基于感知滤波器与统计方法的语音增强新算法  

New Algorithm Based on Perceptual Filter and Statistical Method for Phonetic Speech Enhancement

作　　者：王娜 郑德忠 孙海艳[2] 陈伟[2] 

机构地区：[1]河北省测试计量技术及仪器重点实验室,河北秦皇岛066004 [2]河北建材职业技术学院信息机电系,河北秦皇岛066004

出　　处：《自动化仪表》2010年第12期8-10,14,共4页Process Automation Instrumentation

摘　　要：根据人耳感知特性,提出了一种小波包感知滤波器与统计方法相结合的语音增强新算法。小波包感知滤波器根据人耳Bark域频率感知特性,将含噪语音频带划分成24个频率群,每个频率群内信号进行最小均方误差对数谱幅度(MMSE-LSA)的估计。通过估计各频率群的先验信噪比得到待估计语音与含噪语音的增益方程,从而得到该频率群内的估计语音,最后将所得的分段估计语音重建即得到增强后的语音。实验结果表明,在各种噪声情况下,该方法均优于其他方法。In accordance with the perceptual characteristics of human＇s ear,the new method of combining wavelet packet filter and statistics method for enhancing phonetic speech is proposed.The frequency bands of noisy phonetic speech are divided into 24 of frequency group by the wavelet packet filter based on human ear＇s perceptual characteristics in Bark domain;then the minimum mean square error-log spectral amplitude（MMSE-LSA） estimation is made for each frequency group.The gain equation of the phonetic speech to be estimated and noisy phonetic speech can be obtained through estimating priori SNR of each frequency group,thus estimated phonetic speech in this frequency group is achieved.Finally,the enhanced phonetic speech will be achieved by re-establish segmental estimated phonetic speech.The experiments indicate that under various noises,this method is better than other methods.

关 键 词：语音增强 滤波器 谱幅度估计 信噪比 小波包变换 

分 类 号：TN912.3[电子电信—通信与信息系统]