基于多脉冲激励和卡尔曼滤波的语音增强算法  

Speech Enhancement Method Based on Kalman Filtering and Multipluse Excitation

作　　者：柳林[1] 李辉[1] 戴蓓蒨[1] 陆伟[1] 

机构地区：[1]中国科学技术大学多媒体计算与通信教育部微软重点实验室,合肥230027

出　　处：《数据采集与处理》2009年第1期100-104,共5页Journal of Data Acquisition and Processing

摘　　要：给出了一种基于卡尔曼滤波的语音增强算法,并对语音信号的清浊音加以区分。通过在浊音段提取多脉冲激励信号加入语音的状态方程,有效地重建语音的高频谐波。采用对数谱测度LSD(Log-spectral distortion)和语音感知质量评价PESQ(Perceptual evaluation of speech quality)评分对语音增强的效果进行评测。实验表明,算法提高了增强语音的质量,对白噪声和汽车噪声均有满意的效果。A speech enhancement algorithm based on Kalman filtering （KF） is described, a speech model for descrbing the voiced and unvoiced speech is proposed. By extracting the multipulse excitations during voiced speech, and adding the excitations into the process equation the harmonic of high-frequency area can be reconstructed. The quality of the enhanced speech is evaluated by LDS（Log-speetral distortion） and PESQ （Perceptual evaluation of speech quality） scores. Experimental results show that the proposed algorithm improves the output speech quality compared with conventional KF algorithm and subband spectral subtraction.

关 键 词：语音增强 卡尔曼滤波 多脉冲激励 

分 类 号：TN912.35[电子电信—通信与信息系统]