-
Notifications
You must be signed in to change notification settings - Fork 0
/
DataLoader.py
228 lines (203 loc) · 9.95 KB
/
DataLoader.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
import os
from tqdm import tqdm
import numpy as np
import librosa
from Preprocessor import *
from DataManager import *
class DataLoader():
def __init__(self, sample_rate=32000, dim_to_conform=3000):
self.class_id_mapping = {'Hi-hat': 0, 'Saxophone': 1, 'Trumpet': 2, 'Glockenspiel': 3, 'Cello': 4, 'Knock': 5, 'Gunshot_or_gunfire': 6, 'Clarinet': 7, 'Computer_keyboard': 8, 'Keys_jangling': 9, 'Snare_drum': 10, 'Writing': 11, 'Laughter': 12, 'Tearing': 13, 'Fart': 14, 'Oboe': 15, 'Flute': 16, 'Cough': 17, 'Telephone': 18, 'Bark': 19, 'Chime': 20, 'Bass_drum': 21,
'Bus': 22, 'Squeak': 23, 'Scissors': 24, 'Harmonica': 25, 'Gong': 26, 'Microwave_oven': 27, 'Burping_or_eructation': 28, 'Double_bass': 29, 'Shatter': 30, 'Fireworks': 31, 'Tambourine': 32, 'Cowbell': 33, 'Electric_piano': 34, 'Meow': 35, 'Drawer_open_or_close': 36, 'Applause': 37, 'Acoustic_guitar': 38, 'Violin_or_fiddle': 39, 'Finger_snapping': 40}
self.classes_frequency = {}
self.classes_percent = {}
self.classes_verified = {}
self.files = []
self.labels = []
self.verified = []
self.files_loaded = set()
self.train_csv = False
self.sample_rate = sample_rate
self.preprocessor = Preprocessor(dump=True)
self.dim_to_conform = dim_to_conform
def load_files_labels(self, file_csv):
with open(file_csv, 'r') as fp:
file_list = fp.read()
fp.close()
print("load labels and files name from csv...")
file_list = file_list.split("\n")
if(file_list[0].split(",")[2].strip() == "manually_verified"):
self.train_csv = True
for line in file_list[1:]:
split_line = line.split(",")
if(split_line[0] == ''):
continue
file_name = split_line[0].strip()
file_label = split_line[1].strip()
if(file_label == "None"):
continue
if(self.train_csv):
file_verified = np.bool(split_line[2].strip() == '1')
else:
file_verified = np.bool(True)
self.verified.append(file_verified)
self.files.append(file_name)
self.labels.append(self.class_id_mapping[file_label])
self.classes_frequency[
file_label] = self.classes_frequency.setdefault(file_label, 0) + 1
if(file_verified):
self.classes_verified[
file_label] = self.classes_verified.setdefault(file_label, 0) + 1
print("finish loading csv")
tot = len(self.labels)
for key in self.classes_frequency.keys():
self.classes_percent[key] = self.classes_frequency[key] / tot
return self.files, self.labels
def __load_spectrogram(self, file_name, version):
if(version == 1):
data_path = "./dataset/spec/ver1/"
elif(version == 2):
data_path = "./dataset/spec/ver2/"
self.preprocessor.set_spectrogram_path(data_path)
self.preprocessor.set_version(version)
spec_file_name = file_name.replace(".wav", ".npy")
spec_file_name = os.path.join(data_path, spec_file_name)
try:
spec = np.load(spec_file_name)
except FileNotFoundError:
print(
file_name, " spectrogram not exist, compute spectrogram from the original file")
signal = self.__load_audio_signal(file_name)
spec = self.preprocessor.compute_spectrogram(
signal, os.path.basename(spec_file_name))
return spec
def __load_audio_signal(self, file_name):
if(self.train_csv):
audio_path = "./dataset/audio_train/"
else:
audio_path = "./dataset/audio_test/"
audio_file_name = os.path.join(audio_path, file_name)
signal, sample_rate = librosa.load(
audio_file_name, sr=self.sample_rate, mono=True)
signal = self.preprocessor.normalize_and_trim_silence(signal)
return signal
def load_verified_spectrograms(self, version):
if(len(self.files) == 0):
raise NameError("load the file name list before from csv file")
labels = []
spectrograms = []
print("loading verified spectrograms...")
