-
Notifications
You must be signed in to change notification settings - Fork 1
/
main.py
415 lines (314 loc) · 15.1 KB
/
main.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
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
'''
@author: Christopher Jacoby, Tlacael Esparza
MIR Fall 2012: Prof. Juan Bello
Final Project - Feature Extraction & Clustering of Noise Data
'''
import argparse
import numpy as np
import marlib.matlab as M
from datetime import datetime
from mirlib import mir_utils
from mirlib import featurevector
import mirlib.audiofile as af
import mirlib.FFTParams as fftparams
import mirlib.feature_extraction.eventDetect as ed
import mirlib.feature_extraction.calcLoudness as cl
import mirlib.feature_extraction.lowlevel_spectral as llspect
from mirlib.feature_analysis import kmeans
import plot
from matplotlib import pyplot as plt
FEATURE_VECTOR_FILENAME = "features.npy"
SONE_VECTOR_FILENAME = "sones.npy"
def getfeatures(args):
''' write the extracted features from the input audio file into numpy files for reading in the other steps. '''
debug = args.debug
filepath = args.audiofile
chunk_len = args.audio_seg_length
afm = af.audiofile_manager(filepath, chunk_len)
# FFT Parameters
fs = afm.afReader.samplerate()
N = 2048
hopDenom = 2
hopSize = N/hopDenom
zp = 0
winfunc=np.hamming
fftParams = fftparams.FFTParams(fs, N, hopDenom, zp, winfunc)
# MFCC Paramters
nFilters = 40
nDCTCoefs = 20
minFreq = 50
maxFreq = 8000
nIndexSkip = 2
seglen = 1
mfccParams = fftparams.MFCCParams(nFilters, nDCTCoefs, minFreq, maxFreq, nIndexSkip)
# Feature Vector parameters
# Template : ('name', order index, length)
vector_template = [('sones', 0, 1),
('mfcc', 1, nDCTCoefs - nIndexSkip)]
sone_template = [('sones', 0, 1)]
# Initialize the feature vector holder
feature_holder = featurevector.feature_holder(vector_template, filepath)
sone_holder = featurevector.feature_holder(sone_template, filepath)
envelopeHolder = []
audioHolder = []
sonesHolder = []
maxEnvelope = 0;
count =0
print "Feature Extraction Mode\n"
print datetime.now()
# For each chunk of audio
while afm.HasMoreData():
count +=1
audioChunk, chunkIndex = afm.GetNextSegment()
if debug: print "Read %d sample chunk of audio (%0.2fs)" % (len(audioChunk), len(audioChunk) / fs)
# Get Events
eventTimes, envelope = GetEvents(audioChunk, fftParams, debug)
if maxEnvelope < envelope.max():
maxEnvelope = envelope.max()
if debug: print "EVENTTIMES:", eventTimes
envelopeHolder.append(envelope)
eventTimesSamps = np.asarray(np.multiply(eventTimes,fs),dtype=int)
# Get event audio segments
eventSegments = GetEventAudioSegments(eventTimesSamps, audioChunk, debug)
#get sones
eventSegmentSones = GetEventSones(eventSegments, fftParams, debug)
# Get the MFCCs for each segment / event
eventSegmentMFCCs = GetEventMFCCs(eventSegments, fftParams, mfccParams, debug)
# Time-average for each segment / event
averagedEventSegmentMFCCs, averagedEventSegmentSones = AverageEventFeatures(eventSegmentMFCCs, eventSegmentSones, seglen, fftParams, debug)
# Store these vectors in the feature_holder, labelled with their time
StoreFeatureVector(feature_holder, sone_holder, averagedEventSegmentMFCCs, averagedEventSegmentSones, chunkIndex, chunk_len, eventTimes, debug)
# Write features to disk
print datetime.now()
fileSize = feature_holder.save(FEATURE_VECTOR_FILENAME)
print "Wrote", fileSize, "bytes to disk. (%s)" % (FEATURE_VECTOR_FILENAME)
fileSize = sone_holder.save(SONE_VECTOR_FILENAME)
print "Wrote", fileSize, "bytes to disk. (%s)" % (SONE_VECTOR_FILENAME)
def GetEvents(audiodata, fftParams, debug):
''' Given the audio data, return lists of form [ (start time, length), ...]
