Abstract: Recognition performances of traditional feature extraction methods degenerate dramatically in noisy environment. In this paper, a new approach called sparse discriminant analysis is developed and its algorithm derived by gradient descent method is given. It combines Linear Discriminant Analysis(LDA) with sparse constraint. Experimental results demonstrate that the method improves the speaker recognition performance in noisy environment.

Key words: speaker recognition, noisy environment, Linear Discriminant Analysis(LDA), sparse representation

摘要： 在噪音环境下，传统说话人识别特征参数的性能会大幅降低。针对该问题，提出一种说话人特征提取方法，将线性判别分析加上稀疏性的约束，并采用梯度下降的优化方法得出相应算法。实验结果表明，该方法不仅在纯净语音的情况下具有较好的鲁棒性，对含有噪音的语音也有较高的识别率。

关键词: 说话人识别, 噪音环境, 线性判别分析, 稀疏表示

CLC Number: 

• TP18