论文部分内容阅读
文中研究了在基于矢量量化的说话人识别方法中采用加权的失真测度对识别率的影响。在采用加权欧氏距离失真测度时,分别对采用传统的标准差来描述的类内聚合程度,引入标量量化的失真测度来描述类间离散程度,并产生相应的权值对参数通道进行预加重来进行识别,通过试验,通过实验验证了基于标准差(SD-WDMVQ)、基于标量量化的(SQQ-WDMVQ)和综合考虑类内类间距离的(WDMVQ)可以得到更高的识别率。
In this paper, we study the influence of weighted distortion measure on speaker recognition based on vector quantization. When using the weighted Euclidean distance distortion measure, we introduce the degree of in-class polymerization, which is described by the traditional standard deviation, and introduce the scalar quantization distortion measure to describe the degree of class discretization and generate the corresponding weight to pre-emphasize the parameter channel Through experiments, it is verified through experiment that a higher recognition rate can be obtained based on standard deviation (SD-WDMVQ), scalar quantization (SQQ-WDMVQ) and considering the distance between classes (WDMVQ).