SPEAKER EMOTION RECOGNITION BASED ON SPEECH FEATURES AND CLASSIFICATION TECHNIQUES

Abstract

The smain sprinciple sbehind sthe sspeaker srecognition ssystem sis sthat sof sfeature sextraction sand smatching. sFeature sextraction sdeals swith sthe sextraction sor sreading sof simportant scharacteristics sor sfeature sfrom sa shuman sspeech ssignal. sThose scharacteristics smight sbe spitch, sor sfrequency, swhich sare sunique sto sdifferent spersons. sSpeaker srecognition smethods scan salso sbe sdivided sinto stextindependent sand stext-dependent smethods. sIn sa stext-independent ssystem, sspeaker smodels scapture scharacteristics sof ssomebody’s sspeech swhich sshow sup sirrespective sof swhat sone sis ssaying. sIn sa stext-dependent ssystem, son sthe sother shand, sthe srecognition sof sthe sspeaker’s sidentity sis sbased son shis sor sher sspeaking sone sor smore sspecific sphrases, slike spasswords, scard snumbers, sPIN scodes, setc. sAll sspeaker srecognition ssystems shave sto sserve stwo sphases. sThe sfirst sone sis sreferred sto sthe senrolment ssession sor straining sphase swhile sthe ssecond sone sis sreferred sto sas sthe soperation ssession sor stesting sphase. sIn sthe straining sphase, seach sregistered sspeaker shas sto sprovide ssamples sof stheir sspeech sso sthat sthe ssystem scan sbuild sor strain sa sreference smodel sfor sthat sspeaker. sIt sconsists sof stwo smain sparts. sThe sfirst spart sconsists sof sprocessing seach spersons sinput svoice ssample sto scondense sand ssummarize sthe scharacteristics sof stheir svocal stracts. sThe ssecond spart sinvolves spulling seach sperson's sdata stogether sinto sa ssingle, seasily smanipulated smatrix. sThe stesting ssystem smirrors sthe straining ssystem sarchitecture. sFirst sthe sinput ssignal sis sanalyzed, sand sthen sit sis scompared sto sthe sdata sstored sin sthe scodebook. sThe sdifference sis sused sto smake srecognition sdecision.