Application of Improved AdaBoost.M2-SVM Algorithm in Low SNR Speech Recognition

An algorithm using Geese particle swarm optimization (PSO) algorithm to improve the performance of AdaBoost.M2 with SVM was proposed in this paper. The algorithm first trains some support vector machines as weak classifiers, and then uses AdaBoost.M2 algorithm to embody the weak classifiers into a strong classifie, achieving multi-class classification. this algorithm uses GeesePSO to optimize the weights of SVM weak classifiers, leading to improve lift capacity of strong classifier, making up for the shortcomings of local optimization, and has better performance in global optimization. Experimental result demonstrates that in the low SNR speech recognition the improved AdaBoost.M2-SVM achieved better generalization performance and higher identification rate than SVM.