Abstract: In order to improve the recognition rate and suppress channel interference of speaker verification, a novel classification algorithm based on i-vector and sparse representation is proposed. Firstly, weighted linear discriminant analysis is utilized to obtain low dimensional and discriminant feature vector from i-vector that is based on total variability space of GMM-UBM, meanwhile the influence of channel variation is suppressed. And then, the paper establishes the over-complete dictionary on the new i-vector set. The sparse coefficients of the testing speech are computed in this dictionary. Finally, the target speaker is judged by reconstruction error of sparse coefficients for testing speech. The experiment results verify the effectiveness of this proposed algorithm.

Key words: speaker verification, i-vector, sparse representation, weighted linear discriminant analysis, support vector machine

摘要： 为了提高信道变化下说话人确认系统的识别率和鲁棒性，提出一种基于i-向量和加权线性判别分析的稀疏表示分类算法。首先借助于加权线性判别分析的信道补偿和降维性能，消除i-向量中信道干扰信息并降低i-向量的维数；紧接着在i-向量集上构建训练语音样本过完备字典矩阵，采用MAP算法求解测试语音在字典矩阵上的稀疏系数向量，最后利用稀疏系数向量重构测试语音样本，根据重构误差确定目标说话人。仿真实验结果验证了该算法的有效性和可行性。

关键词: 说话人确认, i-向量, 稀疏表示, 加权线性判别分析, 支持向量机