基于改进的Kalman滤波的语音增强算法  

Speech enhancement based on improved Kalman filtering method

作　　者：余华[1,2] 陈国明[2] 赵力[2] 邹采荣[2] 

机构地区：[1]南京信息职业技术学院,南京210023 [2]东南大学信息科学与工程学院,南京210096

出　　处：《声学技术》2009年第6期763-767,共5页Technical Acoustics

基　　金：国家自然科学基金项目(60472058;60975017);江苏省自然科学基金项目(BK2008291)

摘　　要：传统的kalman滤波方法在推导过程中假定观测噪声为白噪声。通常对于有色噪声需要用白噪声激励的方法予以模拟,并且需要以牺牲运算量作为代价。本文提出了一种改进的基于kalman滤波的语音增强算法,可以处理白噪声和有色噪声情况,不需要增加计算量,仿真结果表明了该算法对有色噪声的语音增强性能要优于基于传统kalman滤波方法。In this paper a single channel speech enhancement method for noisy speech signals is proposed, which is based on improved Kalman filter. Due to the high computation cost and the need of accurate estimation of AR model of speech in colored noise, a subspace based speech enhancement method is introduced. The proposed method can make tradeoff between residual noise and speech distortion. Experimental results demonstrate the improved performance of the proposed algorithms.

关 键 词：语音增强 卡尔曼滤波 有色噪声 

分 类 号：TN912.35[电子电信—通信与信息系统]