-
Notifications
You must be signed in to change notification settings - Fork 0
/
AudioNetwork.py
241 lines (189 loc) · 8.53 KB
/
AudioNetwork.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
from serpent.machine_learning.context_classification.context_classifier import ContextClassifier
from serpent.utilities import SerpentError
import tensorflow as tf
try:
from keras.applications.inception_v3 import InceptionV3, preprocess_input
from keras.layers import (Activation, Convolution1D, Dense, Dropout, GlobalAveragePooling1D,
GlobalMaxPool1D, Input, MaxPool1D, concatenate)
from keras.models import Model, load_model
from keras.callbacks import ModelCheckpoint
from keras.utils import Sequence, to_categorical
from keras import backend as K
except ImportError:
raise SerpentError("Setup has not been been performed for the ML module. Please run 'serpent setup ml'")
import skimage.transform
import serpent.cv
import numpy as np
import random
import os
import shutil
import IPython
import pandas as pd
# To load and read audio files
import librosa
SAMPLE_RATE = 44100
from Config import Config, DataGenerator
config = Config(sampling_rate=SAMPLE_RATE, audio_duration=2, use_mfcc=False)
def auc_roc(y_true, y_pred):
# any tensorflow metric
value, update_op = tf.contrib.metrics.streaming_auc(y_pred, y_true)
# find all variables created for this metric
metric_vars = [i for i in tf.local_variables() if 'auc_roc' in i.name.split('/')[1]]
# Add metric variables to GLOBAL_VARIABLES collection.
# They will be initialized for new session.
for v in metric_vars:
tf.add_to_collection(tf.GraphKeys.GLOBAL_VARIABLES, v)
# force to update metric values
with tf.control_dependencies([update_op]):
value = tf.identity(value)
return value
class ContextClassifierError(BaseException):
pass
class AudioNetwork(ContextClassifier):
def __init__(self, input_shape=None):
super().__init__()
self.input_shape = input_shape
self.training_generator = None
self.validation_generator = None
def train(self, epochs=3, autosave=False, validate=True):
if validate and (self.training_generator is None or self.validation_generator is None):
self.prepare_generators()
inp = Input(shape=self.input_shape)
x = Convolution1D(16, 9, activation='tanh', padding="valid")(inp)
x = Convolution1D(16, 9, activation='tanh', padding="valid")(x)
x = MaxPool1D(16)(x)
x = Dropout(rate=0.1)(x)
x = Convolution1D(32, 3, activation='tanh', padding="valid")(x)
x = Convolution1D(32, 3, activation='tanh', padding="valid")(x)
x = MaxPool1D(4)(x)
x = Dropout(rate=0.1)(x)
x = Convolution1D(32, 3, activation='tanh', padding="valid")(x)
x = Convolution1D(32, 3, activation='tanh', padding="valid")(x)
x = MaxPool1D(4)(x)
x = Dropout(rate=0.1)(x)
x = Convolution1D(256, 3, activation='tanh', padding="valid")(x)
x = Convolution1D(256, 3, activation='tanh', padding="valid")(x)
x = GlobalMaxPool1D()(x)
x = Dropout(rate=0.2)(x)
x = Dense(64, activation='tanh')(x)
x = Dense(1028, activation='tanh')(x)
predictions = Dense(2, activation='softmax')(x)
self.classifier = Model(inputs=inp, outputs=predictions)
self.classifier.compile(
optimizer="adam",
loss="categorical_crossentropy",
metrics=["accuracy",auc_roc]
)
callbacks = []
if autosave:
callbacks.append(ModelCheckpoint(
"datasets/audio_classifier_{epoch:02d}-{val_loss:.2f}.model",
monitor='val_loss',
verbose=0,
save_best_only=False,
save_weights_only=False,
mode='auto',
period=1
))
self.classifier.fit_generator(
self.training_generator,
samples_per_epoch=self.training_sample_count,
nb_epoch=epochs,
validation_data=self.validation_generator,
nb_val_samples=self.validation_sample_count,
