Title: Acoustic study of dysarthria

Authors: Eduardo Castillo-Guerra

Addresses: Department of Electrical and Computer Engineering, University of New Brunswick, P.O. Box 4400, 15 Dineen Dr., D36 Head Hall, Fredericton, NB E3B 5A3, Canada

Abstract: This paper evaluates a biomedical speech processing tool to perform acoustic studies of eight types of dysarthria based on a multidimensional approach. The performance of three different assessment methods is evaluated with 20 acoustic features derived perceptually and objectively. Results: An assessment method based on self-organised maps neural networks provides the highest assessment index (88.1%) and the most accurate analysis of relevant features. It also enables the creation of bi-dimensional maps that are of great significance for tracking the evolution of the disorders. The tool provides reference judgements of difficult-to-perceive acoustic dimensions that are relevant to clinical practice, rehabilitation and research.

Keywords: biomedical speech processing; acoustic modelling; pathologic speech; self-organised maps; dysarthria; neural networks; self-organised maps.

DOI: 10.1504/IJBET.2009.027799

International Journal of Biomedical Engineering and Technology, 2009 Vol.2 No.4, pp.352 - 369

Published online: 11 Aug 2009 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article