Anoveltextindependentspeakeridentificationsystemisproposed.Intheproposedsystem,the12-orderperceptuallinearpredictivecepstrumandtheirdeltacoefficientsinthespanoffiveframesareextractedfromthesegmentedspeechbasedonthemethodofpitchsynchronousanalysis.TheFisherratiosoftheoriginalcoefficientsthenbecalculated,andthecoefficientswhoseFisherratiosarebiggerareselectedtoformthe13-dimensionalfeaturevectorsofspeaker.TheGaussianmixturemodelisusedtomodelthespeakers.TheexperimentalresultsshowthattheidentificationaccuracyoftheproposedsystemisobviouslybetterthanthatofthesystemsbasedonotherconventionalcoefficientslikethelinearpredictivecepstralcoefficientsandtheMel-frequencycepstralcoefficients.
