论文部分内容阅读
为了提高基于G auss混合模型通用背景模型(GMM-U BM)的说话人辨认系统的运算效率,提出一种基于树的核心挑选算法(TBK S),通过将U BM中的各个G auss分布按组织成树形结构,来减少从中挑选核心分布的运算量。实验结果表明:对1 000个说话人进行辨认,TBK S与现有的基于特征矢量重排序的剪枝算法(ORBP)相结合,将基于GMM-U BM的辨认系统的运算速度提高21.9倍,误识率却只上升不到4%;TBK S和ORBP相结合,可大幅度提高GMM-U BM系统的运算效率,而基本不降低识别率。
In order to improve the computational efficiency of the speaker recognition system based on GMM-U BM, a tree-based core selection algorithm (TBK S) is proposed. By distributing each G auss in U BM by Organize into a tree structure to reduce the amount of computations needed to pick core distributions. The experimental results show that combining TBK S with the existing ORBP (Pruning Algorithm based on eigenvector rearrangement) to identify 1000 speakers increases the computation speed of the recognition system based on GMM-U BM by 21.9 times, The misclassification rate only increased by less than 4%. The combination of TBK S and ORBP can greatly improve the computational efficiency of the GMM-U BM system without reducing the recognition rate.