Efficient Acoustic Parameters for Speaker Recognition

In a scheme for the mechanical recognition of speakers, it is desirable to use acoustic parameters that are closely related to voice characteristics that distinguish speakers. This paper describes an investigation of an efficient approach to selecting such parameters, which are movitated by known relations between the voice signal and vocal-tract shapes and gestures. Rather than general measurements over the extent of an utterance, only significant features of selected segments are used. A simulation of a speaker recognition system was performed by manually locating speech events within utterances and using parameters measured at these locations to classify the speakers. Useful parameters were found in fundamental frequency, features of vowel and nasal consonant spectra, estimation of glottal source spectrum slope, word duration, and voice onset time. These parameters were tested in speaker recognition paradigms using simple linear classification procedures. When only 17 such parameters were used, no errors were made in speaker identification from a set of 21 adult male speakers. Under the same conditions, speaker verification errors of the order of 2% were also obtained.