-
Notifications
You must be signed in to change notification settings - Fork 10
/
SpeechModels.py
133 lines (102 loc) · 4.68 KB
/
SpeechModels.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
from tensorflow.keras.models import Model, load_model
from tensorflow.keras import layers as L
from tensorflow.keras import backend as K
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint, LearningRateScheduler
from tensorflow.keras import backend as K
from tensorflow.keras import optimizers
from vggish import VGGish
from kapre.time_frequency import Melspectrogram, Spectrogram
from kapre.utils import Normalization2D
def AttRNNSpeechModel(nCategories, samplingrate=16000,
inputLength=16000, unet = False, rnn_func=L.LSTM):
# simple LSTM
sr = samplingrate
iLen = inputLength
inputs = L.Input((inputLength,), name='input')
x = L.Reshape((1, -1))(inputs)
m = Melspectrogram(n_dft=1024, n_hop=128, input_shape=(1, iLen),
padding='same', sr=sr, n_mels=80,
fmin=40.0, fmax=sr / 2, power_melgram=1.0,
return_decibel_melgram=True, trainable_fb=False,
trainable_kernel=False,
name='mel_stft')
m.trainable = False
x = m(x)
x = Normalization2D(int_axis=0, name='mel_stft_norm')(x)
# note that Melspectrogram puts the sequence in shape (batch_size, melDim, timeSteps, 1)
# we would rather have it the other way around for LSTMs
x = L.Permute((2, 1, 3))(x)
if unet == True:
x = L.Conv2D(16, (5, 1), activation='relu', padding='same')(x)
up = L.BatchNormalization()(x)
x = L.Conv2D(32, (5, 1), activation='relu', padding='same')(up)
x = L.BatchNormalization()(x)
x = L.Conv2D(16, (5, 1), activation='relu', padding='same')(x)
down = L.BatchNormalization()(x)
merge = L.Concatenate(axis=3)([up,down])
x = L.Conv2D(1, (5, 1), activation='relu', padding='same')(merge)
x = L.BatchNormalization()(x)
else:
x = L.Conv2D(10, (5, 1), activation='relu', padding='same')(x)
x = L.BatchNormalization()(x)
x = L.Conv2D(1, (5, 1), activation='relu', padding='same')(x)
x = L.BatchNormalization()(x)
# x = Reshape((125, 80)) (x)
# keras.backend.squeeze(x, axis)
x = L.Lambda(lambda q: K.squeeze(q, -1), name='squeeze_last_dim')(x)
x = L.Bidirectional(rnn_func(64, return_sequences=True)
)(x) # [b_s, seq_len, vec_dim]
x = L.Bidirectional(rnn_func(64, return_sequences=True)
)(x) # [b_s, seq_len, vec_dim]
xFirst = L.Lambda(lambda q: q[:, -1])(x) # [b_s, vec_dim]
query = L.Dense(128)(xFirst)
# dot product attention
attScores = L.Dot(axes=[1, 2])([query, x])
attScores = L.Softmax(name='attSoftmax')(attScores) # [b_s, seq_len]
# rescale sequence
attVector = L.Dot(axes=[1, 1])([attScores, x]) # [b_s, vec_dim]
x = L.Dense(64, activation='relu')(attVector)
x = L.Dense(32)(x)
output = L.Dense(nCategories, activation='softmax', name='output')(x)
model = Model(inputs=[inputs], outputs=[output])
return model
def VggishModel(nCategories = 36,iLen= 16000, sr= 16000, audioset = False):
inputs = L.Input((iLen,), name='input')
x = L.Reshape((1, iLen))(inputs)
if audioset == True:
para_m = [128, 3800, 29]
else: # Google pre-trained
para_m = [32, 7500, 31]
m = Melspectrogram(n_dft=1024, n_hop=para_m[0], input_shape=(1, iLen),
padding='same', sr=sr, n_mels=64,
fmin=125.0, fmax=para_m[1], power_melgram=1.0,
return_decibel_melgram=True, trainable_fb=False,
trainable_kernel=False,
name='mel_stft')
m.trainable = False
mfs = m(x)
mfs = L.Permute((2,1,3))(mfs)
mfs = L.Cropping2D(cropping=((0, para_m[2]), (0, 0)))(mfs)
if audioset == True:
vgg_model = VGGish(include_top=True, load_weights=True)
else:
vgg_model = VGGish(include_top=False, load_weights=False)
vgg_emd = vgg_model(mfs)
if audioset == True:
output = vgg_emd
else:
output = L.Dense(nCategories, activation='softmax', name='output')(vgg_emd)
model = Model(inputs=[inputs], outputs=[output], name='VGG_pre')
# model.summary()
return model
## Updated from 2021 March with TF Hub
import y_params as yamnet_params
import yamnet as yamnet_model
def Yamnet():
params = yamnet_params.Params(sample_rate=16000, patch_hop_seconds=0.1)
# Set up the YAMNet model.
class_names = yamnet_model.class_names('weight/yamnet_class_map.csv')
yamnet = yamnet_model.yamnet_frames_model(params)
yamnet.load_weights('weight/yamnet.h5')
return yamnet