SDBM-Based Speaker Recognition for Speaking Style Variations
There are many factors corresponding to performance degradation of an actual speaker recognition system. Mismatch in speaking style of a target speaker during training and testing is an important one. When a client enrolls in a system, it is natural for him/her to speak in a spontaneous way. However, it is difficult to maintain the same speaking style throughout test phases. In view of this situation, this paper, based on a database with multiple speaking styles, proposes the concept of Speaking-style-Dependent Background Model (SDBM). The SDBM-based system is presented to train speaking style featured speaker models aiming to alleviate the speaking style mismatch between training and testing. Experimental results show that EER can be reduced by 35.40％.
Linlin Wang Mingxing Xu
Department of Computer Science and Technology, Tsinghua University State Key Laboratory of Intellig Department of Computer Science and Technology, Tsinghua University State Key Laboratory of Intellig
国际会议
2011亚太信号与信息处理协会年度峰会(APSIPAASC 2011)
西安
英文
1-4
2011-10-18（万方平台首次上网日期，不代表论文的发表时间）