分布式压缩感知麦克风阵列多声源方位估计*

麦克风阵列已被广泛应用于音/视频会议等人机交互领域中时，多声源应用场景对声源方位估计性能提出了更高的要求。压缩感知(CS)声源定位算法将声源定位问题转化为信号的稀疏重构问题，相比传统的定位算法如相位变换加权（SRP-PHAT）和时延累加定位（DS）能够获得较高的定位性能，但多声源的存在一定程度上降低了稀疏程度，影响了CS重构性能。考虑到传统的CS定位算法并未利用多个连续语音帧之间声源空间向量的共同稀疏性，提出采用分布式压缩感知（DCS）理论以改善多声源的稀疏恢复估计的性能。仿真和实验结果表明，相比于传统定位算法和CS-OMP算法，DCS-SOMP算法在不同信噪比和不同声源强度的环境中，对多声源的方位估计都具有更好的定位性能和定位稳健性。

Distributed compressed sensing microphone array multi-source azimuth estimation

Microphone arrays have been widely used in the field of human-computer interaction such as audio/video conferencing. It is necessary to make higher-resolution azimuth estimation performance for scenes with multiple sound sources in different orientations. The Compressed Sensing (CS) sound source localization algorithm transforms the sound source localization problem into a sparse reconstruction problem of the signal, thus achieves better estimation performance compared to traditional localization algorithms such as phase shift weighting (SRP-PHAT) and time delay-sum(DS). However, the existence of multiple sound sources reduces the sparsity, to some extent degrades the performance of CS reconstruction. Considering that the traditional CS localization algorithm does not utilize the common sparsity of the sound source space vector between multiple consecutive speech frames, in this paper the distributed compressed sensing (DCS) theory is proposed to improve the performance of sparse recovery estimation of multiple sound sources. The simulation and experimental results show that compared with traditional positioning algorithm and CS-OMP algorithm, DCS-SOMP algorithm has better positioning performance and robustness for multi-sound source azimuth estimation under different SNR and different sound source intensity environments.