View article

The present invention relates to a pattern recognition system which uses data fusion to combine data from a plurality of extracted features and a plurality of classi? ers. Speaker patterns can be accurately veri? ed with the combination of discriminant based and distortion based classi? ers. A novel approach using a training set of a “leave one out” data can be used for training the system with a reduced data set. Extracted features can be improved with a pole? ltered method for reducing channel effects and an af? ne transfor mation for improving the correlation between training and testing data.