Recent Progress in Speech Analysis Methods

Toward Clinical Applications

Abstract

Speech analysis methods can be divided into two classes, parametric and nonparametric. The parametric method includes a speech production model in the analysis process, while the nonparametric method does not. The former is exemplified by the well-established linear prediction analysis and the latter often utilizes short-term Fourier analysis. Recent progress in speech analysis methods can be found in an ARX (Auto-regressive with exogenous input) method, a parametric approach, as well as STRAIGHT (Speech Transformation and Representation using Adaptive Interpolation of weighted spectrum) method, a nonparametric approach. The ARX method incorporates a sophisticated ARX-based speech production model, making it possible to separately estimate glottal flow parameters and formants. Basic and clinical applications of the ARX method are presented in the field of voice and speech disorders, including acoustic evaluation of dysarthria, esophageal voice and dysphonia.