Improving Acoustic Localization Accuracy by Applying Interaural Level Difference and Support Vector Machine for AoA Outlier Removal

In this paper, we propose a novel method for gunshot sound source localization based on range detection by combining steered response power phase transformation (SRP-PHAT) and interaural level difference (ILD) localization for improved localization performance. The conventional SRP-PHAT method is accurate in estimating the Angle of Arrival (AoA). However, the range estimation from SRP-PHAT has low accuracy and high variance. For this reason, we use only the AoA estimate from SRP-PHAT and apply a support vector machine (SVM) for AoA outlier detection. We then use a ILD-based method for range localization to decrease the localization distance error. In contrast to other sound source localization methods which are inadequate for detecting gunshot sound because of its very short impulse, our method is able to detect gunshot sound.