摘要
提出了针对说话人识别的GMM模型训练的新方法。理论推导和实验结果表明,与GMM常用的传统EM算法相比,提出的新算法能够解决训练中会出现奇异阵的问题,并能提高系统识别率。
A new method which aims at the GMM model on speaker recognition is proposed. The results of the experiments show that the new method can solve the problem of odd matrix in training and the recognition rate is improved when compared with traditional EM algorithm.
出处
《电声技术》
北大核心
2004年第12期51-53,共3页
Audio Engineering
基金
教育部科学技术重点项目资助(03082)
关键词
说话人识别
高斯混合模型
EM算法
speaker recognition
Gauss Mixture Model(GMM)
EM algorithm
