-
Notifications
You must be signed in to change notification settings - Fork 0
/
features.py
163 lines (128 loc) · 6.71 KB
/
features.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
from preprocessing import *
from argparse import ArgumentParser
import pickle
import time
from numba import jit, cuda
import concurrent.futures
def calculate_note_features(note, sr, n_fft, pitch):
hop_length = int(n_fft / 2)
# zerocrossingrates of all windows of all notes are put together
ZCR = librosa.feature.zero_crossing_rate(note, frame_length=2048, hop_length=512) # 46ms like in paper.
Spectrogramnote = np.abs(librosa.stft(note, n_fft=n_fft, hop_length=hop_length))
centroids = librosa.feature.spectral_centroid(S=Spectrogramnote, sr=sr)
bandwidths = librosa.feature.spectral_bandwidth(S=Spectrogramnote, sr=sr)
harmonicpercentage = np.empty((0))
inharmonicity = np.empty(0)
for frame in FrameGenerator(note, frameSize=n_fft, hopSize=hop_length, startFromZero=True):
# print('frame' + str(idx))
window = Windowing(type='blackmanharris92')(frame)
windowfilt = DCRemoval(sampleRate = sr)(window)
spectrum = Spectrum(size=n_fft)(windowfilt)
specdb = (20 * np.log10(spectrum))/(-60)
frequencies, magnitudes = SpectralPeaks(maxPeaks=10, sampleRate=sr)(
specdb) # should be in dB, and best with blackmanharriswindow with 92db
magnitudes = np.delete(magnitudes, np.where(frequencies == 0))
frequencies = np.delete(frequencies, np.where(frequencies == 0))
harmonicfreq, harmonicmag = HarmonicPeaks(maxHarmonics=4, tolerance=0.3)(frequencies, magnitudes, float(
pitch)) # we feed frequencies, magnitudes and pitch.
percentage_bandwidth = pitch / 12 # in paper, 1/12 octave
for k in range(harmonicfreq.shape[0]):
# print(EnergyBandRatio(sampleRate = sr, startFrequency = harmonicfreq[k] - (percentage_bandwidth/2), stopFrequency = harmonicfreq[k] + (percentage_bandwidth/2))(spectrum))
harmonicpercentage = np.append(harmonicpercentage, EnergyBandRatio(sampleRate=sr,
startFrequency=harmonicfreq[k] - (
percentage_bandwidth / 2),
stopFrequency=harmonicfreq[k] + (
percentage_bandwidth / 2))(
spectrum))
#print(harmonicpercentage)
inharmonicity = np.append(inharmonicity, Inharmonicity()(harmonicfreq,
harmonicmag)) # should we first mean the frequencies and magnitudes, than give this to inharmonicity?
harmonicpercentage = harmonicpercentage.reshape(-1,
4) # gives second dimension, otherwise we do not know which harmonic peak it was.
envelope = Envelope()(note)
logtime, start, stop = LogAttackTime(sampleRate = sr)(envelope)
envflat = FlatnessSFX()(envelope)
tempcentroid = TCToTotal()(envelope)
return ZCR, centroids, bandwidths, inharmonicity, harmonicpercentage, logtime, envflat, tempcentroid
def calculate_track_features(filename, sr, C, n_fft):
st=time.time()
audio = MonoLoader(filename = filename, sampleRate =sr)()
#audio = librosa.core.load(path = filename, sr = sr) #cannot read mp3
end = time.time()
print(str(end-st)+ ' seconds to load')
audio = normalize(audio)
#we get limits and pitches from librosa
start = time.time()
limits, pitchdisc = extractpitchlimitslibrosa(audio,sr,C)
stop = time.time()
print('preprocessing ' + str(stop-start))
#noteharmonicpercentage = np.empty((limits.shape[0], 4, 2))
features = np.empty((limits.shape[0], 19))
for i in range(limits.shape[0]):
#note splitting
note = audio[int(limits[i, 0]*sr): int(limits[i, 1]*sr)]
ZCR, centroid, bandwidth, inharmonicity, harmonicpercentage, logtime, envflat, tempcentroid = calculate_note_features(note, sr, n_fft, pitchdisc[i])
features[i] = np.array([np.mean(ZCR), np.std(ZCR), np.mean(centroid), np.std(centroid),
np.mean(bandwidth), np.std(bandwidth), np.mean(inharmonicity), np.mean(inharmonicity),
np.mean(harmonicpercentage[:,0]), np.std(harmonicpercentage[:,0]),
np.mean(harmonicpercentage[:, 1]), np.std(harmonicpercentage[:, 1]),
np.mean(harmonicpercentage[:, 2]), np.std(harmonicpercentage[:, 2]),
np.mean(harmonicpercentage[:, 3]), np.std(harmonicpercentage[:, 3]),
logtime, envflat, tempcentroid ])
return features
def calculate_tracks_features(songname):
filenames = os.listdir(str(args.indir) + '/' + songname)
song = {}
if not os.path.exists(args.outdir + '/' + songname +'.pkl'):
start = time.time()
print('computing ' + str(args.indir) + '/' + songname)
for filename in filenames:
song[filename] = calculate_track_features(str(args.indir) + '/' + songname + '/' + filename, sr, C, n_fft)
stop = time.time()
print(songname)
print('computed in '+ str(stop-start) + 's')
picklename = args.outdir +'/' +songname + '.pkl'
filehandler = open(picklename, 'wb')
pickle.dump(song, filehandler)
print('file written at ' + str(picklename))
filehandler.close()
if __name__ == '__main__':
argparser = ArgumentParser()
argparser.add_argument('--indir', type=str, default='data',
help='directory where the tracks are')
argparser.add_argument('--outdir', type=str, default='outdicts',
help='directory where the dictionaries are written')
args = argparser.parse_args()
sr =22050
C = 300
n_fft = 1024
if not os.path.exists(args.outdir):
os.makedirs(args.outdir)
songnames = os.listdir(args.indir)
start = time.time()
calculate_tracks_features(songnames[0])
stop = time.time()
print(start-stop)
#instruments = calculate_tracks_features(songnames, sr, C, n_fft)
'''
#WITH THREADS
with concurrent.futures.ThreadPoolExecutor(max_workers=8) as executor:
start1 = time.time()
executor.map(calculate_tracks_features, songnames)
'''
'''
for song in songs:
for inst in song.keys():
instruments[inst] = song[inst]
print(inst + 'written in instruments')
stop1 = time.time()
'''
'''
#WITHOUT THREADS:
start1 = time.time()
for songname in songnames:
song = calculate_tracks_features(songname)
for inst in song.keys():
instruments[inst] = song[inst]
stop1 = time.time()
'''