An Automatic Data-driven Technique for Selecting Background Dataset in GMM-SVM Speaker Verification System

摘要：

In this paper, we propose an automatic data-driven technique for selecting proper background dataset. By the technique, impostor confidence(IC) is proposed as a metric and more discriminative background dataset is automatically chose by impostor confidence(IC) to train more discriminative model. Experiment results on NIST 2008 SRE corpus in GMM-SVM speaker verification system show that the proposed approach obtains better performance. Relative decline in mincost of 8.9％ in female and 4.6％ in male is obtained. with female and male combined, 5.4％ relative decline in mincost is obtained over Heuristically selected background dataset.

作者: Jinchao Yang Haipeng Wang Jianping Zhang Yonghong Yan

作者单位: ThinkIT Speech Lab, Institute of Acoustics, Chinese Academy of Sciences, Beijing, P.R.China

会议类型: 国际会议

会议名称: 第十届中国虚拟现实年会

会议地点: 上海

会议语种:英文

页码: 85-89

在线出版日期: 2010-10-20（万方平台首次上网日期，不代表论文的发表时间）