中文摘要：

探讨了说话人认证特征级融合策略的可行性.根据关系度量融合框架,构建该策略认证系统.通过与传统融合、单模态算法比较,本算法性能优于以上算法.为进一步分析特征级融合算法优于现有融合算法的原因,本文利用最大Kullback-Leibler距离计算融合算法融合信息量.该距离弥补了传统Kullback-Leibler距离不具有对称性的缺憾,更加精准地获取信息量.分析结果验证了本算法实验结论,说明特征级融合可获取比现有匹配分数级融合更多的信息量,从而取得更优精度.

英文摘要：

This paper investigates the possibility of the feature level fusion based on speaker verification. According to the robustness and availability of the relation measurement fusion framework,the feature level fusion based on speaker verification is established. Through comparison of performances among the feature level fusion,traditional matching-score level fusion,and unimodal algorithms,the experimental results show that the proposed method is the best. To further analyze its correctness,this paper introduces the maximum Kullback-Leibler distance of the aspect of information theory to measure the information content. This distance overcomes the shortcoming of the asymmetry by traditional Kullback-Leibler distance and improves the precision of the information content computation. And the computational results verify the effectiveness of our algorithm,indicating that the feature level fusion can hold more discriminative information than the existing matching-score level fusion to yield a better performance.