Development of an Acoustic Simulation Method during Phonation of the Japanese Vowel /a/ by the Boundary Element...

The purpose of the present study was to establish the method for an acoustic simulation of a vocal tract created from CT data during phonation of the Japanese vowel /a/ and to verify the validity of the simulation.

The subjects were 15 healthy adults (8 males, 7 females). The vocal tract model was created from CT data acquired during sustained phonation of the Japanese vowel /a/. After conversion to a mesh model for analysis, a wave acoustic analysis was performed with a boundary element method. The wall and the bottom of the vocal tract model were regarded as a rigid wall and a nonrigid wall, respectively. The acoustic medium was set to 37°C, and a point sound source was set in the place corresponding to the vocal cord as a sound source. The first and second formant frequencies (F1 and F2) were calculated. For 1 of the 15 subjects, the range from the upper end of the frontal sinus to the tracheal bifurcation was scanned, and 2 models were created: model 1 included the range from the frontal sinus to the tracheal bifurcation; and model 2 included the range from the frontal sinus to the glottis and added a virtually extended trachea by 12 cm cylindrically. F1 and F2 calculated from models 1 and 2 were compared. To evaluate the validity of the present simulation, F1 and F2 calculated from the simulation were compared with those of the actual voice and the sound generated using a solid model and a whistle-type artificial larynx. To judge the validity, the vowel formant frequency discrimination threshold reported in the past was used as a criterion. Namely, the relative discrimination thresholds (%), dividing ▵F by F, where F was the formant frequency calculated from the simulation, and ▵F was the difference between F and the formant frequency of the actual voice and the sound generated using the solid model and artificial larynx, were obtained.

F1 and F2 calculated from models 1 and 2 were similar. Therefore, to reduce the exposure dose, the remaining 14 subjects were scanned from the upper end of the frontal sinus to the glottis, and model 2 with the trachea extended by 12 cm virtually was used for the simulation. The averages of the relative discrimination thresholds against F1 and F2 calculated from the actual voice were 5.9% and 4.6%, respectively. The averages of the relative discrimination thresholds against F1 and F2 calculated from the sound generated by using the solid model and the artificial larynx were 4.1% and 3.7%, respectively.

The Japanese vowel /a/ could be simulated with high validity for the vocal tract models created from the CT data during phonation of /a/ using the boundary element method.