Abstract: For high computational complexity of Support Vector Machine(SVM), this paper uses normalized and Principal Component Analysis(PCA) transform algorithm for voice data pretreatment, combining K-fold cross validation with grid searching in speech recognition. Analysis results show that compared with GA and PSO, this method can effectively improve the SVM parameters optimization efficiency when the recognition rate is basically unchanged.

Key words: Support Vector Machine(SVM), speaker recognition, Principal Components Analysis(PCA), cross validation, normalization

摘要： 针对支持向量机(SVM)计算复杂度高的问题，采用归一化和主元分析变换算法对语音数据进行预处理，并把K倍交叉验证与网络搜索法相结合应用到语音识别中。分析结果表明，与遗传算法和粒子群优化算法相比，该方法可以在识别率基本不变的情况下有效提高 SVM的参数寻优效率。

关键词: 支持向量机, 说话人识别, 主元分析, 交叉验证, 归一化

CLC Number: 

• TP391