-
Notifications
You must be signed in to change notification settings - Fork 3
/
extractor.py
272 lines (200 loc) · 8.22 KB
/
extractor.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
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
import os
import glob
import warnings
import numpy as np
import lws
import subprocess as sp
#TODO: messy file, clean up? lmao no
# set this to music directory
audio_dir = 'fma_medium'
os.environ['AUDIO_DIR'] = os.path.join(os.path.curdir, audio_dir)
# import their custom utils.py - state path to utils.py
import utils
print('Loading tracks.csv and genres.csv...')
tracks = utils.load('tracks.csv')
genres = utils.load('genres.csv')
genres.reset_index(level=0, inplace=True)
# this is the sauce
class Extractor:
"""
Input: Path to .mp3 file (path)
Output: Two npz files:
1) targets: fname_targets.npz
- melspectrogram
- genres
- subgenres
2) raw: fname_raw.npz
- all other features
"""
def __init__(self, target_dir, features, targets):
self.target_dir = target_dir
self.n_fft = 1024
self.hop_sz = 512
self._lws = lws.lws(self.n_fft, self.hop_sz, mode='music')
if not targets:
self.targets = ['genres', 'mel']
else:
self.targets = targets
available_features = [
'mel',
'subgenres',
'zcr',
'chroma',
'spectral_centroid',
'spectral_bandwidth',
'spectral_contrast',
'spectral_rolloff',
'mfcc']
if not features:
self.features = available_features
else:
self.features = [p for p in features if p in available_features]
def load_audio(self, path):
command = [
'ffmpeg',
'-i', path,
'-f', 'f32le', # float32
'-acodec', 'pcm_f32le', # float32
'-ac', '1', # mono
'-ar', '22050', # 22050 sampling rate
'-' # send output to stdout
]
try:
proc = sp.run(command, stdout=sp.PIPE, bufsize=10 ** 7, stderr=sp.DEVNULL, check=True)
return np.frombuffer(proc.stdout, dtype=np.float32)
except sp.CalledProcessError as e:
print('Error occurred with file', path)
print(e)
def compute_features(self, path, sr=22050):
import librosa
features_dict = dict()
warnings.filterwarnings('error', module='librosa')
x = self.load_audio(path)
S = np.abs(self._lws.stft(x)).astype(np.float32)
mel = librosa.amplitude_to_db(
librosa.feature.melspectrogram(sr=sr, S=S.T, n_fft=self.n_fft, hop_length=self.hop_sz)).astype(
np.float32)
tid = int(os.path.splitext(os.path.basename(path))[0])
mel = mel[:, :1290]
# TODO: reduce mel spectograms float size to float16?
if 'subgenres' in self.features:
genre_ids = tracks['track']['genres_all'].loc[tid]
indices = genres['genre_id'].loc[genres['genre_id'].isin(genre_ids)].index
f = np.array(indices, dtype=np.uint32)
features_dict['subgenres'] = f
if 'mel' in self.features:
# # drop the last 2 samples out of 1292 to be divisible by 43
# frame_size = 43
# num_splits = mel.shape[1] // frame_size
#
# # drop excess samples (2 in case of 43 samples in frame) to be divisble by 43
# mel = mel[:, :frame_size * num_splits]
# frame_shift = frame_size // 2 # shift by half a frame for 50% overlap
# sp1 = np.hsplit(mel, num_splits)
# sp2 = np.hsplit(mel[:, frame_shift:frame_shift + frame_size * (num_splits - 1)], num_splits - 1)
# f = np.array(sp1 + sp2)
#
# assert f.shape == (59, 128, 43)
# features_dict['mel_frames'] = f
f = mel
assert f.shape == (128, 1290)
features_dict['mel'] = f
# TODO: dMFCC?
if 'chroma' in self.features:
f = librosa.feature.chroma_stft(S=S ** 2, n_chroma=12)
assert f.shape == (12, 513)
features_dict['chroma'] = f.astype(np.float32)
if 'mfcc' in self.features:
# expects log-power mel spectrogram (amplitude-to-db computes log-power)
f = librosa.feature.mfcc(S=mel, n_mfcc=12)
assert f.shape == (12, 1290)
features_dict['mfcc'] = f
if 'spectral_contrast' in self.features:
f = librosa.feature.spectral_contrast(S=np.abs(S))
assert f.shape == (7, 513)
features_dict['spectral_contrast'] = f.astype(np.float32)
if 'spectral_centroid' in self.features:
f = librosa.feature.spectral_centroid(S=np.abs(S))
features_dict['spectral_centroid'] = f.astype(np.float32)
if 'spectral_bandwidth' in self.features:
f = librosa.feature.spectral_bandwidth(S=np.abs(S))
features_dict['spectral_bandwidth'] = f.astype(np.float32)
if 'spectral_rolloff' in self.features:
f = librosa.feature.spectral_rolloff(S=S)
features_dict['spectral_rolloff'] = f.astype(np.float32)
return features_dict
def write_feature_files(self, path):
features = self.compute_features(path)
basepath, _ = os.path.splitext(path)
basename = os.path.basename(basepath)
# for k, v in features_dict.items():
# print(k, v.dtype)
# save all features
np.savez(os.path.join(self.target_dir, basename + '_raw.npz'), **features)
from multiprocessing import cpu_count, Process, Queue
# %% If no arguements are given to the extractor it will:
# 1) calculate all possible features
# 2) write (genres, subgenres, mel_frames) features to target file
# 3) write all remaning features to raw (aka remainder) file
# %% The extractor takes two optional arguements
# Features to calculate and features to target file
# The difference between the two sets is written to the raw (or remainder) file
features = [ # all of these features will be calculated
'mel',
'subgenres',
'chroma',
'spectral_contrast',
'mfcc']
# %% following features will be written to target file
targets = ['mel']
# reminaing features will be written to remainder file
song_paths = [*glob.iglob(os.path.join(audio_dir, '*/*.mp3'), recursive=True)]
target_dir = os.path.join(audio_dir, 'raw')
import shutil
if os.path.exists(target_dir):
shutil.rmtree(target_dir)
os.mkdir(target_dir)
def apply_all():
queue = Queue()
for i, filename in enumerate(song_paths):
queue.put((i, filename))
count = i + 1
def taker():
while not queue.empty():
i, filename = queue.get()
try:
e.write_feature_files(filename)
except BaseException as ex:
if type(ex) is IndexError:
raise ex
print('Error extracting features for', filename, type(ex), ex)
if (i + 1) % 50 == 0:
print('Applied procedure to', i + 1, '/', count, 'files')
ps = []
for _ in range(cpu_count()):
p = Process(target=taker)
p.start()
ps.append(p)
print('Starting...')
import signal
def handler(sig, frame):
for p in ps:
p.terminate()
signal.signal(signal.SIGINT, handler)
for p in ps:
p.join()
if __name__ == '__main__':
multicore = True
e = Extractor(target_dir, features, targets)
if multicore:
apply_all()
else:
for i, filename in enumerate(song_paths):
try:
print('Applying procedure to', filename)
e.write_feature_files(filename)
print('Applied procedure to', filename)
except Exception as ex:
print('Error extracting features for', filename, ex)
if (i + 1) % 50 == 0:
print('Applied procedure to', i + 1, 'files')