'''
# Get Onsets
onsetDetector = ed.onsetDetect(fftParams)
# Get Time-Segments from those offsets
return (onsetDetector.findEventLocations(audiodata))
def GetEventAudioSegments(eventTimes, audiodata, debug):
''' Given event times, return the audio segments associated to those events.
eventTimes must be in samples!!! '''
segments = []
for i in np.arange(len(eventTimes)):
segments.append(audiodata[eventTimes[i,0]:eventTimes[i,1]])
if debug:
print "\tEvent Detected. Start: %0.2fs, End: %0.2fs, Length: %d samps" % (eventTimes[i,0], eventTimes[i,1], len(segments[i]))
return segments
def GetEventSones(eventSegments, fftParams, debug):
''' For audio event segments, get the sones for each. '''
soneSegments = []
for i in np.arange(len(eventSegments)):
calcSones = cl.SoneCalculator(eventSegments[i], fftParams)
soneSegments.append(calcSones.calcSoneLoudness())
return soneSegments
def GetEventMFCCs(eventSegments, fftParams, mfccParams, debug):
''' For audio event segments, get the mfccs for each. '''
mfccSegments = []
for i in np.arange(len(eventSegments)):
X = M.spectrogram(eventSegments[i], fftParams.N, fftParams.h, fftParams.winfunc(fftParams.N))
#take log magnitude
mfcc = llspect.MFCC_Normalized(X, mfccParams, fftParams)
mfccSegments.append(mfcc)
if debug:
print "\t MFCC:", mfcc.shape
#print "\t ",mfcc
return mfccSegments
def AverageEventFeatures(mfccSegments, soneSegments, seglen, fftParams, debug):
''' Given the raw feature data, get time averaged versions, averaged to seglen, based on the FS from fftParams. '''
spect_fs = fftParams.fs / fftParams.h
averaged_mfcc_segs = []
averaged_sone_segs = []
for i in np.arange(len(mfccSegments)):
averaged_segment = mir_utils.AverageFeaturesInTime(mfccSegments[i], spect_fs, seglen)
averaged_mfcc_segs.append(averaged_segment)
averaged_segment = mir_utils.AverageFeaturesInTime(soneSegments[i], spect_fs, seglen)
averaged_sone_segs.append(averaged_segment)
return averaged_mfcc_segs, averaged_sone_segs
def StoreFeatureVector(feature_holder, sone_holder, averagedEventSegmentMFCCs, averagedEventSegmentSones, chunkIndex, chunk_len, eventTimes, debug):
''' given the feature vectors, add them to the feature_holders, with dicts to point back to the original audio '''
chunk_start_time = chunkIndex * chunk_len # in seconds
for i in range(len(averagedEventSegmentMFCCs)):
chunk_start = eventTimes[i][0]
chunk_length = eventTimes[i][1] - eventTimes[i][0]
timekey = (chunk_start_time + chunk_start, chunk_length)
thissone = averagedEventSegmentSones[i]
thismfcc = averagedEventSegmentMFCCs[i]
if debug:
print "\t Storing Vector at key:", timekey
sone_holder.add_feature('sones', thissone, timelabel=timekey)
feature_holder.add_feature('mfcc', thismfcc, timelabel=timekey)
def clustering(args):
''' run clustering on a single k'''
print "Feature Analysis/Clustering Mode: single k"
feature_holder = featurevector.feature_holder(filename=FEATURE_VECTOR_FILENAME)
sones_holder = featurevector.feature_holder(filename=SONE_VECTOR_FILENAME)
k = args.k
print feature_holder
mfccs = feature_holder.get_feature('mfcc')
print sones_holder
sones = sones_holder.get_feature('sones')
centroids, distortion = Get_Best_Centroids(k, 1)
print "Distortion for this run: %0.3f" % (distortion)
classes,dist = kmeans.scipy_vq(mfccs, centroids)
# Get the inter class dist matrix
inter_class_dist_matrix = mir_utils.GetSquareDistanceMatrix(centroids)
