ISCA Archive ECST 1987
ISCA Archive ECST 1987

Using condensed nearest neighbor rule for speaker independent word recognition

N. Yalabik, A. Mansur

In this study, a combination of Editing and modified Condensed Nearest Neighbor (CNN) rule is used for classification of speaker independent isolated words. The approach is compared to clustering techniques used for template selection. The implementation is done with an 18-word Turkish vocabulary where Linear Preditive Coding parameters and Dynamic Time Warping are used for classification. The results show that Editing and CNN together can be used for template selection as a better alternative to clustering.