A New Weighted Decision Making Method for Accurate Sound Source Localization

Abstract

Sound source localization is a challenging task in adverse environments with high reverberation and low signal-to-noise ratio. To accurately localize the source through classification methods, the number of sub-spaces for classification decision should be large. However, this also causes high misclassification rate, leading to larger localization error, especially in adverse environments. In this paper, we propose a new weighted decision making method (WDMM), which can effectively improve the localization accuracy of the likelihood based classification algorithms, by revisiting and accessing the probabilities of the adjacent sub-spaces. The synthetic experimental results have shown that the average mean and average standard deviation of the localization errors from the 20 different acoustic environments by the proposed WDMM are only 0.42 and 0.21 m respectively in a 4.0 m × 4.0 m × 4.0 m room. The 20 different acoustic environments include the high reverberation T ₆₀ up to 0.6 s and low signal-to-noise ratio to −10 dB. Compared to localization results without WDMM, the proposed method has reduced averages of mean and standard deviation localization errors by 35.8 and 55.2% respectively.