Articulatory speech synthesis and the analysis of boundary conditions

Abstract

In this work, a computational fluid dynamics (CFD) approach was used to model the unsteady fluid flow in idealized human vocal tracts. The speech waveform synthesis was based on the slightly compressible Reynolds-averaged Navier–Stokes (RANS) equations. A K-epsilon turbulence model has been used to represent the effects of turbulence. The vocal tract geometry was determined from the mid-sagittal cut through the three-dimensional vocal tract shape. The moving vocal tract shapes during the speech articulation were estimated using interpolation technique. The excitation signal for the simulation was accomplished by specifying a time-varying area at the inlet to the bottom of the pharyngeal cavity and a sawtooth waveform was used to represent the area variation in time. Different boundary conditions were applied at the inlet of the vocal tract. One is based on specifying the volume velocity and the other is based upon specifying the particle velocity, respectively. We will further analyze the effects of different boundary conditions using an analogy of transmission line T-network. Finally, we will present the waveform and spectrum of some synthesized voiced sounds based on the above CFD approach and boundary conditions. [This work was supported by Motorola and NSF.]