for index in tqdm(range(len(self.files))):
if(not self.verified[index]):
continue
file_name = self.files[index]
spec = self.__load_spectrogram(file_name, version)
spec = DataManager.conform_dim(spec, self.dim_to_conform, 1)
spectrograms.append(spec)
labels.append(self.labels[index])
self.files_loaded.add(file_name)
return np.asarray(spectrograms, dtype=np.float32), np.asarray(labels, dtype=np.int32)
def load_verified_audio_signal(self):
if(len(self.files) == 0):
raise NameError("load the file name list before from csv file")
labels = []
audio_signals = []
print("loading audio signals...")
for index in tqdm(range(len(self.files))):
if(not self.verified[index]):
continue
file_name = self.files[index]
signal = self.__load_audio_signal(file_name)
signal = DataManager.conform_dim(signal, self.dim_to_conform, 0)
audio_signals.append(signal)
labels.append(self.labels[index])
self.files_loaded.add(file_name)
return np.asarray(audio_signals, dtype=np.float32), np.array(labels, dtype=np.int32)
def __get_file_name_of(self, clas):
files_of = set()
for index in range(len(self.files)):
if(self.labels[index] == clas and self.files[index] in self.files_loaded):
files_of.add(self.files[index])
return files_of
def get_next_spectrograms(self, version, how_many):
file_per_class = {}
loaded = 0
labels = []
verified = []
spectrograms = []
classes_percent = {}
for key in self.classes_percent.keys():
classes_percent[self.class_id_mapping[
key]] = self.classes_percent[key]
print("loading the next " + str(how_many) + " files...")
while(loaded < how_many):
inex = 0
for index in tqdm(range(len(self.files))):
if(self.files[index] in self.files_loaded):
continue
if(file_per_class.setdefault(self.labels[index], 0) > classes_percent[self.labels[index]] * how_many):
continue
if(loaded >= how_many):
loaded += how_many
break
spec = self.__load_spectrogram(self.files[index], version)
spec = DataManager.conform_dim(spec, self.dim_to_conform, 1)
spectrograms.append(spec)
labels.append(self.labels[index])
verified.append(self.verified[index])
self.files_loaded.add(self.files[index])
loaded += 1
file_per_class[self.labels[index]] = file_per_class.setdefault(self.labels[
index], 0) + 1
for key in file_per_class.keys():
if(file_per_class[key] <= (classes_percent[key] * how_many)):
files_to_free = self.__get_file_name_of(key)
self.files_loaded = self.files_loaded.difference(files_to_free)
print(len(self.files_loaded))
print(len(spectrograms))
return np.asarray(spectrograms, dtype=np.float32), np.asarray(labels, dtype=np.int32), verified
def get_next_audio_signals(self, how_many):
file_per_class = {}
loaded = 0
labels = []
audio_signals = []
verified = []
for key in self.classes_percent.keys():
classes_percent[self.class_id_mapping[
key]] = self.classes_percent[key]
print("loading the next " + str(how_many) + " files...")
while(loaded < how_many):
for index in tqdm(range(len(self.files))):
if(self.files[index] in self.files_loaded):
continue
if(file_per_class.setdefault(self.labels[index], 0) > classes_percent[self.labels[index]] * how_many):
continue
if(loaded >= how_many):
break
signal = self.__load_audio_signal(self.files[index], version)
signal = DataManager.conform_dim(
signal, self.dim_to_conform, 1)
audio_signals.append(signal)
labels.append(self.labels[index])
verified.append(self.verified[index])
self.files_loaded.add(self.files[index])
loaded += 1
file_per_class[self.labels[index]] = file_per_class.setdefault(self.labels[
index], 0) + 1
for key in file_per_class.keys():
if(file_per_class[key] < classes_percent[self.labels[index]] * how_many):
files_to_free = self.__get_file_name_of(key)
self.files_loaded = self.files_loaded.difference(
files_to_free)
return np.asarray(audio_signals, dtype=np.float32), np.asarray(labels, dtype=np.int32), verified
def get_label_id_mapping(self):
return self.class_id_mapping
def get_label(self):
return self.class_id_mapping.keys()
def get_general_statistics(self):
tot = len(self.labels)
classes_percent_verified = {}
for key in self.classes_frequency.keys():
classes_percent_verified[key] = self.classes_verified[
key] / self.classes_frequency[key]
verif_num = self.verified.count(True)
return self.classes_frequency, self.classes_percent, verif_num, verif_num / tot, self.classes_verified, classes_percent_verified