A general framework of feature extraction: application to speaker recognition

Extracting a good feature set is important to pattern recognition. A new formulation of integrating the feature extraction into the model training is proposed. The intraframe weighting, the interframe weighting and the feature reduction schemes can be obtained from this new formulation. According to the dependence of the class model parameters, three types of feature extraction are derived. Some experiments for the speaker recognition application are given to show the effectiveness of the new proposed feature extraction method