利用多点互相关值均值的实时声源定位算法

基于可控响应功率的麦克风阵列声源定位算法在噪声和混响环境中具有较好的鲁棒性，但是利用网格搜索方法寻找可控响应功率的最大值计算复杂度很高。故提出一种新的实时声源定位算法，使用网格中多互相关值均值代替单点的可控响应功率值作为代价函数，通过多级区域收缩方法实现了代价函数最大值的空间搜索，并给出一种不显著增加计算量的互相关值求和范围的计算方式。由于利用了网格中更多的空间采样信息，因此可以使用更大尺度的搜索网格，从而降低了整个搜索过程的计算复杂度。

Real-time acoustic source localization algorithm based on the average of multiple cross correlation values

The Steered Response Power(SRP)-based acoustic source localization algorithm with microphone arrays has been shown to be robust in noisy and reverberant environments. However, grid-search methods used to find a global maximum of the SRP image are computationally-intensive. In this paper, we propose a different strategy where, instead of the SRP at discrete spatial positions, the average of multiple Cross-Correlation(CC) values is evaluated as a cost function. The multi-stage region contraction is then used to find the maximum of the cost function to reduce the computational burden. The determination of CC accumulation limits does not increase the amount of calculation significantly. For incorporating additional spatial knowledge at each search grid, the proposed algorithm allows for a coarser spatial grid and reduces the computational cost with almost no loss of accuracy. Experiments carried out under different acoustic conditions confirm the validity of the proposed approach.