On the optimization of multiclass support vector machines dedicated to speech recognition

Abstract

We present in this paper an interesting approach to enhance the performance of multi-classification using Genetic Algorithm. Two systems for an instance selection and feature selection are respectively introduced. We combined Genetic Algorithm with multiclass Support Vector Machines in order to reduce the learning set. The goal is to simplify the learning process and to improve the generalization. The results obtained on speech corpus show encouraging improvements in terms of processing time and classification accuracies. © 2012 Springer-Verlag.