Feature Selection of Abnormal Speech using Binary Tree

Abstract

Speech problems are faced by Articulatory handicapped people. To understand the problems it is necessary to create database and develop new and improved feature selection technique for reliable, robust and accurate recognition of spoken word. This paper describes novel method of automatic feature selection for abnormal speech which helps to improve performance and accuracy of system. The database created is eleven digits from zero to ten for ten persons each being recorded ten times. Twelve features considered with 183 parameter gives 4096 binary combination of features for eleven digits. The results obtained using MATLAB 12B gives 70% accuracy for digit two for sixteen combinations of feature