说话人确认中基于无监督聚类的得分规整  

Unsupervised Clustering Score Normalization in Speaker Verification

作　　者：古斌 郭武[1] Gu Bin;Guo Wu(University of Science and Technology of China,National Engineering Laboratory for Speech and Language Information Processing,Hefei,230026,China)

机构地区：[1]中国科学技术大学语音及语言信息处理国家工程实验室

出　　处：《数据采集与处理》2019年第5期837-843,共7页Journal of Data Acquisition and Processing

摘　　要：在说话人确认任务中,得分规整可有效调整测试得分分布,使每个说话人的得分分布接近同一分布,从而提升系统整体性能。直接从开发集中获得针对待识别目标说话人的大量冒认者得分,利用无监督聚类手段对这些得分进行筛选,并采用混合高斯模型来拟合得分分布,挑选均值最大的高斯单元作为得分规整的参数并将其应用于说话人的得分规整。在NIST SRE 2016测试集上的测试结果表明,相对于其他得分规整算法,采用无监督聚类得分规整的方法可有效提升系统性能。In the speaker verification(SV)task,score normalization can improve the system performance by adjusting the score distribution of each speaker to a similar distribution.Here,a large number of imposter scores for the target speakers are obtained from the development set firstly,then these scores are clustered by unsupervised clustering algorithm and the Gaussian mixture models(GMM)are used to fit the score distribution.The mean and standard deviation of Gaussian component with maximum mean value are used in the SV score normalization method.Experiments are conducted on the NIST SRE 2016 test set and results show that compared with the conventional score normalization methods,the proposed method can effectively improve the system performance.

关 键 词：说话人确认 得分规整 无监督聚类 

分 类 号：TN912[电子电信—通信与信息系统]