| 基于端点检测和高斯滤波器组的MFCC说话人识别 |
| |
| 引用本文： | 王萌,王福龙. 基于端点检测和高斯滤波器组的MFCC说话人识别[J]. 计算机系统应用, 2016, 25(10): 218-224 |
| |
| 作者姓名： | 王萌 王福龙 |
| |
| 作者单位： | 广东工业大学 应用数学学院, 广州 510520,广东工业大学 应用数学学院, 广州 510520 |
| |
| 基金项目： | 广东省自然科学基金（S2011040004273） |
| |
| 摘 要： | 在上下文无关的说话人识别应用中，针对传统MFCC特征参数在语音预处理方面不足以及三角滤波器组的缺陷，提出一种改进的MFCC特征参数提取方法.一方面在传统算法上加入端点检测，去除与说话人语音特征无关的静音段；另一方面用高斯滤波器组（Gaussian shaped filters GF）代替三角滤波器组进行频率到Mel频率的转换，提高识别准确率.说话人识别模型使用流行的高斯混合模型（GMM）.实验结果显示，高斯滤波器组的引入相比于传统三角滤波器组识别率有4.45%的提升，本文改进后的MFCC特征参数相比于传统方法识别率也提升了6.43%，能更好的代表说话人的语音特征.
|
| 关 键 词： | MFCC特征参数 端点检测 高斯滤波器组(GF) 高斯混合模型(GMM) 说话人识别 |
| 收稿时间： | 2016-02-17 |
| 修稿时间： | 2016-04-05 |
Speaker Identification with Improved MFCC Based on Endpoint Detection and Gaussian Shaped Filters |
| |
| WANG Meng and WANG Fu-Long. Speaker Identification with Improved MFCC Based on Endpoint Detection and Gaussian Shaped Filters[J]. Computer Systems& Applications, 2016, 25(10): 218-224 |
| |
| Authors: | WANG Meng and WANG Fu-Long |
| |
| Affiliation: | Applied Mathematics, Guangdong University of Technology, Guangzhou 510520, China and Applied Mathematics, Guangdong University of Technology, Guangzhou 510520, China |
| |
| Abstract: | In the application of text-independent speaker recognition, this paper puts forward an improved feature extraction of MFCC parameters to supply the inefficient traditional MFCC. Endpoint detection is added in traditional algorithm to remove silence parts. Gaussian shaped filters are used to replace triangular filter banks to improve the accuracy of speaker identification. Gauss mixed model is for speaker recognition. Experiments show that Gaussian shaped filters gain 9.63% performance improvement while proposed MFCC can significantly improve recognition rate by 11.07%. The result indicates that the new method is an effective feature extraction algorithm. |
| |
| Keywords: | MFCC parameters endpoint detection Gaussian shaped filters(GF) Gauss mixed model(GMM) speaker identification |
|
| 点击此处可从《计算机系统应用》浏览原始摘要信息 |
|
点击此处可从《计算机系统应用》下载全文 |
