Finite-difference time-domain method for vocal tract acoustics

The finite-difference time-domain (FDTD) method effectively analyzes the acoustic properties of static 3-D vocal tract shapes. Its simple algorithm enables acceleration via parallel computing. Using the FDTD method with a message-passing interface, the vocal tract acoustics of five Japanese vowels were investigated based on MRI data. The results showed that the laryngeal cavity uniquely generates the fourth formant, and the bilateral piriform fossae function as a two-coupled oscillator system, producing two spectral dips in the transfer function. This method was also applied to the vocal tracts of male opera singers to examine the acoustic role of the hypopharynx in formant generation. In these cases, calculating a 20-ms impulse response of a 0.5-mm resolution vocal tract required approximately 1 h. This method is now being extended to assess the acoustic effects of sinonasal surgery by simulating acoustic changes in virtually modified sinonasal cavities. Further acceleration using general-purpose computing on graphics processing units (GPGPU) reduced the computation time for the same simulation to 93 s. These findings demonstrate the potential of FDTD with GPGPU for efficient and detailed acoustic analysis in clinical applications.