Improvements on isolated word recognition using FLDA [FLDA kullanarak ayrik kelime tanimada yapilan iyileştirmeler]

Abstract

The purpose of this study is to increase the recognition rates of the isolated word using improved Fisher's Linear Discriminant Analysis (FLDA). Therefore, at first, unique subspace for each class is constructed by defining between class covariance matrix in two ways and the recognition performance is investigated using this subspace. Then, separate subspaces for each class are constructed by defining between class covariance matrix in two different ways and the results obtained from these subspaces are combined to increase recognition performance. These results are slightly greater than the results given in literature for the FLDA method