基于频率信噪比加权的麦克风阵列声源定位算法  

Sound Source Localization Using SNR-based Frequency Weighting with Microphone Array

作　　者：赵小燕 陈书文 周琳[3] Zhao Xiaoyan;Chen Shuwen;Zhou Lin(School of Information and Communication Engineering, Nanjing Institute of Technology, Nanjing, Jiangsu 211167, China;School of Mathematics and Information Technology, Jiangsu Second Normal University, Nanjing, Jiangsu 210013, China;School of Information Science and Engineering, Southeast University, Nanjing, Jiangsu 210096, China)

机构地区：[1]南京工程学院信息与通信工程学院,江苏南京211167 [2]江苏第二师范学院数学与信息技术学院,江苏南京210013 [3]东南大学信息科学与工程学院,江苏南京210096

出　　处：《信号处理》2020年第3期449-456,共8页Journal of Signal Processing

基　　金：国家自然科学基金(61571106);江苏省自然科学基金(BK20170757);江苏省高校自然科学基金(17KJD510002)。

摘　　要：为了提高噪声和混响环境下麦克风阵列的声源定位算法性能,提出了一种基于频率信噪比加权的可控响应功率定位算法。该算法首先根据每帧阵列信号的频域协方差矩阵估计每个频率的信噪比;然后通过激活函数将频率信噪比映射为加权值,并修正传统的相位变换可控响应功率计算公式;最后利用修正公式计算每个候选位置的可控响应功率值,通过搜索可控响应功率的最大值实现声源定位。该算法无需噪声先验信息,根据实时估计的频率信噪比自适应地调整各频率分量对可控响应功率的贡献。仿真环境和实际环境测试结果表明,与传统的相位变换可控响应功率算法、维纳预滤波波束形成算法相比,在噪声和混响的复杂声学环境下,本文算法的定位正确率更高,均方根误差更小,对噪声的鲁棒性更强。In order to improve the performance of sound source localization with microphone array in reverberant noisy environments,a frequency weighted sound source localization algorithm based on signal-to-noise ratio(SNR)is proposed.First,SNR of each frequency is estimated from the covariance matrix of the array signals in each frame.Then,the SRP-PHAT(steered response power-phase transform)formula is revised by the weight mapped from the frequency SNR by activation function.Finally,the SRP of each candidate location is calculated by the revised formula,and the sound source location is estimated by searching the maximum value of all the SRPs.The proposed method adaptively adjusts the contribution of frequency component to SRP according to the frequency SNR estimated in real time without the requirement of prior knowledge about noise.Simulation and real environment test results demonstrate that,compared with the conventional SRP-PHAT algorithm and Wiener pre-filtering beamformer algorithm,the proposed method obtains a higher percentage of correct estimates and a lower root mean square error,and is more robust against noise in complex noisy and reverberation environments.

关 键 词：麦克风阵列 声源定位 频率信噪比加权 相位变换加权可控响应功率 

分 类 号：TN912.3[电子电信—通信与信息系统]