Kinematic modelling of dipthong articulation

Abstract

The Sigma-Lognormal model has successfully been applied to handwriting modeling but never to acoustic-phonetic articulation. The hypothesis of this paper is that vocal tract dynamics, which includes jaw and tongue can be approached by the Kinematic Theory. In speech analysis, the movement of the tongue and jaw has been linked to the variation on first and second formants. In this paper, we explore the Kinematic hypothesis, based on diphthong pronunciation, which invoke the most extreme tongue gestures in the vowel triangle, estimation of their formants, and transformation of these to space for evaluating the speed profile. The estimated speed profile is modelled by the sigma lognormal model of the Kinematic Theory. An average reconstruction error of 20 dB has been obtained in the experiments carried out with 20 different volunteers. This result validates the work hypothesis, opening a new research line in speech processing.