首页 | 本学科首页   官方微博 | 高级检索  
     

一种基于MFCC与韵律特征的说话人确认方法
引用本文:骆启帆,章坚武,吴震东. 一种基于MFCC与韵律特征的说话人确认方法[J]. 杭州电子科技大学学报, 2013, 0(5): 134-137
作者姓名:骆启帆  章坚武  吴震东
作者单位:杭州电子科技大学通信工程学院,浙江杭州,310018
摘    要:梅尔倒谱系数是一种常用于说话人识别的特征参数,韵律特征是一种描述人的声门特性的参数。为融合MFCC与韵律特征,以图优化说话人确认系统性能,该文采用二次判决的方法来处理这两个特征;参与第二次判决的语音则由通过大量实验制定的判决空间来确定。实验结果表明,采用二次判决时,系统等错误率从仅使用MFCC时的5.56%的下降至4.37%。

关 键 词:说话人确认  梅尔倒谱系数  韵律特征

A Speaker Verification Method Based on MFCC and Prosodic Features
LUO Qi-fan , ZHANG Jian-wu , WU Zhen-dong. A Speaker Verification Method Based on MFCC and Prosodic Features[J]. Journal of Hangzhou Dianzi University, 2013, 0(5): 134-137
Authors:LUO Qi-fan    ZHANG Jian-wu    WU Zhen-dong
Affiliation:(School of Communication Engineering, Hangzhou Dianzi University, Hangzhou Zhejiang 310018, China )
Abstract:Mel cepstral coefficients (MFCC) is a common feature of speaker recognition, it describes speakers' channel from the view of how do human perceive the sound. Prosodic features is a feature which describes the characteristics of the human glottis, it is used for speaker recognition, too. For fusing MFCC and prosodic features (logarithmic energy and logarithmic fundamental frequency), to optimize the speaker verification system performance, using the method of quadratic judgment to deal with these two features; the voice take part in secondary judgment was decided by threshold which was determined by a large number of experi- ments. Experimental results show that by using 5.56% of only using MFCC to 4.37%.
Keywords:speaker verification  Mel frequency cepstral coefficient  prosodic features
本文献已被 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号