论文部分内容阅读
本文利用主分量分析神经网络(PCANN)方法,得到一种新的说话人语音特征。该特征通过对相继几帧语音特征参数组成的特征向量作主分量分析得到.新的特征能有效的引入帧间相关信息,减小冗余度,削弱噪声的影响。实验表明,新特征提高了系统的识别性能。
In this paper, a new speaker’s speech feature is obtained by PCANN method. The feature is obtained by the principal component analysis of the eigenvectors composed of several speech feature parameters in succession.The new feature can effectively introduce the inter-frame correlation information, reduce the redundancy and weaken the influence of noise. Experiments show that the new features improve the recognition performance of the system.