Comparison of the finite element method, the multimodal method and the transmission-line model for the computation of vocal tract transfer functions

The acoustic properties of vocal tract are usually characterized by its transfer function from the input acoustic volume flow at the glottis to the radiated acoustic pressure. These transfer functions can be computed with acoustic models. Three-dimensional acoustic simulation are used to take into account accurately the three-dimensional vocal tract shape and to generate valid results even at high frequency. Finite element models, finite difference methods, three-dimensional waveguide meshes, or the multimodal method have been used for this purpose. However, these methods require much more computation time than simple one-dimensional models. Among these methods, the multimodal method can achieve the shortest computation times. However, all the previous implementations had limitations regarding the geometrical shapes and the losses. In this work, we evaluate a new implementation that intends to overcome these limitations. Vowel transfer functions obtained with this new implementation are compared with a transmission-line model and a proven, robust and highly accurate method: the finite element method. While the finite element method remains the most reliable, the multimodal method generates similar transfer functions in much less time. The transmission line model gives valid results for the four first resonances.