speakerdiarisation.py

def speakerdiarisationdf(hyp, frameRate, wavFile):
    audioname=[]
    starttime=[]
    endtime=[]
    speakerlabel=[]
            
    spkrChangePoints = np.where(hyp[:-1] != hyp[1:])[0]
    if spkrChangePoints[0]!=0 and hyp[0]!=-1:
        spkrChangePoints = np.concatenate(([0],spkrChangePoints))
    spkrLabels = []    
    for spkrHomoSegI in range(len(spkrChangePoints)):
        spkrLabels.append(hyp[spkrChangePoints[spkrHomoSegI]+1])
    for spkrI,spkr in enumerate(spkrLabels[:-1]):
        if spkr!=-1:
            audioname.append(wavFile.split('/')[-1].split('.')[0]+".wav")
            starttime.append((spkrChangePoints[spkrI]+1)/float(frameRate))
            endtime.append((spkrChangePoints[spkrI+1]-spkrChangePoints[spkrI])/float(frameRate))
            speakerlabel.append("Speaker "+str(int(spkr)))
    if spkrLabels[-1]!=-1:
        audioname.append(wavFile.split('/')[-1].split('.')[0]+".wav")
        starttime.append(spkrChangePoints[-1]/float(frameRate))
        endtime.append((len(hyp) - spkrChangePoints[-1])/float(frameRate))
        speakerlabel.append("Speaker "+str(int(spkrLabels[-1])))
    #
    speakerdf=pd.DataFrame({"Audio":audioname,"starttime":starttime,"endtime":endtime,"speakerlabel":speakerlabel})
    
    spdatafinal=pd.DataFrame(columns=['Audio','SpeakerLabel','StartTime','EndTime'])
    i=0
    k=0
    j=0
    spfind=""
    stime=""
    etime=""
    for row in speakerdf.itertuples():
        if(i==0):
            spfind=row.speakerlabel
            stime=row.starttime
        else:
            if(spfind==row.speakerlabel):
                etime=row.starttime        
            else:
                spdatafinal.loc[k]=[wavFile.split('/')[-1].split('.')[0]+".wav",spfind,stime,row.starttime]
                k=k+1
                spfind=row.speakerlabel
                stime=row.starttime
        i=i+1
    spdatafinal.loc[k]=[wavFile.split('/')[-1].split('.')[0]+".wav",spfind,stime,etime]
    return spdatafinal
  
 pass1hyp = -1*np.ones(len(vad))
pass1hyp[vad] = frameClust
spkdf=speakerdiarisationdf(pass1hyp, frameRate, wavFile)