一种改进的高斯混合模型算法

An Improved Algorithm of GMM

建立声学模型是说话人识别技术的重要环节,一种好的建模方法对说话人识别系统的识别率具有极其重大的影响。本文介绍了一种改进的高斯混合模型算法———将聚类算法与传统高斯混合模型结合起来的建模方法,并对此种建模方法得出的识别效果与传统的高斯混合模型进行了比较。从对比结果可以看出,基于聚类的高斯混合模型的说话人识别相对于传统的高斯混合模型在识别率上有所提高。

Characteristic modeling is an important link in technology of speaker identification, a nice modeling method impact on the performance of speaker identification system. This article introduces an improved algorithm of GMM which combines classical GMM with clustering algorithm, and compares their performance. From the results, the article concludes that GMM based on clustering algorithm has higher performance than classic GMM algorithm.