Constructing A Time Varying Multi Vowel Synthesizer In Matlab

A digital filter simulation of the vocal cavity, positioned to create a vowel, that was excited by a glottal pulse train with the filter output played through a sound card can be accomplished using MATLAB with a modest effort. The simulation of a time-varying vocal tract that generated sequential vowels, with a natural transition, required the modification of the digital filter coefficients as the glottal pulse train excited the vocal tract model. This paper presents a procedure for the implementation of a pseudo time-varying digital filter that simulated the vocal tract producing sequential vowels. The procedure allowed the controlled movement of the poles associated with one vowel into the correct position for the second vowel. The pole locations of the vowels were represented in polar form using the formant frequencies, scaled by the folding frequency. The radial locations of the poles were arrived at by matching the spectral analysis of the digital filter simulation with the actual vowel spectrum. Since MATLAB does not allow time-varying digital filters, the time scale used for the incremental changes in the digital filter coefficients was chosen to be the glottal pulse period, which was set for a male speaker. The transition interval from one vowel to the next was selected so the listener would sense a natural change. Simulation results were generated for a /u/ /i/ vowel combination. The M-files necessary for the simulation were included as an Appendix.