Calculations of the time varying vocal tract

Abstract Two models for simulating the time-varying vocal tract by means of wave digital filters (WDF) are described. In one model, the coefficients of the WDF are merely updated to simulate area changes. In the other, a “time-varying WDF” is used which allows a physically consistent description of the time-varying vocal tract. In order to test if the first model is sufficient for speech synthesis, the two models are compared one to another. Furthermore, three different models for simulating a glottal source are described. One model is based on an ordinary WDF, the second on a time-varying WDF, and the third on a model in which the glottis is assumed to be purely resistive. The signals of these models and their interactions with the vocal tract are compared one to another. It is shown that the differences between the two models of the vocal tract are very small in the simulation of speech, and not audible in hearing experiments. The differences are larger between the three glottis models, and also audible in hearing experiments. Simple simulation by means of a purely resistive source produces effects, in the same way as the other models do.