A stochastic model of fundamental period perturbation and its application to perception of pathological voice quality

Proposes a stochastic model of fundamental period perturbation and applies it to the perception of pathological voice quality. A fundamental period perturbation is generated by a second-order auto-regressive moving average (ARMA) model which includes excitation by a random white noise. Standard deviation, pole frequency and pole bandwidth were used as parameters of the model. Sustained vowels were synthesized by systematically manipulating the model parameters and subjected to the perceptual experiment to understand the roles of the parameters in the perceived quality of a pathological voice. Five subjects participated in the experiment and were asked to judge whether a vowel sample was normal or pathological. It was found that the perceptual impression of the pathological voice was associated not only with the magnitude but also the pole frequency and bandwidth of the fundamental period perturbation