A STUDY OF HARMONIC FEATURES FOR THE SPEAKER RECOGNITION

In this paper the harmonic features based on the harmonic decompositio n of the Hildebrand-Prony line spectrum are introduced. A Hildebrand-P rony method of spectral analysis was applied because of its high resol ution and accuracy. Comparative tests with the LP and LP-cepstral feat ures were made with 50 speakers from the Slovene database SNABI (isola ted words corpus) and 50 speakers of the German database BAS Siemens 1 00 (utterances of sentences). With both databases the advantages of th e harmonic features were noticed especially for the speaker identifica tion while for the speaker verification the harmonic features have per formed better on the SNABI database and as good as the LP cepstral fea tures on the BAS Siemens 100 database. (C) 1997 Elsevier Science B.V.