[13c3fba] | 1 | from aubio.aubioclass import * |
---|
| 2 | from onset import taskonset |
---|
| 3 | |
---|
| 4 | class taskbeat(taskonset): |
---|
| 5 | def __init__(self,input,params=None,output=None): |
---|
| 6 | """ open the input file and initialize arguments |
---|
| 7 | parameters should be set *before* calling this method. |
---|
| 8 | """ |
---|
| 9 | taskonset.__init__(self,input,output=None,params=params) |
---|
| 10 | self.btwinlen = 512**2/self.params.hopsize |
---|
| 11 | self.btstep = self.btwinlen/4 |
---|
| 12 | self.btoutput = fvec(self.btstep,self.channels) |
---|
| 13 | self.dfframe = fvec(self.btwinlen,self.channels) |
---|
| 14 | self.bt = beattracking(self.btwinlen,self.channels) |
---|
| 15 | self.pos2 = 0 |
---|
[9bec8fe] | 16 | self.old = -1000 |
---|
[13c3fba] | 17 | |
---|
| 18 | def __call__(self): |
---|
| 19 | taskonset.__call__(self) |
---|
[ba11e53] | 20 | #results = taskonset.__call__(self) |
---|
[13c3fba] | 21 | # write to current file |
---|
| 22 | if self.pos2 == self.btstep - 1 : |
---|
| 23 | self.bt.do(self.dfframe,self.btoutput) |
---|
| 24 | for i in range (self.btwinlen - self.btstep): |
---|
| 25 | self.dfframe.set(self.dfframe.get(i+self.btstep,0),i,0) |
---|
| 26 | for i in range(self.btwinlen - self.btstep, self.btwinlen): |
---|
| 27 | self.dfframe.set(0,i,0) |
---|
| 28 | self.pos2 = -1; |
---|
| 29 | self.pos2 += 1 |
---|
| 30 | val = self.opick.pp.getval() |
---|
[ba11e53] | 31 | #if not results: val = 0 |
---|
| 32 | #else: val = results[1] |
---|
[13c3fba] | 33 | self.dfframe.set(val,self.btwinlen - self.btstep + self.pos2,0) |
---|
| 34 | i=0 |
---|
| 35 | for i in range(1,int( self.btoutput.get(0,0) ) ): |
---|
| 36 | if self.pos2 == self.btoutput.get(i,0) and \ |
---|
| 37 | aubio_silence_detection(self.myvec(), |
---|
| 38 | self.params.silence)!=1: |
---|
[9bec8fe] | 39 | now = self.frameread-0 |
---|
| 40 | period = (60 * self.params.samplerate) / ((now - self.old) * self.params.hopsize) |
---|
| 41 | self.old = now |
---|
[6b0ca87] | 42 | return now*self.btstep*self.params.step,period |
---|
[13c3fba] | 43 | |
---|
[9bec8fe] | 44 | def eval(self,results,tol=0.20,tolcontext=0.25): |
---|
| 45 | obeats = self.gettruth() |
---|
| 46 | etime = [result[0] for result in results] |
---|
| 47 | otime = [obeat[0] for obeat in obeats] |
---|
| 48 | CML_tot, CML_max, CML_start, CML_end = 0,0,0,0 |
---|
| 49 | AML_tot, AML_max, AML_start, AML_end = 0,0,0,0 |
---|
| 50 | AMLd_tot, AMLd_max, AMLd_start, AMLd_end = 0,0,0,0 |
---|
| 51 | AMLh_tot, AMLh_max, AMLh_start, AMLh_end = 0,0,0,0 |
---|
| 52 | AMLo_tot, AMLo_max, AMLo_start, AMLo_end = 0,0,0,0 |
---|
| 53 | # results iteration |
---|
| 54 | j = 1 |
---|
| 55 | # for each annotation |
---|
| 56 | for i in range(2,len(otime)-2): |
---|
| 57 | if j+1 >= len(etime): break |
---|
| 58 | count = 0 |
---|
| 59 | # look for next matching beat |
---|
| 60 | while otime[i] > etime[j] - (otime[i] - otime[i+1])*tol: |
---|
| 61 | if count > 0: |
---|
| 62 | #print "spurious etime" |
---|
| 63 | if CML_end - CML_start > CML_max: |
---|
| 64 | CML_max = CML_end - CML_start |
---|
| 65 | CML_start, CML_end = j, j |
---|
| 66 | if AMLh_end - AMLh_start > AMLh_max: |
---|
