基于话者聚类的多系统输出评分融合话者确认  

Speaker clustering based likelihood scores fusion robust speaker verification

作　　者：上官葳[1] 戴蓓蒨[1] 

机构地区：[1]中国科学技术大学电子科学与技术系,安徽合肥230027

出　　处：《兰州大学学报（自然科学版）》2008年第3期81-86,共6页Journal of Lanzhou University(Natural Sciences)

基　　金：国家自然科学基金(60272039);多媒体计算和通信MOE-微软联合实验室科研基金(05071810)资助.

摘　　要：针对短电话语音条件下文本无关说话人确认问题中语音数据不充分和电话信道失配问题,利用说话人特征参数之间的互补性,提出了一种对多子系统输出评分采用多个网络的融合策略,提高了话者确认系统的整体性能.实验结果表明:与传统的所有话者共享一个输出评分融合网络相比,应用多个融合网络,在采用KLD和GLR模型相似性测度时,系统等误识率分别下降了12.7%和10.8%.An approach of multi-fusion networks for the likelihood scores fusion of multi-subsystems in the text-independent speaker verification system with limited speech data （e.g. 10 s training data for each speaker） in various telephone microphone channels was presented. The subsystems complement each other with the fusion network and enhance the system performance. With the proposed method, experimental results on NIST SRE 05＇s database show a relative equal error rate reduction of 12.7% and 10.8% for KLD and GLR, which are two similarity measurement between speaker models, respectively, with respect to an all-speaker-shared fusion network.

关 键 词：说话人确认 话者聚类 输出评分融合 多网络融合 

分 类 号：TN912.34[电子电信—通信与信息系统]