Contributions of Vowel and Nasal Sounds to Speaker Identification

Abstract

Speaker identification techniques which extract speaker characteristics from single sound units can be applicable to both text-dependent and text-independent speech. Experiments have shown that identification accuracy improves if decisions are based upon combined information from several different sound units. With such techniques, it would be helpful if potential sound units could be rank ordered according to how much each contributes toward successful speaker identification. In text-independent speaker identification there is no control over the number of sound units in the test speech sample, so it would be beneficial if the probability of success could be predicted not only from the population size but also from the number and type of sound units available. Speech samples from ten male speakers are being used to evaluate ten non-diphthongized vowels and three nasals. Experimental results are presented on rank ordering these sounds and the determination of conditional probabilities based upon the selection of sounds. [Sponsored by Air Force Systems Command, Rome Air Development Center, Griffiss Air Force Base, New York 13441.]