Mercurial > hg > chourdakisreiss2016

annotate experiment-reverb/code/segmentation2.py @ 2:c87a9505f294 tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Added LICENSE for code, removed .wav files
author Emmanouil Theofanis Chourdakis <e.t.chourdakis@qmul.ac.uk>
date Sat, 30 Sep 2017 13:25:50 +0100
parents 246d5546657c
children
rev   line source
e@0 1 # -*- coding: utf-8 -*-
e@0 2 """
e@0 3 Created on Mon Jun 1 11:42:06 2015
e@0 4
e@0 5 @author: Emmanouil Theofanis Chourdakis
e@0 6 """
e@0 7
e@0 8 # Note, reference everything!
e@0 9
e@0 10 from sys import argv
e@0 11
e@0 12
e@0 13 if __name__=="__main__":
e@0 14 if len(argv) != 3:
e@0 15 print("Incorrect number of arguments:")
e@0 16 print("Usage: ")
e@0 17 print("%s <input> <output>")
e@0 18 print("")
e@0 19 print("Arguments:")
e@0 20 print("<input>\tThe input filename. Can be .wav, .mp3, etc...")
e@0 21 print("<parameters>\t The parameters filename, in .yaml format" )
e@0 22 print("<output>\tThe output filename in .yaml format")
e@0 23 sys.exit(-1)
e@0 24 else:
e@0 25
e@0 26 print("[II] Loading libraries")
e@0 27
e@0 28 import essentia
e@0 29 from essentia import Pool
e@0 30 from essentia.standard import *
e@0 31 import yaml
e@0 32
e@0 33
e@0 34 # reqyures matplotlib
e@0 35 from pylab import *
e@0 36
e@0 37 #requires numpy
e@0 38 from numpy import *
e@0 39
e@0 40 #requires scikit-learn
e@0 41 from sklearn.metrics import pairwise_distances
e@0 42
e@0 43 d = {}
e@0 44 v = {}
e@0 45
e@0 46 fname = argv[1]
e@0 47 outfname = argv[2]
e@0 48
e@0 49
e@0 50
e@0 51 trackname = fname.split('.')[0].split('/')[-1]
e@0 52
e@0 53
e@0 54 if outfname.partition('.')[-1].lower() not in ['json', 'yaml']:
e@0 55 print("Please choose a .json or .yaml as an output file.")
e@0 56 sys.exit(-1)
e@0 57 else:
e@0 58 if outfname.partition('.')[-1].lower() == 'json':
e@0 59 output = YamlOutput(filename = outfname, format='json')
e@0 60 else:
e@0 61 output = YamlOutput(filename = outfname, format='yaml')
e@0 62
e@0 63 print("Feature extraction of `%s\'" % fname)
e@0 64
e@0 65 # Sampling Rate
e@0 66 SR = 16000.0
e@0 67
e@0 68 # Sampling Frequency
e@0 69 T = 1.0/SR
e@0 70
e@0 71 # FrameSize
e@0 72 tframeSize = 23 #ms
e@0 73 frameSize = int(ceil(tframeSize*SR/1000)) if mod(ceil(tframeSize*SR/1000),2) == 0 \
e@0 74 else int(floor(tframeSize*SR/1000))
e@0 75
e@0 76 # HopSize
e@0 77 hopSize = frameSize/2
e@0 78
e@0 79 # Load Audio
e@0 80 audio = MonoLoader(filename = fname, sampleRate=SR)()
e@0 81
e@0 82
e@0 83 #Window Frames
e@0 84 w = Windowing(size = frameSize, type = 'hamming')
e@0 85
e@0 86 # Spectrum
e@0 87 spec = Spectrum(size=1024)
e@0 88
e@0 89 # Pool to append mean and variance
e@0 90 pool = Pool()
e@0 91 globalPool = Pool()
e@0 92
e@0 93 # Below are Features to be used in the feature extraction stage
e@0 94 # We use, Spectral Contrast, MFCCs, Zero-Crossing rate, RMS,
e@0 95 # Crest Factor, Spectral Centroid, Spectral Occupation, Spectral Flux
e@0 96
e@0 97 # Spectral Contrast
e@0 98 sc = SpectralContrast(frameSize = frameSize, highFrequencyBound = 8000, sampleRate = SR)
e@0 99
e@0 100 # MFCCs
e@0 101 mfccs = MFCC(highFrequencyBound = 8000, sampleRate = SR)
e@0 102
e@0 103 # Spectral Centroid
e@0 104 centroid = Centroid(range = SR/2)
e@0 105
e@0 106 # Spectral Roll-Off
e@0 107 rolloff = RollOff(sampleRate = SR, cutoff = 0.9)
e@0 108
e@0 109 # Spectral Flux
e@0 110 flux = Flux()
e@0 111
e@0 112 # Zero Crossing Rate
e@0 113 zcr = ZeroCrossingRate()
e@0 114
e@0 115 # RMS
e@0 116 rms = RMS()
e@0 117
e@0 118 # Crest Factor
e@0 119 crest = Crest()
e@0 120
e@0 121
e@0 122
e@0 123
e@0 124 # Segmentation based on Onset detection-based temporal modeling
e@0 125 print("[II] Calculating features for %s, please wait..." % fname)
e@0 126 # Onset Detection
e@0 127
e@0 128 print("[II] Splitting to onsets...")
