FEM analyses of three-dimensional vocal tract models after tongue and mouth floor resection

Abstract

The accuracy of vocal tract transfer functions estimated by Finite Element Method (FEM) analyses was evaluated to find relationships between structural characteristics of vocal tracts and their acoustical characteristics. Three-dimensional vocal tract models were constructed from MR Images of four speakers (two normal persons and two patients after tongue and mouth floor resection), uttering the Japanese vowel /i/. Vocal tract transfer functions of the models were estimated by FEM and one-dimensional models, and peak positions of these functions were compared with formant frequencies of these speaker’s speech waves. The patient speech waves had a peculiar formant at around 1500 Hz. The peak positions of the patients estimated by FEM were similar to the formant frequencies of the patients, especially the formant at 1500 Hz. However, the transfer functions by the one-dimensional models had no peak at around 1500 Hz and the number of peaks was fewer than that by FEM. Since the patient vocal tract shapes were not symmetrical and bent right and left, sound pressure distributions estimated by FEM showed that the waves in the patient vocal tracts cannot be assumed as plane waves. Then, the one-dimensional models cannot describe acoustical characteristics of the patient vocal tracts.