Indexical cues to talker sex and individual talker identity extracted from vowels produced in sentence-length utterances

Abstract

The purpose of this study was to replicate and extend a previous study of indexical cuing [J.-A. Bachorowski and M. J. Owren, J. Acoust. Soc. Am. 106, 1054–1063 (1999)] by including more vowel sounds spoken in more diverse contexts. Specific combinations of acoustic parameters that should represent talker sex and individual talker identity were identified using predictions based on known sources of variation in vocal production-related anatomy. This study utilized 100 recordings of sentence-length utterances, produced by each of 43 male and 44 female undergraduates, as well as 22 stock-phrase recordings produced by these same participants. One of five vowel sounds (/æ, ε, i, ■, u/) was isolated from each sentence and analyzed for F0, F1, F2, F3, F4, vowel duration, jitter, shimmer, and harmonicity. Classification by talker sex was nearly perfect using a combination of cues related to both vocal-fold and vocal-tract anatomy. The accuracy of classification by individual identity depended strongly on cues relating to vocal tract-variation within sex.