【作者】 郭武； 戴礼荣； 王仁华；

【Author】 Guo Wu, Dai Li-Rong, Wang Ren-Hua iFly Speech Lab, Department of Electronic Engineering and Information Science, University of Science and Technology of China, Heifei, Anhui 230027

【机构】 中国科技大学电子工程与信息科学系科大讯飞语音实验室；

【摘要】 在与文本无关的说话人识别研究中,因子分析是减少信道影响的最有效的方法之一,但是因子分析复杂的运算使得因子分析在实时中很难得到应用。本文提出采用主成分分析的方法简化因子分析的过程。首先在模型域中估计出信道因子所在的空间,然后通过映射的方法在特征参数域中减去信道因子的影响。在2006年数据库上,采用本文推荐方法的系统相对基线系统在等错误率上有24%的降低。更多还原

【Abstract】 In the speaker-independent speaker verification research, factor analysis is one of the most effective methods to reduce the bias by the channel, but the computation load of factor analysis prevents its application in real-time cases. In this paper, the traditional principal component analysis(PCA) method is proposed to reduce the computation of factor analysis. The subspace of the channel is estimated by PCA, then the bias of the channel is subtracted from the speech feature. In the 2006 SRE corpus, the equal error rate(EER) of the proposed system can reduce by 24% against the baseline GMM system.更多还原