TheGlottal ModulationComponents ForSpeakerVoiceRecognition

Abstract

Glottalsignalsmaycarrymoreinformationaboutspeakersthanthespeech signalitselfwhentheyusedinthecon- textoftext-dependentspeakeridentificationsystem.Source signalsthatcamefromvoice-containing utteranceappearto improvethesystemperformanceupto10%comparingwith thespeechsignalitselfwhilethosecamefromnon-voicedutteranceappeartohavesimilarorprettylessaccuracycomparingwiththe speechsignals.Featuresare extractedfromeach speech/source signalbythemethodofAM-FMsignalmodulation.Thismethodisabletointroduceasetofdescriptors thatareencodedasaparameters forspeakeridentification. Thepurposes ofthispaperistoshowthatthevoicedspeech sourcesignalcontainsmorespeakerdependenceinformation thanthe speechsignalin the contextof speakeridentification system.Thespeechsignalanditssourceareparameterisedin themethodofAM-FM signalmodulation.Theresultsshow thatthesourcesignalwillgivebetterperformancejustwhen itcomesfrom voicedsource.