Fast Sound Source Localization Using Two-Level Search Space Clustering

Abstract

Steered response power phase transform (SRP-PHAT) is a method that is widely used for robust sound source localization (SSL). However, since SRP-PHAT searches over a large number of candidate locations, it is too slow to run in real-time for large-scale microphone array systems. In this paper, we propose a robust two-level search space clustering method to speed-up SRP-PHAT-based SSL. The proposed method divides the candidate locations of the sound source into a set of groups and finds a small number of groups that are likely to contain the maximum power location. By searching within the small number of groups, the computational costs are reduced by 61.8% compared to a previously proposed method without loss of accuracy.

Parallel SRP-PHAT for GPUs

“…The 32-sensor data were generated using a similar method to that described by (Allen and Berkley, 1979). For some applications such as a spoken language based humanoid robot interface (Cho et al, 2009;Yook et al, 2015), estimated angles within ±10 • from the true sound source direction are generally acceptable. The sound source localization accuracy was 99.99% when the estimated angles within ±10 • of the sound source were considered correct.…”

“…The sound source localization accuracy was 99.99% when the estimated angles within ±10 • of the sound source were considered correct. When estimated DOAs that were within ±5 • of the true azimuth and ±3 • of the true elevation were considered correct, the sound source localization accuracy was 76% (Yook et al, 2015). Note that the heights of the sound sources were 170 cm in (Yook et al 2015).…”

“…When estimated DOAs that were within ±5 • of the true azimuth and ±3 • of the true elevation were considered correct, the sound source localization accuracy was 76% (Yook et al, 2015). Note that the heights of the sound sources were 170 cm in (Yook et al 2015). The FFT and IFFT were implemented using the CUFFT library (NVIDIA, 2014).…”

“…However, since it performs a grid search examining a large number of candidate sound source locations, the SRP-PHAT is computationally expensive and may not be used in real time. Many approaches have been proposed for a fast SRP-PHAT (Zotkin and Duraiswami, 2004;Peterson and Kyriakakis, 2005;Do and Silverman, 2007;Dmochowski et al, 2007;Cho et al, 2009;Yook et al, 2015). For instance, a hierarchical search method (Zotkin and Duraiswami, 2004;Do and Silverman, 2007) gradually prunes the candidate sound source locations in a coarse-to-fine search.…”

“…Although it is guaranteed to find a global maximum power location, this method is suitable for small-scale microphone array systems because there may be too many candidate coordinates in the look-up table for a large-scale microphone array system (Cho et al, 2009). Recently, improvement of this method was proposed by adopting the idea of the hierarchical search method (Yook et al, 2015). All of the above methods focus on removing repetitive computation from the algorithm or reducing the number of candidate sound source locations that must be searched.…”