一种噪声谱快速跟踪的语音增强方法  

A speech enhancement method based on quickly tracking the noise spectrum

作　　者：陈文钢[1] 田岚[2] 姜晓庆[2] 孙英明[3] 

机构地区：[1]山东理工大学电气与电子工程学院,山东淄博255049 [2]山东大学信息科学与工程学院,山东济南250100 [3]济南大学

出　　处：《山东大学学报（工学版）》2006年第4期26-28,43,共4页Journal of Shandong University（Engineering Science）

摘　　要：提出了基于多窗谱法和小波阈值法对带噪语音的短时功率谱进行估计的语音增强改进方法.针对软、硬阈值法自身的缺点,采用了一种改进方法,即模平方处理方法.噪声估计过程中通过跟踪带噪语音帧来更新噪声估计.仿真结果表明,本文提出的方法能够快速跟踪突变噪声,具有较强的降噪能力.把此算法综合到语音增强中时,其增强性能要好于STSA-MMSE算法.A new speech enhancement method based on multitaper spectrum method and wavelet thresholding is proposed so that a noisy speech power spectrum is obtained. As for the faults of the hard-thresholding and soft-thresholding, an improved model of threshold （the modulus squared model） is presented. During the noise estimation, the estimation of its spectrum is updated by tracking the speech-absent frames. The experimental results show that this method presented here can quickly track the changes of the speech and has better denoise function. When combining the method with speech enhancement, the enhancement result is better than that of STSA algorithm.

关 键 词：语音增强 多窗谱法 小波阈值 噪声估计 

分 类 号：TN912.35[电子电信—通信与信息系统]