Computer Engineering and Applications ›› 2015, Vol. 51 ›› Issue (15): 206-210.

Previous Articles     Next Articles

Study of speaker gender identification based on modified Citation-KNN

ZHU Junmei1, GU Mingliang1,2, ZHANG Shixing2, JIA Jingjing1   

  1. 1.School of Linguistic Science, Jiangsu Normal University, Xuzhou, Jiangsu 221116, China
    2.School of Physics & Electronic Engineering, Jiangsu Normal University, Xuzhou, Jiangsu 221116, China
  • Online:2015-08-01 Published:2015-08-14

基于改进Citation-KNN算法的性别识别研究

朱俊梅1,顾明亮1,2,张世形2,贾晶晶1   

  1. 1.江苏师范大学 语言科学学院,江苏 徐州 221116
    2.江苏师范大学 物理与电子工程学院,江苏 徐州 221116

Abstract: To simplify the method of model training and improve efficiency of gender identification system, an algorithm of speaker gender identification based on modified Citation-KNN algorithm and a new speech Multi-Instance(MI) bag generating method are introduced. Continuous speech is segmented and sub-paragraph speech is modeled with a Gaussian Mixed Model(GMM). The generated GMM is treated as an MI bag and the components of the GMM are the instances of the corresponding bag. Modified Hausdorff distance is used to measure the distance between two bags. Experimental results show that this method can identify speakers’ gender effectively and is superior to traditional algorithms in the correct recognition rate.

Key words: gender identification, modified Citation-K-Nearest Neighbor(KNN) algorithm, Gaussian Mixed Model(GMM), modified Hausdorff distance

摘要: 为了简化系统模型训练方法,提高性别识别系统的整体效率,提出了一种基于改进Citation-KNN算法的说话人性别识别方法。该方法将连续语音切分,训练每段语音的高斯混合模型(Gaussian Mixture Model,GMM)作为多示例包,其所有混合元为相应包中示例;采用改进的Hausdorff距离作为包与包之间的距离测度,通过Citation-KNN算法进行性别识别。该方法以多示例包间距离为分类依据,简化了系统训练,且识别率优于一些传统算法。

关键词: 性别识别, 改进Citation-K最近邻(KNN)算法, 高斯混合模型, 改进Hausdorff距离