A Vectorized Method for Computationally Efficient SRP-PHAT Sound Source Localization

摘要

Most microphone array applications require sound source localization for subsequent multichannel signal processing algorithms. Steered response power-based source localization has been the most popular method due to its robustness against reverberation and background noise. Unfortunately, it requires a grid search method whose computational cost is proportional to the number of grid points, which makes it challenging for real-time applications. This paper proposes an efficient method for computing the steered response power by using two different techniques. Firstly we propose to separate preruntime computations from runtime computations. Secondly we vectorize the computations and group similar computations with a multi-threaded library (IPP) to efficiently exploit standard processing architectures. Experimental results show that the proposed method significantly reduces the computational load of microphone array processing, making it viable for real-time implementations on constrained devices.