论文部分内容阅读
针对传统的说话人分割聚类系统中,由于聚类时话者信息不足而影响切分准确度的问题,本文提出了一种基于进化隐马尔科夫模型和交叉对数似然比距离测度的多层次说话人分割聚类算法,在传统的话者分割聚类算法的基础上引入了重分割和重聚类的机制,以及基于距离测度和贝叶斯信息准则的分层聚类算法,有效的解决了传统方法中切分准确度受到话者信息制约的问题.在美国国家标准技术署(NIST)2003 Spring RT数据库上的实验结果表明,本文提出的算法比传统算法系统性能相对提高了41%.
Aiming at the problem of segmentation accuracy in traditional speaker segmentation clustering system because of the lack of information of speaker in clustering, this paper proposes an evolutionary Hidden Markov Model and cross-logarithmic likelihood ratio distance measure Multi-level speaker segmentation clustering algorithm, based on the traditional speaker segmentation clustering algorithm, introduces the mechanism of re-segmentation and re-clustering, as well as hierarchical clustering algorithm based on distance measure and Bayesian information criterion, which is effective Which solves the problem that the segmentation accuracy is limited by the speaker information in the traditional method.Experimental results on 2003 Spring RT database of the National Institute of Standards and Technology show that the performance of the proposed algorithm is 41% .