数字助听器中单通道语音增强算法的研究  

Research on Single-Channel Speech Enhancement Algorithm for Hearing Aids

作　　者：曹旭来 张玲华[1] 林志敏[1] 郑宝玉[1] 

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《信号处理》2014年第4期405-412,共8页Journal of Signal Processing

基　　金：江苏省普通高校研究生科研创新计划(CXZZ12_0474)资助课题;江苏省高校自然科学研究重大项目(13KJA510003);江苏高校优势学科建设工程资助项目(PAPD)

摘　　要：传统多通道补偿算法忽略保护语音特征,容易造成语音结构变形和识别率低等问题。为了解决上述问题,本文提出了一种基于多分辨率小波的单通道语音增强算法。利用多分辨率小波对信号进行分解与重构,提取出语音信号的频谱包络,得到特征点初期信息并依此计算补偿增益,再利用插值算法计算出整个频谱的增益并对其进行响度补偿。仿真实验以及主观性能测试的各项结果均表明该算法能够在对语音进行补偿的同时有效地保护语音特征,提高言语识别率,达到比较理想的效果。At present,multi-channel loudness compensation algorithm is widely used for the hearing aids.However,the conventional multi-channel algorithm ignores the protection of speech feature,which will lead to the loss of recognition rate and deformation of speech structure.To solve this problem,a new single-channel speech enhancement algorithm based on multi-resolution wavelet is proposed in this paper.Multi-resolution wavelet is involved to decompose and reconstruct the signal＇ s spectrum,which can help to extract the envelope of signal spectrum.Then,calculate the required gain of each feature point based on the input-output curve.After that,the gain of whole signal spectrum is obtained by using the interpolation algorithm.The proposed algorithm shows fairly good performance in both stimulation and subjective test.The experimert and simulation indicate that the proposed algorithm has high practical value.It can not only compensate the loudness of voice,but also effectively protect the speech feature,which improves the speech recognition rate.

关 键 词：语音特征 多分辨率小波 插值算法 响度补偿 

分 类 号：TN912.3[电子电信—通信与信息系统]