AN APPLICATION OF MINIMUM CLASSIFICATION ERROR TO FEATURE SPACE TRANSFORMATIONS FOR SPEECH RECOGNITION

The use of signal transformations is a necessary step for feature extr action in pattern recognition systems. These transformations should ta ke into account the main goal of pattern recognition: the error-rate m inimization, In this paper we propose a new method to obtain feature s pace transformations based on the Minimum Classification Error criteri on. The goal of these transformations is to obtain a new representatio n space where the Euclidean distance is optimal for classification. Th e proposed method is tested on a speech recognition system using diffe rent types of Hidden Markov Models, The comparison with standard pre-p rocessing techniques shows that our method provides an error-rate redu ction in all the performed experiments.