基于改进MFCC的说话人特征参数提取算法  

An Efficient Speaker Feature Parameter Extraction Method Based on Improved of MFCC

作　　者：高铭 孙仁诚 GAO Ming;SUN Ren-cheng(School of Computer Science and Technology,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学计算机科学技术学院,青岛266071

出　　处：《青岛大学学报（自然科学版）》2019年第1期61-65,73,共6页Journal of Qingdao University(Natural Science Edition)

基　　金：国家自然科学基金(批准号:41476101)资助

摘　　要：在说话人识别系统中,传统梅尔倒频谱系数(MFCC)所提取特征不能够很好的反映说话人动态特征,尤其在噪声环境中,识别率较低,鲁棒性不足。针对以上问题,提出一种基于改进梅尔倒频谱系数(MFCC)的方法,通过多窗谱估计和一阶、二阶差分的方法提升识别性能。实验结果证明,在纯净语音和添加信噪的情况下,改进后方法的识别准确率都有所提升。当训练集为纯净语音,只为测试集添加噪声时,实验结果依然有较高的准确率。In speaker recognition system,the traditional Mel frequency cepstral coefficients(MFCC)can't reflect speaker dynamic characteristics very well.Especially in the noisy environment,the recognition capability and robustness are insufficient.To solve the above issues,an algorithm based on improved Mel frequency cepstral coefficients(MFCC)is proposed.One way that using multiple windows spectral estimation,first-order differential and second-order differential parameters to enhance its recognition capability.Experimental results show that the recognition accuracy of the proposed method has significantly improved in case of pure speech and different signal noise ratios(SNR).Furthermore,when the training set is pure speech and only noise is added to the test set only,the experimental results of the proposed method still have high accuracy.

关 键 词：MFCC 多窗谱估计 一阶二阶差分 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN912.34[自动化与计算机技术—计算机科学与技术]