| 67 | AMLh_max = AMLh_end - AMLh_start |
---|
| 68 | AMLh_start, AMLh_end = j, j |
---|
| 69 | if AMLd_end - AMLd_start > AMLd_max: |
---|
| 70 | AMLd_max = AMLd_end - AMLd_start |
---|
| 71 | AMLd_start, AMLd_end = j, j |
---|
| 72 | if AMLo_end - AMLo_start > AMLo_max: |
---|
| 73 | AMLo_max = AMLo_end - AMLo_start |
---|
| 74 | AMLo_start, AMLo_end = j, j |
---|
| 75 | j += 1 |
---|
| 76 | count += 1 |
---|
| 77 | if j+1 >= len(etime): break |
---|
| 78 | #print otime[i-1],etime[j-1]," ",otime[i],etime[j]," ",otime[i+1],etime[j+1] |
---|
| 79 | prevtempo = (otime[i] - otime[i-1]) |
---|
| 80 | nexttempo = (otime[i+1] - otime[i]) |
---|
[13c3fba] | 81 | |
---|
[9bec8fe] | 82 | current0 = (etime[j] > otime[i] - prevtempo*tol) |
---|
| 83 | current1 = (etime[j] < otime[i] + prevtempo*tol) |
---|
| 84 | |
---|
| 85 | # check correct tempo |
---|
| 86 | prev0 = (etime[j-1] > otime[i-1] - prevtempo*tolcontext) |
---|
| 87 | prev1 = (etime[j-1] < otime[i-1] + prevtempo*tolcontext) |
---|
| 88 | next0 = (etime[j+1] > otime[i+1] - nexttempo*tolcontext) |
---|
| 89 | next1 = (etime[j+1] < otime[i+1] + nexttempo*tolcontext) |
---|
| 90 | |
---|
| 91 | # check for off beat |
---|
| 92 | prevoffb0 = (etime[j-1] > otime[i-1] - prevtempo/2 - prevtempo*tolcontext) |
---|
| 93 | prevoffb1 = (etime[j-1] < otime[i-1] - prevtempo/2 + prevtempo*tolcontext) |
---|
| 94 | nextoffb0 = (etime[j+1] > otime[i+1] - nexttempo/2 - nexttempo*tolcontext) |
---|
| 95 | nextoffb1 = (etime[j+1] < otime[i+1] - nexttempo/2 + nexttempo*tolcontext) |
---|
| 96 | |
---|
| 97 | # check half tempo |
---|
| 98 | prevhalf0 = (etime[j-1] > otime[i-1] + prevtempo - prevtempo/2*tolcontext) |
---|
| 99 | prevhalf1 = (etime[j-1] < otime[i-1] + prevtempo + prevtempo/2*tolcontext) |
---|
| 100 | nexthalf0 = (etime[j+1] > otime[i+1] - nexttempo - nexttempo/2*tolcontext) |
---|
| 101 | nexthalf1 = (etime[j+1] < otime[i+1] - nexttempo + nexttempo/2*tolcontext) |
---|
| 102 | |
---|
| 103 | # check double tempo |
---|
| 104 | prevdoub0 = (etime[j-1] > otime[i-1] - prevtempo - prevtempo*2*tolcontext) |
---|
| 105 | prevdoub1 = (etime[j-1] < otime[i-1] - prevtempo + prevtempo*2*tolcontext) |
---|
| 106 | nextdoub0 = (etime[j+1] > otime[i+1] + nexttempo - nexttempo*2*tolcontext) |
---|
| 107 | nextdoub1 = (etime[j+1] < otime[i+1] + nexttempo + nexttempo*2*tolcontext) |
---|
| 108 | |
---|
| 109 | if current0 and current1 and prev0 and prev1 and next0 and next1: |
---|
| 110 | #print "YES!" |
---|
| 111 | CML_end = j |
---|
| 112 | CML_tot += 1 |
---|
| 113 | else: |
---|
| 114 | if CML_end - CML_start > CML_max: |
---|
| 115 | CML_max = CML_end - CML_start |
---|
| 116 | CML_start, CML_end = j, j |
---|
| 117 | if current0 and current1 and prevhalf0 and prevhalf1 and nexthalf0 and nexthalf1: |
---|
| 118 | AMLh_end = j |
---|
| 119 | AMLh_tot += 1 |
---|
| 120 | else: |
---|
| 121 | if AMLh_end - AMLh_start > AMLh_max: |
---|
| 122 | AMLh_max = AMLh_end - AMLh_start |
---|
| 123 | AMLh_start, AMLh_end = j, j |
---|
| 124 | if current0 and current1 and prevdoub0 and prevdoub1 and nextdoub0 and nextdoub1: |
---|
| 125 | AMLd_end = j |
---|
| 126 | AMLd_tot += 1 |
---|
| 127 | else: |
---|
| 128 | if AMLd_end - AMLd_start > AMLd_max: |
