source: python/aubio/task/onset.py @ 67d0a8b

from aubio.task.task import task
from aubio.aubioclass import *

class taskonset(task):
        def __init__(self,input,output=None,params=None):
                """ open the input file and initialize arguments 
                parameters should be set *before* calling this method.
                """
                task.__init__(self,input,params=params)
                self.opick = onsetpick(self.params.bufsize,
                        self.params.hopsize,
                        self.channels,
                        self.myvec,
                        self.params.threshold,
                        mode=self.params.onsetmode,
                        dcthreshold=self.params.dcthreshold,
                        derivate=self.params.derivate)
                self.olist = [] 
                self.ofunc = []
                self.maxofunc = 0
                self.last = 0
                if self.params.localmin:
                        self.ovalist   = [0., 0., 0., 0., 0.]

        def __call__(self):
                task.__call__(self)
                isonset,val = self.opick.do(self.myvec)
                if (aubio_silence_detection(self.myvec(),self.params.silence)):
                        isonset=0
                if self.params.storefunc:
                        self.ofunc.append(val)
                if self.params.localmin:
                        if val > 0: self.ovalist.append(val)
                        else: self.ovalist.append(0)
                        self.ovalist.pop(0)
                if (isonset > 0.):
                        if self.params.localmin:
                                # find local minima before peak 
                                i=len(self.ovalist)-1
                                while self.ovalist[i-1] < self.ovalist[i] and i > 0:
                                        i -= 1
                                now = (self.frameread+1-i)
                        else:
                                now = self.frameread
                        # take back delay
                        if self.params.delay != 0.: now -= self.params.delay
                        if now < 0 :
                                now = 0
                        if self.params.mintol:
                                # prune doubled 
                                if (now - self.last) > self.params.mintol:
                                        self.last = now
                                        return now, val
                        else:
                                return now, val 


        def fprint(self,foo):
                print self.params.step*foo[0]

        def eval(self,inputdata,ftru,mode='roc',vmode=''):
                from aubio.txtfile import read_datafile 
                from aubio.onsetcompare import onset_roc, onset_diffs, onset_rocloc
                ltru = read_datafile(ftru,depth=0)
                lres = []
                for i in range(len(inputdata)): lres.append(inputdata[i][0]*self.params.step)
                if vmode=='verbose':
                        print "Running with mode %s" % self.params.onsetmode, 
                        print " and threshold %f" % self.params.threshold, 
                        print " on file", self.input
                #print ltru; print lres
                if mode == 'local':
                        l = onset_diffs(ltru,lres,self.params.tol)
                        mean = 0
                        for i in l: mean += i
                        if len(l): mean = "%.3f" % (mean/len(l))
                        else: mean = "?0"
                        return l, mean
                elif mode == 'roc':
                        self.orig, self.missed, self.merged, \
                                self.expc, self.bad, self.doubled = \
                                onset_roc(ltru,lres,self.params.tol)
                elif mode == 'rocloc':
                        self.v = {}
                        self.v['orig'], self.v['missed'], self.v['Tm'], \
                                self.v['expc'], self.v['bad'], self.v['Td'], \
                                self.v['l'], self.v['labs'] = \
                                onset_rocloc(ltru,lres,self.params.tol)

        def plot(self,onsets,ofunc,wplot,oplots,nplot=False):
                import Gnuplot, Gnuplot.funcutils
                import aubio.txtfile
                import os.path
                from numpy import arange, array, ones
                from aubio.onsetcompare import onset_roc

                x1,y1,y1p = [],[],[]
                oplot = []
                if self.params.onsetmode in ('mkl','kl'): ofunc[0:10] = [0] * 10

                self.lenofunc = len(ofunc) 
                self.maxofunc = max(ofunc)
                # onset detection function 
                downtime = arange(len(ofunc))*self.params.step
                oplot.append(Gnuplot.Data(downtime,ofunc,with_='lines',title=self.params.onsetmode))

                # detected onsets
                if not nplot:
                        for i in onsets:
                                x1.append(i[0]*self.params.step)
                                y1.append(self.maxofunc)
                                y1p.append(-self.maxofunc)
                        #x1 = array(onsets)*self.params.step
                        #y1 = self.maxofunc*ones(len(onsets))
                        if x1:
                                oplot.append(Gnuplot.Data(x1,y1,with_='impulses'))
                                wplot.append(Gnuplot.Data(x1,y1p,with_='impulses'))

                oplots.append((oplot,self.params.onsetmode,self.maxofunc))