class_weight={0: 4., 1: 1.},
callbacks=callbacks
)
def validate(self):
pass
def predict(self, input_frame):
def audio_norm(data):
np.nan_to_num(data, copy=False)
max_data = np.max(data)
min_data = np.min(data)
data = (data - min_data) / (max_data - min_data + 1e-6)
return data - 0.5
source_min = 0
input_frame = np.array(serpent.cv.normalize(
input_frame,
source_min,
source_max=1,
target_min=-1,
target_max=1
), dtype="float32")
np.nan_to_num(input_frame,copy=False)
class_probabilities = self.classifier.predict(input_frame[None, :, :])[0]
print(class_probabilities)
max_probability_index = np.argmax(class_probabilities)
max_probability = class_probabilities[1]
return max_probability
def save_classifier(self, file_path):
if self.classifier is not None:
self.classifier.save(file_path)
def load_classifier(self, file_path):
self.classifier = load_model(file_path, custom_objects={'auc_roc': auc_roc})
def prepare_generators(self):
trainingLabels = []
trainingIDs = []
files = os.listdir('datasets/current/training/yes_jump/')
for file in files:
if file.endswith(".wav"):
trainingLabels.append('yes_jump')
trainingIDs.append('/yes_jump/' +file)
files = os.listdir('datasets/current/training/no_jump/')
for file in files:
if file.endswith(".wav"):
trainingLabels.append('no_jump')
trainingIDs.append('/no_jump/' +file)
validationLabels = []
ValidtionIDS = []
files = os.listdir('datasets/current/validation/yes_jump/')
for file in files:
if file.endswith(".wav"):
validationLabels.append('yes_jump')
ValidtionIDS.append('/yes_jump/' +file)
files = os.listdir('datasets/current/validation/no_jump/')
for file in files:
if file.endswith(".wav"):
validationLabels.append('no_jump')
ValidtionIDS.append('/no_jump/' + file)
print(trainingIDs);
def audio_norm(data):
np.nan_to_num(data, copy=False)
max_data = np.max(data)
min_data = np.min(data)
data = (data-min_data)/(max_data-min_data+1e-6)
return data-0.5
self.training_generator = DataGenerator(config, 'datasets/current/training', trainingIDs,
trainingLabels, batch_size=32, preprocessing_fn=audio_norm)
self.validation_generator = DataGenerator(config, 'datasets/current/validation', ValidtionIDS,
validationLabels, batch_size=32, preprocessing_fn=audio_norm)
def executable_train(epochs=3, autosave=False, classifier="AudioNetwork", validate=True):
context_paths = list()
for root, directories, files in os.walk("datasets/audio/collect_frames_for_training".replace("/", os.sep)):
if root != "datasets/audio/collect_frames_for_training".replace("/", os.sep):
break
for directory in directories:
context_paths.append(f"datasets/audio/collect_frames_for_training/{directory}".replace("/", os.sep))
if not len(context_paths):
raise ContextClassifierError("No Context Frames found in 'datasets/audio/collect_frames_for_training'...")
serpent.datasets.create_training_and_validation_sets(context_paths)
context_path = random.choice(context_paths)
frame_path = None
for root, directories, files in os.walk(context_path):
for file in files:
if file.endswith(".wav"):
frame_path = f"{context_path}/{file}"
break
if frame_path is not None:
break
frame, _ = librosa.core.load(frame_path, sr=SAMPLE_RATE)
np.nan_to_num(frame,copy=False)
frame.shape
audionetwork = AudioNetwork(input_shape=(config.audio_length, 1))
audionetwork.train(epochs=epochs, autosave=autosave, validate=validate)
audionetwork.validate()
AudioNetwork.save_classifier(audionetwork, "datasets/pretrained_audio_classifier.model")
print("Success! Model was saved to 'datasets/pretrained_audio_classifier.model'")
# check this:
#(X,y) = self.training_generator[0]