Energy distribution analysis and nonlinear dynamical analysis of phonation in patients with Parkinson's disease

Abstract

Patients with Parkinson's disease (PD) have been reported to exhibit vocal impairment during the course of PD. Recently, development of automatic PD severity assessment based on acoustical characteristics from voice recordings has been attempted. However, objective extraction of appropriate features that can characterize PD symptoms faces many problems, due to the prevalence of aperiodicity in PD voices, rendering traditional perturbation analysis unreliable. The present study attempted to examine the validity of more advanced acoustic analysis techniques based on energy distribution measures and nonlinear dynamical measures. All of the features were extracted from sustained phonations of the vowel /a/ produced by 16 PD patients and 20 age-matched non- pathologic subjects. Results revealed that the energy distribution measures, such as glottal-to-noise excitation (GNE), and empirical mode decomposition excitation ratio (EMD-ER), as well as nonlinear dynamical measures including correlation dimension (D2), permutation entropy (PE), and detrended fluctuation analysis (DFA) were effective in discerning between PD and normal voices. This finding suggests that both energy distribution and nonlinear dynamical analyses could be appropriate measures in determining the status of PD voice.