A new method for structural analysis of perturbed pitch period series

Abstract

In medicine, analysis of pitch period series is extremely important for the detection and treatment of voice pathologies. This is because these sequences bring information of the physiological condition of the speech production system. In this article we propose a state space based approach for the analysis of pitch period series, combining structural analysis and the principal features of real disturbed periods. State space models were developed, considering different structures and complexity levels, and applied to a corpus of real periods. Statistical tests were applied to objectively evaluate the performance of this method. Most signals in the corpus were correctly analyzed with this methodology, and optimally estimated structural components properly modeled the involved dynamics. Experimental results suggest that the proposed approach is a suitable strategy for the analysis of perturbed pitch period series.