Parameterization of a computational physical model for glottal flow using inverse filtering and high-speed videoendoscopy

Abstract

High-speed videoendoscopy, glottal inverse filtering, and physical modeling can be used to obtain complementary information about speech production. In this study, the three methodologies are combined to pursue a better understanding of the relationship between the glottal air flow and glottal area. Simultaneously acquired high-speed video and glottal inverse filtering data from three male and three female speakers were used. Significant correlations were found between the quasi-open and quasi-speed quotients of the glottal area (extracted from the high-speed videos) and glottal flow (estimated using glottal inverse filtering), but only the quasi-open quotient relationship could be represented as a linear model. A simple physical glottal flow model with three different glottal geometries was optimized to match the data. The results indicate that glottal flow skewing can be modeled using an inertial vocal/subglottal tract load and that estimated inertia within the glottis is sensitive to the quality of the data. Parameter optimisation also appears to favour combining the simplest glottal geometry with viscous losses and the more complex glottal geometries with entrance/exit effects in the glottis.