eventBeginnings = feature_holder.get_event_start_indecies()
# write audio if given -w
if args.plot_segments:
PlotWaveformWClasses(k, feature_holder,classes)
if args.write_audio_results:
WriteAudioFromClasses(k, feature_holder, classes)
plot.plot(mfccs, sones, eventBeginnings, centroids, inter_class_dist_matrix, classes)
def calcJ(mfccs, classes, centroids, k):
''' calculates J_0 from the class labels. '''
Sw = np.zeros((mfccs.shape[1],mfccs.shape[1]))
Sb = np.zeros((mfccs.shape[1],mfccs.shape[1]))
for i in range(k):
#sw
if len(mfccs[(classes == i)]) == 0:
#print i
continue
proportion = np.sum(classes==i)/float(classes.size)
curClass = mfccs[classes==i]
if np.ndim(curClass) ==1:
curClass.shape = (1,curClass.size)
covar = np.outer(curClass,curClass)
Sw += np.multiply(proportion,covar)
elif len(curClass) == 1:
covar = np.outer(curClass,curClass)
Sw += np.multiply(proportion,covar)
else:
covar = np.cov(curClass.T)
Sw += np.multiply(proportion,covar)
#Sb
globalMean = np.mean(mfccs, 0)
meanOfClass = np.mean(mfccs[classes==i],0)
diff = meanOfClass - globalMean
Sb += np.multiply(np.outer(diff,diff), proportion)
SWsumDiag = sum(np.diag(Sw))
SBsumDiag = sum(np.diag(Sb))
return SBsumDiag/SWsumDiag
def feature_selection(args):
''' run clustering on a range of k's'''
print "Feature Analysis/Clustering Mode - feature selection from multiple k's"
feature_holder = featurevector.feature_holder(filename=FEATURE_VECTOR_FILENAME)
kMin = args.k_min
kMax = args.k_max
kHop = args.k_hop
mfccs = feature_holder.get_feature('mfcc')
nmfcc = len(mfccs)
print "N MFCCS:", nmfcc
results = []
for k in range(kMin, kMax, kHop):
print "Running k-Means with k=%d" % (k)
if k >= nmfcc:
print "WARNING! k is greater than the number of samples!"
centroids, distortion = kmeans.scipy_kmeans(mfccs, k)
classes, dist = kmeans.scipy_vq(mfccs, centroids)
J0 = calcJ(mfccs, classes, centroids, k)
results.append( (k, distortion, dist, J0) )
plot.plot_feature_selection(kMin, kMax, kHop, results)
def Get_Best_Centroids(k, iterations):
print "Feature Analysis/Clustering Mode - feature selection from multiple k's"
feature_holder = featurevector.feature_holder(filename=FEATURE_VECTOR_FILENAME)
mfccs = feature_holder.get_feature('mfcc')
j_measures = np.zeros(iterations)
max = 0;
bestCentroids = 0
bestDistortion = 0
for i in range(iterations):
centroids, distortion = kmeans.scipy_kmeans(mfccs, k)
classes, dist = kmeans.scipy_vq(mfccs, centroids)
j_measures[i] = calcJ(mfccs, classes, centroids, k)
if j_measures[i] > max:
max = j_measures[i]
bestCentroids = centroids
bestDistortion = distortion
return bestCentroids, bestDistortion
def PlotWaveformWClasses(k, feature_holder, classes):
index_time_map = feature_holder.get_index_time_map()
print "Original File:", feature_holder.filename
# for each class k
events = np.zeros((60*6*44100,1))
segment_classes = GetClassFromSegment(k, index_time_map, classes)
for i in sorted(segment_classes.keys()):
# Find all time segments that go with this class
timeSegments = [ index_time_map[j] for j in sorted(segment_classes[i])]
# Write all these time segments to a single file
audioSegments, fs = af.get_arbitrary_file_segments(feature_holder.filename, timeSegments)
for j in range(len(timeSegments)):
print j, timeSegments[j]
if (timeSegments[j][0]+timeSegments[j][1]) < 60*6:
