Sex differences manifested in Japanese vowel formants : a likelihood ratio-based analysis

Abstract

This research aims to describe individual differences manifested in voice with a particular focus on the reflected individual differences within gender framework. The following is the central goal of the current research to investigate; differences in the ways and the degree of individualising information manifested in their speech acoustics between the sexes. The multi-speaker database of Japanese compiled by the National Research Institute of Police Science (Japan) was employed for the research. Out of more than 600 speakers stored in the database, 28 male and 28 female speakers were selected. Their age variable was controlled so that age groups are approximately balanced. The speech samples were recorded from two separate occasions for each speaker. Higher formant frequencies (F1/F2/F3) extracted from five Japanese vowels (/a/,/i,/u/,/e/,/o/) were labelled on Emu and the formant values were automatically extracted by the speech-analysis software. On the basis of the extracted formant values, speaker verification techniques were used as an analytical tool in measuring difference/similarity exhibited in the speakers' voice. A statistical inference called kernel-density multivariate likelihood ratio (abbreviated as MVLR) was used to quantify the difference/similarity manifested by comparing two speech samples either from the same speaker (same-speaker comparison, abbreviated as SS comparison) or from two different speakers (different-speaker comparison, DS comparison). Overall number of speakers who were correctly evaluated in their SS and DS comparisons was higher in the male group than in the female group. This confirms that gender{u00AD}correlated acoustic differences do manifest in higher formant values in Japanese. Each comparison made for each vowel was discussed in details to investigate the mechanism behind the gender difference shown in the formant value. Likelihood ratios calculated for both SS and DS comparisons contribute to the discussion.