基于计算听觉场景分析的语音增强改进算法  

Improved Speech Enhancement Based on Computational Auditory Scene Analysis

作　　者：王雨[1] 林家骏[1] 袁文浩[1] 

机构地区：[1]华东理工大学信息科学与工程学院,上海200237

出　　处：《华东理工大学学报（自然科学版）》2012年第5期617-621,共5页Journal of East China University of Science and Technology

摘　　要：针对单通道语音增强问题,基于计算听觉场景分析(CASA)的原理,提出了一种基于CASA计算模型的语音增强改进算法。该算法在特征提取中选择了目标语音有效能量、信道互相关等特征,对语谱能量和互相关特征的阈值选取进行了改进。在5种低信噪比噪声干扰条件下的仿真实验结果证明,该算法输出增强语音的信噪比平均提高了9.32dB,有效地抑制了噪声。Based on computational auditory scene analysis （CASA）, this paper proposes an improved algorithm for monaural speech enhancement. In the proposed algorithm, both effective energy of target speech and cross-channel correlation are chosen as extracted feature. Moreover, this algorithm improves the threshold selection on energy spectrum and cross-channel correlation feature. Under the condition of low SNR with 5 different noises, the experimental results show that the proposed algorithm can raise the output SNR by 9.32 dB averagely, and attenuates noise effectively.

关 键 词：语音增强 计算听觉场景分析 语音有效能量 信道互相关 二值掩码 

分 类 号：TN912.3[电子电信—通信与信息系统]