Disordered voice evaluation based on correlogram of auditory model

Abstract

Sustained vowel analysis is an important method in disordered voice evaluation. Conventional method generally depends on pitch detection, which may be invalid in hoarse or seriously disordered vowel samples analysis. In our vowel analysis, correlogram and cross-channel correlation in auditory model are proposed as a robust and accurate criterion. Instead of manual selection, a method of selecting stable part of samples automatically is adopted to improve the feasibility of the vowel analysis. Experiments show the robustness and effectiveness of this method.