Estimation of vocal fold geometry and stiffness from voice acoustics
Zhaoyan Zhang · The Journal of the Acoustical Society of America · 2019
Many speech applications require estimating vocal fold properties from the produced acoustics. While there have been many previous research on solving the inverse problem, they are often based on lumped-element models of phonation, whose model parameters are difficult to relate to realistic vocal fold properties. This study explores the feasibility of inferring physiologically realistic vocal fold properties, including vocal fold length, thickness, depth, anisotropic stiffness moduli, and the subglottal pressure, from the produced acoustics, using a three-dimensional phonation model and a Bayesian inference approach. To reduce the computational cost associated with the use of a three-dimensional model and the large number of inputs to be estimated, we explore the possibility of improving computational efficiency and estimation accuracy using the large amount of three-dimensional simulation data available from our previous research. Preliminary results show that joint likelihood probabilities can be reasonably estimated based on the available simulation data, significantly reducing computational costs of Bayesian inference. The approach is able to estimate the control parameters of voice production from the produced acoustics with reasonable accuracy. It is observed that certain control parameters, particularly vocal fold stiffness, consistently have large estimation errors than others. [Work supported by NIH.]