High-order autoregressive modeling of individual speaker's qualities

Abstract

The modeling of individual speaker's properties is presented in this paper. The classic Autoregressive (AR) model is proposed for this purpose. The employed model order and parameter estimation technique gave much higher model order (up to 200 in some cases) in detailed spectral analysis of speech signals. Comparison of high-order AR model-based and Fourier transform-based spectral density functions suggests an idea that only a high-order AR model yields accurate values of fundamental and overtone frequencies. Results of initial experimental study show the potential of high-order AR model to be applied in estimation of individual speaker's spectral qualities and emotional state, evaluation of recover dynamics of patient's vocal folds.