改进的AdaBoost.M2-SVM在低信噪比语音识别中的应用  

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

作　　者：刘红芬[1] 刘晓峰[2] 张雪英[1] 黄丽霞[1] 王子中[1] 

机构地区：[1]太原理工大学信息工程学院,山西太原030024 [2]太原理工大学理学院数学系,山西太原030024

出　　处：《微电子学与计算机》2015年第2期88-91,共4页Microelectronics & Computer

基　　金：国家自然科学基金项目(61371193);山西省青年科技研究基金项目(2013021016-1);山西省2012年留学回国人员科技活动择优资助(晋人社厅函[2013]68号)

摘　　要：提出了基于雁群启示的粒子群优化算法改进的AdaBoost.M2-SVM算法.首先训练多个支持向量机作为弱分类器,用AdaBoost.M2算法将多个弱分类器集成为最终的强分类器,实现多类分类;采用GeesePSO算法对AdaBoost.M2算法计算出的权值进行优化得到一组最优的权值,提高最终强分类器的提升能力.实验结果表明,在低信噪比语音识别中,与SVM相比,改进的AdaBoost.M2-SVM表现出更好的泛化能力,提高了识别准确率.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.

关 键 词：AdaBoost.M2 支持向量机 权值 GeesePSO 低信噪比 

分 类 号：TN912.34[电子电信—通信与信息系统]