Three-Dimensional Representation of Japanese Phonemes.

Abstract

The spectra, which represent nine consonants /p, t, k, b, d, g, m, n, ŋ/, were defined at the boundary between following vowel. Nine consonants of 20-dimensional vectors, which are projected orthogonally on three-dimensional subspace (that is created by the three directions which promote separation in manner and place of articulation), produce nearly a triangular prism. Although spatial representation of vowels has early been tried, physical and analytic expressions of consonants based on actual speech are supposed to be, for the first time, tried here. Nine Consonants were uttered as C2 of two-syllable C1V1C2V2 (consonantvowel-consonant-vowel) words. In order to inspect the existence of physical spectra characterizing each consonant, multivariate statistical analysis of variance for four-factor design with repeated measurements was performed, considering C1, V1, C2 and V2 as four factors.