2':
#audio, _sample_rate = read_wave('datasets/speech_commands_esp/_background_noise_/20181209190151.wav')
#audio, _sample_rate = read_wave('datasets/speech_commands_esp/_background_noise_/20181209190241.wav')
#audio, _sample_rate = read_wave('datasets/speech_commands_esp_vad/train/_background_noise_/13chunk20181209190151.wav')
audio, _sample_rate = read_wave('datasets/speech_commands_esp_vad/train/speech/13chunk20181209192108.wav')
LOG.debug(vad.is_speech(audio))