---|
| 129 | AMLd_max = AMLd_end - AMLd_start |
---|
| 130 | AMLd_start, AMLd_end = j, j |
---|
| 131 | if current0 and current1 and prevoffb0 and prevoffb1 and nextoffb0 and nextoffb1: |
---|
| 132 | AMLo_end = j |
---|
| 133 | AMLo_tot += 1 |
---|
| 134 | else: |
---|
| 135 | if AMLo_end - AMLo_start > AMLo_max: |
---|
| 136 | AMLo_max = AMLo_end - AMLo_start |
---|
| 137 | AMLo_start, AMLo_end = j, j |
---|
| 138 | # look for next matching beat |
---|
| 139 | count = 0 |
---|
| 140 | while otime[i] > etime[j] - (otime[i] - otime[i+1])*tolcontext: |
---|
| 141 | j += 1 |
---|
| 142 | if count > 0: |
---|
| 143 | #print "spurious etime" |
---|
| 144 | start = j |
---|
| 145 | count += 1 |
---|
| 146 | total = float(len(otime)) |
---|
| 147 | CML_tot /= total |
---|
| 148 | AMLh_tot /= total |
---|
| 149 | AMLd_tot /= total |
---|
| 150 | AMLo_tot /= total |
---|
| 151 | CML_cont = CML_max/total |
---|
| 152 | AMLh_cont = AMLh_max/total |
---|
| 153 | AMLd_cont = AMLd_max/total |
---|
| 154 | AMLo_cont = AMLo_max/total |
---|
| 155 | return CML_cont, CML_tot, AMLh_cont, AMLh_tot, AMLd_cont, AMLd_tot, AMLo_cont, AMLo_tot |
---|
| 156 | |
---|
| 157 | # for i in allfreq: |
---|
| 158 | # freq.append(float(i) / 2. / N * samplerate ) |
---|
| 159 | # while freq[i]>freqs[j]: |
---|
| 160 | # j += 1 |
---|
| 161 | # a0 = weight[j-1] |
---|
| 162 | # a1 = weight[j] |
---|
| 163 | # f0 = freqs[j-1] |
---|
| 164 | # f1 = freqs[j] |
---|
| 165 | # if f0!=0: |
---|
| 166 | # iweight.append((a1-a0)/(f1-f0)*freq[i] + (a0 - (a1 - a0)/(f1/f0 -1.))) |
---|
| 167 | # else: |
---|
| 168 | # iweight.append((a1-a0)/(f1-f0)*freq[i] + a0) |
---|
| 169 | # while freq[i]>freqs[j]: |
---|
| 170 | # j += 1 |
---|
| 171 | |
---|
| 172 | |
---|
| 173 | def gettruth(self): |
---|
| 174 | import os.path |
---|
| 175 | from aubio.txtfile import read_datafile |
---|
| 176 | datafile = self.input.replace('.wav','.txt') |
---|
| 177 | if not os.path.isfile(datafile): |
---|
| 178 | print "no ground truth " |
---|
| 179 | return False,False |
---|
| 180 | else: |
---|
| 181 | values = read_datafile(datafile,depth=0) |
---|
| 182 | old = -1000 |
---|
| 183 | for i in range(len(values)): |
---|
| 184 | now = values[i] |
---|
| 185 | period = 60 / (now - old) |
---|
| 186 | old = now |
---|
| 187 | values[i] = [now,period] |
---|
| 188 | return values |
---|
| 189 | |
---|
| 190 | |
---|
| 191 | def plot(self,oplots,results): |
---|
| 192 | import Gnuplot |
---|
| 193 | oplots.append(Gnuplot.Data(results,with='linespoints',title="auto")) |
---|
| 194 | |
---|
[14aae81] | 195 | def plotplot(self,wplot,oplots,outplot=None,extension=None,xsize=1.,ysize=1.,spectro=False): |
---|
[9bec8fe] | 196 | import Gnuplot |
---|
[14aae81] | 197 | from aubio.gnuplot import gnuplot_create, audio_to_array, make_audio_plot |
---|
[9bec8fe] | 198 | import re |
---|
| 199 | # audio data |
---|
| 200 | #time,data = audio_to_array(self.input) |
---|
| 201 | #f = make_audio_plot(time,data) |
---|
| 202 | |
---|
[14aae81] | 203 | g = gnuplot_create(outplot=outplot, extension=extension) |
---|
[9bec8fe] | 204 | oplots = [Gnuplot.Data(self.gettruth(),with='linespoints',title="orig")] + oplots |
---|
| 205 | g.plot(*oplots) |
---|