论文部分内容阅读
为提高C-SVM的泛化性能,提出一种基于特征分组的多核融合在线自适应识别算法.此算法首先把特征按照待识别样本集的特性分为若干组,然后根据各组特征的特性采用不同的核函数训练C-SVM模型,并分别把各个模型支持向量间的相似度作为其权重系数,通过自适应样本不断调整权重系数和模型参数,使得C-SVM模型的参数能够随着待识别样本特性的变化而自适应地变化.将此算法应用于非特定人语音情感识别系统,与RBF核、多项式核和Sigmoid核的对比证明了多核融合在线自适应识别算法的优越性,通过与中性语句归一化方法相比证明了本文算法的有效性和稳定性.
In order to improve the generalization performance of C-SVM, a feature-based multi-core fusion on-line adaptive recognition algorithm is proposed, which firstly divides the features into several groups according to the characteristics of the sample set to be identified and then adopts the features of each group Different kernel functions are used to train C-SVM model, and the similarity of each model support vector is taken as its weight coefficient. The weight coefficients and model parameters are adjusted continuously by adaptive samples so that the parameters of C-SVM model can be identified Sample characteristics change adaptively.This algorithm is applied to non-specific speech emotion recognition system, compared with the RBF kernel, polynomial kernel and Sigmoid kernel to prove the superiority of multi-core fusion on-line adaptive recognition algorithm, Compared with the normalized sentences, this paper proves the validity and stability of the proposed algorithm.