Perceptual scaling of voice identity: common dimensionsfor different vowels and speakers

摘要

The aims of our study were: (1) to determine ifthe acoustical parameters used by normal subjects to dis-criminate between different speakers vary whencomparisons are made between pairs of two of the same ordifferent vowels, and if they are different for male andfemale voices; (2) to ask whether individual voices canreasonably be represented as points in a low-dimensionalperceptual space such that similarly sounding voices arelocated close to one another. Subjects were presented withpairs of voices from 16 male and 16 female speakersuttering the three French vowels "a", "i" and "u" andasked to give speaker similarity judgments. Multidimen-sional analyses of the similarity matrices were performedseparately for male and female voices and for three types ofcomparisons: same vowels, different vowels and overallaverage. The resulting dimensions were then interpreteda posteriori in terms of relevant acoustical measures. Forboth male and female voices, a two-dimensional perceptualspace was found to be most appropriate, with axes largelycorresponding to contributions of the larynx (pitch) andsupra-laryngeal vocal tract (formants), mirroring the twolargely independent components of source and filter invoice production. These perceptual spaces of male andfemale voices and their corresponding voice samples areavailable at: http://vnl.psy.gla.ac.uk section Resources.