An acoustic model of pathological voice production is presented. It describes the non-linear effects occurring in the acoustic waveform of disordered voices. The noise components such as fundamental frequency and amplitude irregularities and variations, sub-harmonic components, turbulent noise and voice breaks are formally expressed as a result of random time function influences on the excitation function and the glottal filter. A method for quantitative evaluation of these random functions is described. The method computes their statistical characteristics which can be useful in assessing voice in clinical practice. More than 33 acoustic parameters are computed: average fundamental frequency, phonatory frequency range, several frequency and amplitude short- and long-term perturbation and variation measures, noise-to-harmonic ratio, voice turbulence and soft phonation indexes, quantitative measures of voice breaks, sub-harmonic components and vocal tremors. This set of parameters, which corresponds to the model, allows a multi-dimensional voice quality assessment. A computer system based on above model and method was developed for the CSL model 4300 (Kay Elemetrics Corp.). A group of 68 people with normal and disordered voices was analyzed using the system in order to define normative values for the acoustic voice parameters.

Keywords: acoustic voice analysis, signal processing, speech pathology, phoniatrics.