On the model of vocal tract dynamics

Abstract

An attempt to model a true vocal tract shape is presented. The base data were articulatory data taken from the cineradiographic recordings and the speech signal simultaneously registered (Bolla, Földi, 1987). Moreover, information concerning distances between the reference points, labiograms, photopalatograms and photolinguograms were used. A non-uniform three-dimensional vocal tract model was applied. The most difficult problem was how to evaluate the varations of the lateral dimensions along the longitudinal axis of the vocal tract. Only the lip opening dimensions for each speech sound were the objects of direct measurements. After the photo of the subject's hard palate one of the several plaster casts made for other male subjects was chosen as the most similar one. On its basis the dimensions in this region were reconstructed and assume to be constant. The lateral dimensions in other parts of the vocal tract were reconstructed after the formant frequencies measured for each Polish vowel spoken by the subject whose vocal tract was modelled. These dimensions were adjusted in such way that they were common for all vowels and the corresponding formant frequencies were obtained by varying vertical dimensions according to the X-ray data. This model was applied to study the relationship between the shape variation of the vocal tract and its acoustic output, especially in the case of study of transient sounds spoken with vocalic neighbourhood.