e@0 129
e@0 130 onsetdetection = OnsetDetectionGlobal(frameSize = frameSize, hopSize = hopSize, sampleRate = SR)(audio)
e@0 131 onsets = Onsets()(essentia.array([onsetdetection]), [1])
e@0 132
e@0 133
e@0 134
e@0 135 print("[II] done, extracting features...")
e@0 136 for o in range(0, len(onsets)-1):
e@0 137 IOI = audio[onsets[o]*SR:onsets[o+1]*SR]
e@0 138
e@0 139
e@0 140
e@0 141
e@0 142 if len(IOI) == 0:
e@0 143 break;
e@0 144
e@0 145
e@0 146 for frame in FrameGenerator(IOI, frameSize, hopSize):
e@0 147 # Temporal Features
e@0 148
e@0 149 zerocrossingrate = zcr(frame)
e@0 150 rmsvalues = rms(frame)
e@0 151
e@0 152 # Spectral features
e@0 153 framespectrum = spec(w(frame))
e@0 154 framecontrast = sc(framespectrum)
e@0 155 mfcc_coeffs = mfccs(framespectrum)[1]
e@0 156 spectralcentroid = centroid(framespectrum)
e@0 157 spectralrolloff = rolloff(framespectrum)
e@0 158 spectralflux = rolloff(framespectrum)
e@0 159
e@0 160
e@0 161
e@0 162 pool.add('lowlevel.zcr', zerocrossingrate)
e@0 163 pool.add('lowlevel.rms', rmsvalues)
e@0 164 pool.add('lowlevel.spectrum.centroid', spectralcentroid)
e@0 165 pool.add('lowlevel.spectrum.rolloff', spectralrolloff)
e@0 166 pool.add('lowlevel.mfcc.coeffs', mfcc_coeffs)
e@0 167 pool.add('lowlevel.spectrum.magnitude', framespectrum)
e@0 168 pool.add('lowlevel.contrast.contrast', framecontrast[0])
e@0 169 pool.add('lowlevel.contrast.valleys', framecontrast[1])
e@0 170 pool.add('lowlevel.spectrum.flux', spectralflux)
e@0 171
e@0 172
e@0 173
e@0 174 spectrumfull = pool['lowlevel.spectrum.magnitude']
e@0 175 spectralcontrast = pool['lowlevel.contrast.contrast']
e@0 176 spectralvalleys = pool['lowlevel.contrast.valleys']
e@0 177 spectralcentroidfeature = pool['lowlevel.spectrum.centroid']
e@0 178 spectralrollofffeature = pool['lowlevel.spectrum.rolloff']
e@0 179 spectralfluxfeature = pool['lowlevel.spectrum.flux']
e@0 180
e@0 181 spectralfeature = concatenate((spectralcontrast,spectralvalleys),1)
e@0 182 mfccfeature = pool['lowlevel.mfcc.coeffs']
e@0 183 zcrfeature = pool['lowlevel.zcr']
e@0 184 rmsfeature = pool['lowlevel.rms']
e@0 185 crestfeature = crest(rmsfeature)
e@0 186 1
e@0 187
e@0 188 meanspectralfeature = mean(spectralfeature, 0)
e@0 189 for i in range(0, shape(spectralfeature)[1]):
e@0 190 globalPool.add('spectralcontrast_%d' % i , meanspectralfeature[i])
e@0 191 globalPool.add('spectralcentroid', mean(spectralcentroidfeature, 0))
e@0 192 globalPool.add('spectralrolloff', mean(spectralrollofffeature, 0))
e@0 193 globalPool.add('spectralflux', mean(spectralfluxfeature, 0))
e@0 194
e@0 195 # Expand mfccs
e@0 196 meanmfcc = mean(mfccfeature, 0)
e@0 197 for i in range(0, shape(mfccfeature)[1]):
e@0 198 globalPool.add('mfcc_%d' % i, meanmfcc[i])
e@0 199
e@0 200
e@0 201 globalPool.add('zcr', mean(zcrfeature, 0))
e@0 202 globalPool.add('rms', mean(rmsfeature, 0))
e@0 203 globalPool.add('crest', crestfeature)
e@0 204
e@0 205 pool.clear()
e@0 206
e@0 207 print("[II] done.")
e@0 208
e@0 209
e@0 210 print("[II] Saving data to %s:" % outfname)
e@0 211 globalPool.add("metadata.filename", fname)
e@0 212 output(globalPool)
e@0 213
e@0 214