基于有效特征集选择的说话人识别

GMM and MMI Merged Speaker Identification

常用的文岙无关的说话人识别系统中，高斯混合模型是一种常用的有效模型，其在常用的说话人识别系统中具有很广泛的应用，但是典型的GMM模型的识别时间较长，其识别性能尚待提高。为了进一步提高说话人识别系统的性能，练合语音数据的分布特性和log10（·）函数曲线的分布特性提出了一种有效特征集的选择方法，实验结果证明1了提出的选择算法的有效性。

In speaker identificatoin system, GMM is a frequently-sed modal and its application is abroad,but typical GMM modal need much speech used tobe recognized,so we can see that its performance need to be improved..So to improve the performance of speaker recognition system, Condsidering distribution character of speech data and loglO （.） function,an effective feature set extract method is put forward and its effectiveness is proved by experiments.