摘要
在文本无关的说话人识别系统中,采用序列核作为核函数的支持向量机系统已经得到了广泛的应用。本文首先归纳出构造序列核的通用框架,并在此框架之上对高斯序列核和广义线性区分核两种目前运用比较成熟的序列核系统进行分析比较,说明特征空间中不同属性和层次的语音特征是如何通过不同的序列核来表征的。在NIST2006评测数据集中,识别率较传统的混合高斯模型-通用背景模型有显著提高。
In the text- independent speaker recognition system, support vector machine(SVM) equipped with sequence kernel has been widely used. In this paper, a generic structure conceiving sequence kernel has been encapsulated and in the structure we make a analytical comparison between two well used sequence kernel system--GMM supervector kerne(GSK) and generalized linear discrimi- nant sequence(GLDS) showing how different attribute and levels of cues conveyed by speech utterances are being characterized within different sequence kernel. In the NIST 2006 SRE corpus, recognition rate improves significantly compared with the traditional GMM and universal background models( GMM -UBM) system.
出处
《南阳理工学院学报》
2012年第4期14-17,共4页
Journal of Nanyang Institute of Technology
关键词
序列核
高斯超向量核
广义线性区分核
说话人识别
sequence kernel
GMM supervector kernel
generalized linear discriminant sequence kernel
speaker recognition