startIndex = int(timeSegments[j][0]*fs)
endIndex = int(float(timeSegments[j][0]+timeSegments[j][1])*fs)
diff = len(audioSegments[j])-len(events[startIndex:endIndex])
events[startIndex:endIndex+diff] = audioSegments[j]
plt.plot(events[::100]);plt.show()
def WriteAudioFromClasses(k, feature_holder, classes):
''' given the class labels, write the audio associated with each one '''
index_time_map = feature_holder.get_index_time_map()
print "Original File:", feature_holder.filename
# for each class k
segment_classes = GetClassFromSegment(k, index_time_map, classes)
for i in sorted(segment_classes.keys()):
# Find all time segments that go with this class
timeSegments = [ index_time_map[j] for j in sorted(segment_classes[i])]
#print timeSegments
# Write all these time segments to a single file
audioSegments, fs = af.get_arbitrary_file_segments(feature_holder.filename, timeSegments)
resultDir = './results'
af.write_segment_audio("%s/class-%d.wav" % (resultDir, i), audioSegments, fs)
def GetClassFromSegment(k, index_time_map, classes):
''' Determine which class an audio segment belongs to, given the classes of it's components.
Currently just getting the mode from the histogram. This probably should be better...'''
results = {}
for time_seg in sorted(index_time_map.keys()):
start = time_seg[0]
end = start + time_seg[1]
time_seg_classes = classes[ start : end ]
hist, edges = np.histogram( time_seg_classes, np.arange(k) )
segment_class = np.argmax(hist)
if results.has_key(segment_class):
results[segment_class].append(time_seg)
else:
results[segment_class] = [time_seg]
return results
def ParseArgs():
''' Parse the program arguments & run the appropriate functions '''
parser = argparse.ArgumentParser()
# Main program parameters
parser.add_argument("-d", "--debug", action="store_true")
subparsers = parser.add_subparsers(help="Program Mode Help")
# Parameters for Feature Extraction
parser_getfeatures = subparsers.add_parser("getfeatures", help="Feature Extraction Mode")
parser_getfeatures.set_defaults(func=getfeatures)
parser_getfeatures.add_argument("audiofile", help="Input audio file path")
parser_getfeatures.add_argument("-l", "--audio_seg_length", help="Amount of audio data to process at a time", default=30, type=int)
# Parameters for Feature Selection
parser_featureselection = subparsers.add_parser("featureselect", help="Feature Selection Mode")
parser_featureselection.add_argument("k_min", default=2, type=int)
parser_featureselection.add_argument("k_max", default=100, type=int)
parser_featureselection.add_argument("k_hop", default=5, type=int)
parser_featureselection.set_defaults(func=feature_selection)
# Parameters for Clustering mode
parser_clustering = subparsers.add_parser("clustering", help="Feature Analysis Mode")
parser_clustering.add_argument("k", help="Number of classes", type=int)
parser_clustering.add_argument("-w", "--write_audio_results", help="Write audio from clusters", action="store_true")
parser_clustering.add_argument("-plot", "--plot_segments", help="Plot audio segment classes", action="store_true")
parser_clustering.set_defaults(func=clustering)
args = parser.parse_args()
args.func(args)
def main():
''' handle argument parsing '''
print 'Christopher Jacoby & Tlacael Esparza'
print 'MIR-Noise'
print
ParseArgs()
if __name__ == '__main__':
main()