This repository was archived by the owner on Dec 27, 2023. It is now read-only.
Repository navigation
Expand file tree
/
Copy pathmodels.py
More file actions
141 lines (102 loc) · 5.45 KB
/
Copy pathmodels.py
File metadata and controls
141 lines (102 loc) · 5.45 KB
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
from keras.models import Sequential
from keras.layers import Dense, Conv2D, Reshape, Input, Concatenate, AveragePooling1D, GlobalAveragePooling2D, GlobalMaxPooling2D, \
BatchNormalization, MaxPooling2D, Activation, Dropout, CuDNNGRU, ELU
from keras.optimizers import RMSprop
from keras.models import Model
def conv_gru_model(n_classes):
feat_dict = {'mel_input':Input(shape=(128, 1292, 1)), \
'chroma_input':Input(shape=(12, 1292, 1)), \
'mfcc_input':Input(shape=(20, 1292, 1)), \
'spec_input':Input(shape=(7, 1292, 1)), \
'tonnetz_input':Input(shape=(6, 1292, 1)),\
'mel':None, \
'chroma':None, \
'mfcc':None, \
'spec':None, \
'tonnetz':None}
sub_models = ['mel', 'chroma', 'mfcc', 'spec', 'tonnetz']
#sub_models = ['mel']
for sub_model in sub_models:
if sub_model is 'mel': pool_size = [2,(3,2),(3,2),(4,3)]
elif sub_model is 'chroma': pool_size = [(2,2),(2,2),(2,2),(1,3)]
elif sub_model is 'mfcc': pool_size = [(2,2),(2,2),(2,2),(2,3)]
else: pool_size = [(2,2),(2,2),(1,2),(1,3)]
#CONV BLOCK 1
feat_dict[sub_model] = Conv2D(64, 3, padding='same', name=sub_model+'_conv1')(feat_dict[sub_model+'_input'])
feat_dict[sub_model] = BatchNormalization(name=sub_model+'_bn1')(feat_dict[sub_model])
feat_dict[sub_model] = ELU(name=sub_model+'_activation1')(feat_dict[sub_model])
feat_dict[sub_model] = MaxPooling2D(pool_size=pool_size[0],name=sub_model+'_maxpool1')(feat_dict[sub_model])
feat_dict[sub_model] = Dropout(0.25, name=sub_model+'_dropout1')(feat_dict[sub_model])
#CONV BLOCK 2
feat_dict[sub_model] = Conv2D(128, 3, padding='same', name=sub_model+'_conv2')(feat_dict[sub_model])
feat_dict[sub_model] = BatchNormalization(name=sub_model+'_bn2')(feat_dict[sub_model])
feat_dict[sub_model] = ELU(name=sub_model+'_activation2')(feat_dict[sub_model])
feat_dict[sub_model] = MaxPooling2D(pool_size=pool_size[1],name=sub_model+'_maxpool2')(feat_dict[sub_model])
feat_dict[sub_model] = Dropout(0.25, name=sub_model+'_dropout2')(feat_dict[sub_model])
#CONV BLOCK 3
feat_dict[sub_model] = Conv2D(128, 3, padding='same', name=sub_model+'_conv3')(feat_dict[sub_model])
feat_dict[sub_model] = BatchNormalization(name=sub_model+'_bn3')(feat_dict[sub_model])
feat_dict[sub_model] = ELU(name=sub_model+'_activation3')(feat_dict[sub_model])
feat_dict[sub_model] = MaxPooling2D(pool_size=pool_size[2],name=sub_model+'_maxpool3')(feat_dict[sub_model])
feat_dict[sub_model] = Dropout(0.25, name=sub_model+'_dropout3')(feat_dict[sub_model])
#CONV BLOCK 4
feat_dict[sub_model] = Conv2D(256, 3, padding='same', name=sub_model+'_conv4')(feat_dict[sub_model])
feat_dict[sub_model] = BatchNormalization(name=sub_model+'_bn4')(feat_dict[sub_model])
feat_dict[sub_model] = ELU(name=sub_model+'_activation4')(feat_dict[sub_model])
feat_dict[sub_model] = MaxPooling2D(pool_size=pool_size[3],name=sub_model+'_maxpool4')(feat_dict[sub_model])
feat_dict[sub_model] = Dropout(0.25, name=sub_model+'_dropout4')(feat_dict[sub_model])
feat_dict[sub_model] = Reshape((53, 256))(feat_dict[sub_model])
#GRU BLOCK
feat_dict[sub_model] = CuDNNGRU(128, return_sequences=True, name=sub_model+'_gru1')(feat_dict[sub_model])
feat_dict[sub_model] = CuDNNGRU(128, return_sequences=False,name=sub_model+'_gru2')(feat_dict[sub_model])
x = Concatenate()([feat_dict[i] for i in sub_models])
x = Dropout(0.3)(x)
x = Dense(256, activation='elu')(x)
x = Dropout(0.5)(x)
pred = Dense(n_classes, activation='softmax')(x)
#model = Model(inputs=feat_dict['mel_input'], outputs=feat_dict['mel'])
model = Model(inputs=[feat_dict[i+'_input'] for i in sub_models], outputs=pred)
model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
return model
'''
def conv_lstm_model():
sub_model_dict = {'mel_input':Input(shape=(1292, 128)), \
'chroma_input':Input(shape=(1292, 12)), \
'mfcc_input':Input(shape=(1292, 20)), \
'spec_input':Input(shape=(1292, 7)), \
'tonnetz_input':Input(shape=(1292, 6)),\
'mel':None, \
'chroma':None, \
'mfcc':None, \
'spec':None, \
'tonnetz':None}
for feat in ['mel', 'chroma', 'mfcc', 'spec', 'tonnetz']:
sub_model_dict[feat] = Conv1D(filters=128, kernel_size=4)(sub_model_dict[feat+'_input'])
sub_model_dict[feat] = BatchNormalization()(sub_model_dict[feat])
sub_model_dict[feat] = Activation('elu')(sub_model_dict[feat])
sub_model_dict[feat] = MaxPooling1D()(sub_model_dict[feat])
sub_model_dict[feat] = Dropout(0.5)(sub_model_dict[feat])
sub_model_dict[feat] = Conv1D(filters=256, kernel_size=4)(sub_model_dict[feat])
sub_model_dict[feat] = BatchNormalization()(sub_model_dict[feat])
sub_model_dict[feat] = Activation('elu')(sub_model_dict[feat])
sub_model_dict[feat] = MaxPooling1D()(sub_model_dict[feat])
sub_model_dict[feat] = Dropout(0.5)(sub_model_dict[feat])
#sub_model_dict[feat] = CuDNNLSTM(256, return_sequences=True)(sub_model_dict[feat])
sub_model_dict[feat] = Reshape((320, 256, 1))(sub_model_dict[feat])
x = Concatenate(axis=-1)([sub_model_dict['mel'],\
sub_model_dict['chroma'],\
sub_model_dict['mfcc'],\
sub_model_dict['spec'],\
sub_model_dict['tonnetz']])
x = mobnet_remix()(x)
#x = Dense(512, activation='elu')(x)
pred = Dense(8, activation='softmax')(x)
model = Model(inputs=[sub_model_dict['mel_input'], \
sub_model_dict['chroma_input'], \
sub_model_dict['mfcc_input'], \
sub_model_dict['spec_input'], \
sub_model_dict['tonnetz_input']], \
outputs=pred)
model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'] )
return model
'''