                # check if ground truth datafile exists
                datafile = self.input.replace('.wav','.txt')
                if datafile == self.input: datafile = ""
                if not os.path.isfile(datafile):
                        self.title = "" #"(no ground truth)"
                else:
                        t_onsets = aubio.txtfile.read_datafile(datafile)
                        x2 = array(t_onsets).resize(len(t_onsets))
                        y2 = self.maxofunc*ones(len(t_onsets))
                        wplot.append(Gnuplot.Data(x2,y2,with_='impulses'))
                        
                        tol = 0.050 

                        orig, missed, merged, expc, bad, doubled = \
                                onset_roc(x2,x1,tol)
                        self.title = "GD %2.3f%% FP %2.3f%%" % \
                                ((100*float(orig-missed-merged)/(orig)),
                                 (100*float(bad+doubled)/(orig)))


        def plotplot(self,wplot,oplots,outplot=None,extension=None,xsize=1.,ysize=1.,spectro=False):
                from aubio.gnuplot import gnuplot_create, audio_to_array, make_audio_plot, audio_to_spec
                import re
                # prepare the plot
                g = gnuplot_create(outplot=outplot, extension=extension)
                g('set title \'%s\'' % (re.sub('.*/','',self.input)))
                if spectro:
                        g('set size %f,%f' % (xsize,1.3*ysize) )
                else:
                        g('set size %f,%f' % (xsize,ysize) )
                g('set multiplot')

                # hack to align left axis
                g('set lmargin 3')
                g('set rmargin 6')

                if spectro:
                        import Gnuplot
                        minf = 50
                        maxf = 500 
                        data,time,freq = audio_to_spec(self.input,minf=minf,maxf=maxf)
                        g('set size %f,%f' % (1.24*xsize , 0.34*ysize) )
                        g('set origin %f,%f' % (-0.12,0.65*ysize))
                        g('set xrange [0.:%f]' % time[-1]) 
                        g('set yrange [%f:%f]' % (minf,maxf))
                        g('set pm3d map')
                        g('unset colorbox')
                        g('set lmargin 0')
                        g('set rmargin 0')
                        g('set tmargin 0')
                        g('set palette rgbformulae -25,-24,-32')
                        g.xlabel('time (s)',offset=(0,1.))
                        g.ylabel('freq (Hz)')
                        g('set origin 0,%f' % (1.0*ysize) ) 
                        g('set format x "%1.1f"')
                        #if log:
                        #       g('set yrange [%f:%f]' % (max(10,minf),maxf))
                        #       g('set log y')
                        g.splot(Gnuplot.GridData(data,time,freq, binary=1, title=''))
                else:
                        # plot waveform and onsets
                        time,data = audio_to_array(self.input)
                        wplot = [make_audio_plot(time,data)] + wplot
                        g('set origin 0,%f' % (0.7*ysize) )
                        g('set size %f,%f' % (xsize,0.3*ysize))
                        g('set format y "%1f"')
                        g('set xrange [0:%f]' % max(time)) 
                        g('set yrange [-1:1]') 
                        g('set noytics')
                        g('set y2tics -1,1')
                        g.xlabel('time (s)',offset=(0,0.7))
                        g.ylabel('amplitude')
                        g.plot(*wplot)

                # default settings for next plots
                g('unset title')
                g('set format x ""')
                g('set format y "%3e"')
                g('set tmargin 0')
                g.xlabel('')

                N = len(oplots)
                y = 0.7*ysize # the vertical proportion of the plot taken by onset functions
                delta = 0.035 # the constant part of y taken by last plot label and data
                for i in range(N):
                        # plot onset detection functions
                        g('set size %f,%f' % ( xsize, (y-delta)/N))
                        g('set origin 0,%f' % ((N-i-1)*(y-delta)/N + delta ))
                        g('set nokey')
                        g('set xrange [0:%f]' % (self.lenofunc*self.params.step))
                        g('set yrange [0:%f]' % (1.1*oplots[i][2]))
                        g('set y2tics ("0" 0, "%d" %d)' % (round(oplots[i][2]),round(oplots[i][2])))
                        g.ylabel(oplots[i][1])
                        if i == N-1:
                                g('set size %f,%f' % ( xsize, (y-delta)/N + delta ) )
                                g('set origin 0,0')
                                g.xlabel('time (s)', offset=(0,0.7))
                                g('set format x')
                        g.plot(*oplots[i][0])

                g('unset multiplot')