以线性预测系数为特征通过高斯混合模型的迭代算法对训练样本的初始 k 均值聚类结果进行优化,得到语音组成单位的表示.以语音组成单位的模式匹配为基础,提出一种文本无关说话人确认的方法——均值法,以及一种文本无关说话人辨认方法.实验结果表明,即使在短时语音下本文方法都能取得较好效果. The initial k-means clustering result of the training samples is optimized by the iterative algorithm of Gaussian mixture model, and the representation of the speech component units is obtained based on the linear prediction coefficient. Based on the pattern matching of the speech component units, a text-independent speaker The method of confirming - the mean value method, and a text-independent speaker recognition method.The experimental results show that the method in this paper can achieve good results even under short-term speech.