基于语音稀疏性的频域声回波对消算法研究  

Research on frequency domain acoustic echo cancellation(AEC) algorithm based on speech sparseness

作　　者：史兆丰[1] 夏秀渝[1] 

机构地区：[1]四川大学电子信息学院,成都610064

出　　处：《四川大学学报（自然科学版）》2013年第2期311-316,共6页Journal of Sichuan University(Natural Science Edition)

基　　金：国家自然科学基金(61071159)

摘　　要：本文提出一种新的基于语音稀疏性的频域声回波对消算法.新算法首先利用短时傅里叶变换(STFT)将远端和近端信号变换到时频域,然后基于语音在时频域具有近似稀疏分布的特点,将远端信号每个时频单元与预先设置的高低能量门限进行比较,判断相对应的近端信号的时频单元信噪比高低,进而决定频域自适应滤波器权系数的调整方式,最后在频域完成回波抵消.本文所提新算法与NLMS算法、DLMS算法相比,具有计算量小、适应环境变化能力强、不需要双端话音检测等优点.A new frequency-domain Acoustic Echo Cancellation（AEC） algorithm based on speech sparseness is proposed.Firstly,far-end and near-end signals are converted to time-frequency domain by short-time Fourier transform（STFT）.According to the characteristics of approximately sparseness distribution in speech signal processing,the algorithm compares each time-frequency unit of the far-end signal with the predefined energy thresholds and measures the SNR of the corresponding time-frequency unit of near-end signal.Then a new method concerning how to adjust weight coefficients of frequency-domain adaptive filter is presented.Finally,echo cancellation is achieved in the frequency-domain.After comparing performance of the new method with NLMS algorithm and DLMS algorithm,simulation results indicate that the proposed algorithm has advantages of low amount of calculation and strong resistance to environment changes,and doesn＇t need double-talk detection.

关 键 词：频域 声回波对消 稀疏性 NLMS DLMS 

分 类 号：TN912.3[电子电信—通信与信息系统]