【作者】 周静芳； 陈一宁； 刘润生；

【Author】 ZHOU Jingfang, CHEN Yining, LIU Jia, LIU Runsheng(Department of Electronic Engineering, TsinghuaUniversity, Beijing, 100084, China)

【机构】 清华大学电子工程系；

【摘要】 在说话人识别系统中,训练语音与测试玉音的话机类型失配会使说话人识别系统识别性能显著下降。本文在说话人模型合成和话机归一化的基础上提出一种新的信道补偿方法,综合模型和分数两个方面对系统进行信道补偿。NIST1999说话人识别评测库上的实验表明,采用新的信道补偿方法使系统在等错误率和最小检测代价上比仅采用倒谱均值减的基线系统分别降低了39.4%和20.9%,而且优于只采用说话人模型合成或话机归一化补偿的系统。更多还原

【Abstract】 Handset type mismatch between the training and test speech segments causes significant perfonnance degradation in speaker recognition systems. In this paper, a novel channel compensation approach based on Speaker Model Synthesis and Handset Normalization is proposed. Experiments on the 1999 NIST SRE corpus show that the new approach yield better performance than either of the two techniques alone. Around 39.4% relative improvement in EER and 20.9% in DCF is obtained compared with the baseline system with CMS.更多还原