摘要
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。
This article describes the modification of standard vector quantization (VQ) method, and proposes a recognition method, that is, Weight Distortion Measure VQ based on standard deviation and scalar quantity quantization. It calculates the weight values that reflect the contribution of different dimension parameters, and the experiment has proved the recognition efficiency.
出处
《通信技术》
2008年第2期73-75,共3页
Communications Technology
关键词
说话人识别
矢量量化
标准差
标量量化
speaker recognition
vector quantization
standard deviation
scalar quantity quantization