简介:Inthispaperanewtext-independentspeakerverificationmethodGSMSVisproposedbasedonlikelihoodscorenormalization.Inthisnovelmethodaglobalspeakermodelisestablishedtorepresenttheuniversalfeaturesofspeechandnormalizethelikelihoodscore.Statisticalanalysisdemonstratesthatthisnormalizationmethodcanremovecommonfactorsofspeechandbringthedifferencesbetweenspeakersintoprominence.Asaresulttheequalerrorrateisdecreasedsignificantly,verificationprocedureisacceleratedandsystemadaptabilitytospeakingspeedisimproved.