Untersuchungen zur Sprechtraktakustik

TL;DR

This thesis advances the numerical simulation of vocal and nasal tract acoustics using 3D FDTD methods, incorporating MRI and CT data for improved accuracy in modeling resonance and absorption mechanisms.

Contribution

It develops a 3D FDTD simulation framework for vocal tract acoustics, integrating MRI and CT data, and introduces methods for modeling absorption and impedance translation.

Findings

Cross section shape significantly affects resonance characteristics.

Elliptical approximation of cross section is effective for 2D modeling.

Accurate impedance translation improves simulation fidelity.

Abstract

This thesis investigates acoustic properties of the vocal tract. Starting from a historical background (to name a few: Galen, Ibn Sina/Avicenna, Mersenne, Hooke, Euler, Kempelen, Abbe Mical, Kratzenstein, Wheatstone, Helmholtz, Riesz, Dudley, Dunn, Kelly, Lochbaum, Chiba, Kajiyama, Saito, Itakura, Burg, Gray...) finaly an improvement in respect to the tube model of the vocal tract is developed: The acoustics of vocal tract is numerical simulated in 3 dimensions with Finite-difference in time-domain (FDTD) applied on an oral-cavity spatially constructed by parameter estimation of speech signals and laterally by MRI (Magnet Resonance Imaging). Further, the FDTD is applied on the acoustics of the nasal tract, including paranasal cavities. The spatial properties of the nasal tract are drawn for CT (Computer Tomographies). It is shown, that the (absolute) cross section area influences the resonance charateritic of the vocal tract significant, further the actual shape of the cross section introduces zeros in the transfer functions. The shape can be approximated by an elliptical contour, and a 2 dimensional FDTD approach is discussed. Important absorption machanism are modeled, and a method to translate the hounsfield-value of a voxel to an acoustic impedance, is shown, which leads to a more accurate simulation. The thesis further elucidates some aspects of a numerical efficient FDTD-implementation.