| 基于说话人识别的GMM/GA算法 |
| |
| 引用本文： | 邱政权,尹俊勋.基于说话人识别的GMM/GA算法[J].电声技术,2006(8):43-45,49. |
| |
| 作者姓名： | 邱政权 尹俊勋 |
| |
| 作者单位： | 华南理工大学,电子与信息学院,广东,广州,510640 |
| |
| 摘 要： | 在实时平台上,高斯混合模型(GMM)具有计算有效性和易于实现的优点。最大似然规则中,模型参数不断更新,但由于爬山特征,任意的原始模型参数估计通常将导致局部最优;遗传算法(GA)适于求解复杂组合优化问题及非线性函数优化。提出了基于说话人识别的可以解决GMM局部最优问题的GMM/GA新算法,实验结果表明,提出的GMM/GA新算法比纯粹的GMM算法能获得更优的效果。
|
| 关 键 词： | 高斯混合模型 最大期望算法 遗传算法 适应值 |
| 文章编号： | 1002-8684（2006）08-0043-03 |
| 收稿时间： | 2006-05-08 |
| 修稿时间： | 2006-05-08 |
GMM/GA Algorithm Based on Speaker Recognition |
| |
| QIU Zheng-quan,YIN Jun-xun.GMM/GA Algorithm Based on Speaker Recognition[J].Audio Engineering,2006(8):43-45,49. |
| |
| Authors: | QIU Zheng-quan YIN Jun-xun |
| |
| Affiliation: | College of Electronic and Information Engineering, South China University of Technology, Guangzhou 510640, China |
| |
| Abstract: | GMM(Gaussian Mixed Medel)has the advantages of calculating effectiveness and being easy to realiz-ed, especially in the real time condition. Based on ML(Maximum Likelihood)rule, the model parameter can be updated ceaselessly but due to mountain climbing character, the estimation of arbitrary original model parameter can lead local optimization. GA(Genetic Algorithm)is fit for solving complicated combination optimization and no-nlinear function optimization. GMM/GA algorithm based on speaker recognition which can solve the problem of in GMM local optimization is proposed. The experimental results show that the GMM/GA algorithm is better than GMM algorithm. |
| |
| Keywords: | GMM expectation-maximization algorithm genetic algorithm fitness value |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
