面向语音增强的聚焦自适应波束形成方法  

Wideband Adaptive Beamforming Using Adjustable Focusing Method for Speech Enhancement

作　　者：陈鹏[1] 李洋洋 高婧洁 王威 CHEN Peng;LI Yangyang;GAO Jingjie;WANG Wei(School of Information Engineering,Chang’an University,Xi’an 710064,China)

机构地区：[1]长安大学信息工程学院,西安710064

出　　处：《西安交通大学学报》2022年第11期148-155,共8页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(61871059,61901057);陕西省科技计划项目重点研发计划资助项目(2021KWZ-08)。

摘　　要：针对语音信号相对带宽大、麦克风存在误差以及现有自适应波束形成方法宽带干扰抑制能力不足的问题,提出一种面向语音增强的聚焦自适应波束形成方法。首先将Capon空间谱处理为接收信号的概率密度函数,并对不同频带设计了基于阵列观测数据的聚焦矩阵,从而根据干扰功率自动调整频率聚焦程度;然后,计算聚焦后的样本协方差矩阵和各子带频段内的干扰加噪声协方差矩阵,并利用聚焦矩阵重构出宽带数据的干扰加噪声协方差矩阵;最终,通过重构出参考频率下的期望信号协方差矩阵,修正期望信号方向的导向矢量并得到自适应波束形成器的加权向量。数值仿真结果表明:所提方法能够有效地对期望信号进行接收的同时充分抑制宽带干扰;所提方法在强干扰处的聚焦误差仅为现有方法的约1%;在信噪比为15 dB且入射角误差为3°的情况下,所提方法的输出信噪比较现有方法提高约12 dB。This paper proposes an adjustable focusing method for adaptive beamforming to address the problems of large relative bandwidth of speech signals,errors in microphones and insufficient broadband interference suppression with existing adaptive beamforming methods.Firstly,the Capon spectrum is processed as the probability density function of the received signals,and the focusing matrix based on the array observation data is designed for different frequency bands,so as to automatically adjust the frequency focusing degree according to the interference power;then,the focused sample covariance matrix(SCM)and the interference-plus-noise covariance matrix(INCM)in each frequency band are obtained,and the INCM of the broadband data is reconstructed by using the focusing matrix;finally,the desired signal covariance matrix(DSCM)at the reference frequency is reconstructed,the steering vector(SV)is corrected and the weighting vector of the proposed beamformer is obtained.According to the results of several numerical simulations,the proposed method allow the desired signals to be effectively received and the wideband interference sufficiently suppressed;the focusing error of the proposed method under strong interference is only about 1%that of the existing methods;when all signals have an incident angle error of 3°and the signal-to-noise ratio(SNR)is 15 dB,the output signal-to-interference-and-noise-ratio(SINR)of the proposed method is about 12 dB higher than the existing methods.

关 键 词：语音增强 聚焦矩阵 自适应波束形成 宽带干扰 导向矢量 

分 类 号：TN912.35[电子电信—通信与信息系统]