Active model for extraction of nasality and its perceptual evaluation

Abstract

A description of nasality in the acoustic domain and a model for the extraction of nasality are discussed from a perceptual point of view. Nasality is perceptually regarded as some kind of distortion against the vowel quality, and the acoustic correlate of nasality is assumed to be the degree of spectral deviation from the oral sound. As a functional model for the extraction of nasality, a kind of A-b-S model is assumed. In the model the trial oral spectra are synthesized internally, the input speech spectrum is compared with each of the trial spectra, and the minimum value of the matching errors is taken out as the parameter representing nasality after the best matching is obtained. The trial spectra are specified as those calculated from the equation for the spectral description of vowels given by Fant [“Acoustic Theory of Speech Production” (1960)]. The correspondences were examined between the parameters extracted in the model and the results of the identification test on nasality using the oral and nasal sounds in natural speech.