7th International Workshop on Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA 2011)

Florence, Italy
August 25-27, 2011

Automatic GRBAS Assessment Using Complexity Measures and a Multiclass GMM-Based Detector

Julian D. Arias-Londoño (1), Juan Ignacio Godino-Llorente (2), Nicolás Sáenz-Lechón (2), Víctor Osma-Ruiz (2), Juana M. Gutiérrez-Arriola (2)

(1) Bioinstrumentation Research Group, Universidad Antonio Nariño, Bogotá, Colombia
(2) Department of Circuits & Systems Engineering, E.U.I.T de Telecomunicación, Universidad Politécnica de Madrid, Spain.

This paper presents a system for the automatic assessment of pathological voice quality according to the GRBAS protocol, which uses a short time scheme and a characterization based on 9 complexity measures, including conventional nonlinear statistics and 7 entropy based features. The classification is carried out using three different multiclass classification strategies all of them based on Gaussian Mixture Models. The performance of the system is measured in terms of efficiency and a statistical agreement index. The results show that the complexity analysis provides relevant information for the automatic assessment of voice quality according to the GRBAS protocol.

Index Terms. automatic Grbas assessment, complexity measures, multiclass classification

Full Paper (reprinted with permission from Firenze University Press)

Bibliographic reference.  Arias-Londoño, Julian D. / Godino-Llorente, Juan Ignacio / Sáenz-Lechón, Nicolás / Osma-Ruiz, Víctor / Gutiérrez-Arriola, Juana M. (2011): "Automatic GRBAS assessment using complexity measures and a multiclass GMM-based detector", In MAVEBA-2011, 111-114.