This proposal focuses on the development of nonlinear signal processing systems and algorithms that can model or extract information about two types of nonlinear and time-varying phenomena in speech production: modulations and turbulence. A model is proposed for short-time speech resonances that combines both amplitude and frequency modulation. Preliminary experimental findings are consistent with this model. Fractal models for describing the geometric fragmentation of the speech signals will be used to quantify the degree of speech turbulence. A nonlinear filtering algorithm is proposed to measure the short-time fractal dimension of speech signals.
