| 摘要: |
| 传统的基于最大似然估计高斯混合模型参数的方法是一种无导师的学习方法.该方法的主要缺点是学习算法在估计一类模式模型中的参数时只利用了该类模式中的训练样本，而未考虑其他类训练样本分布的影响，因此，这种方法的识别效果往往不够理想.该文提出了利用最大交叉熵估计高斯混合模型参数的方法，这种方法考虑了不同类之间的样本区分性.同时,为了提高获得全局最优解的可能性，文章给出一种利用进化规划求解最优参数的算法，并将这种方法用于非限定文本的话者识别.实验表明，该方法比传统的参数估计方法识别效果要好. |
| 关键词: 最大交叉熵，高斯混合模型，进化规划，话者识别. |
| DOI： |
| 分类号: |
| 基金项目:本文研究得到国家重点自然科学基金和国家863高科技项目基金资助. |
|
| An Approach for Estimating Parameters in Gaussian Mixture Model Based on Maximum Cross Entropy |
|
MA Ji-yong,GAO Wen
|
| Abstract: |
| The traditional approach for estimating parameters in Gaussian mixture models (GMM) based on maximum likelihood is a kind of unsupervised learning method, its shortage is that the parameters in GMM are derived only by the training samples in one class without taking the effect of sample distributions of other classes into account, hence, its recognition is usually not ideal. In this paper, an approach is presented for estimating parameters in GMM based on the maximum cross entropy of different classes, this method takes the discriminations of samples in different classes into account. To increase the possibility of obtaining the global optimal solution, this paper proposes an approach for estimating the optimal parameters in GMM based on evolutionary programming. An experiment has been carried out using the method for the text-independent speaker recognition, the results have shown that the recognition accuracy is higher than that of the traditional approach. The method has also fast convergent speed. |
| Key words: Maximum cross entropy, Gaussian mixture model, evolutionary programming, speaker recognition. |
