基于DCT 与维纳滤波的单通道语音增强算法DCT and Wiener filter.PDF

第 27 卷第 10 期 通 信 学 报 Vol.27 No.10 2006 年 10 月 Journal on Communications October 2006 基于 DCT 与维纳滤波的单通道语音增强算法 欧世峰 ，赵晓晖，顾海军 （吉林大学 通信工程学院信息科学实验室，吉林 长春 130012 ） 摘 要：针对复杂噪声背景下的语音增强 问题 ，基于离散余弦变换(DCT)和 纳滤波提 出了一种新的单通道语音 增强算法 。该算法不依赖任何语音信号模型且无需对噪声的统计特性进行先验假定，它利用 DCT 域中连续时刻 语音信号分量间的相关特性结合最小均方误差算法实现纯净语音分量的最优估计，弥补 了一般算法仅依赖单帧带 噪语音对语音分量估计得不足 。多种噪声背景下的仿真结果表 明，该算法在主观和客观测试 中都具有 良好的语音 增强效果。 关键词：语音增强; 语音分量估计 ；离散余弦变换； 纳滤波；连续分量相关 中图分类号：TN912 文献标识码 ：A 文章编号：1000-436X(2006) 10-0086-08 DCT and Wiener filter based on approach for speech enhancement using a single microphone OU Shi-feng, ZHAO Xiao-hui, GU Hai-jun (Laboratory of Information cience, College of Communication Engineering, Jilin University, Changchun 130012, China) Abstract: Considering the enhance ent of speech corrupted by co plicated additive noise, a novel algorith was proposed based on discrete cosine transfor (DCT) and Wiener filter using a single icrophone. This algorith did not rely on any speech signal odel and could efficiently attain the opti al esti ation of clean speech co ponents using successive noisy speech co ponents and ini u ean square error esti ation in DCT do ain. Further ore, it could overco e the disad- vantage of independent assu ptions in classic ethods for speech co ponents. Si ulation results de onstrate that the pro- posed algorith possesses good perfor ance both in obj ective and subj ective tests wit