On The Use Of Machine Learning In Microphone Array Beamforming For Far-Field Sound Source Localization
Daniele Salvati, University of Udine
Carlo Drioli, University of Udine
Gian Luca Foresti, University of Udine

Abstract:
This paper presents a weighted minimum variance distortionless response (WMVDR) algorithm for far-field sound source localization in a noisy environment. The broadband beamforming is computed in the frequency-domain by calculating the response power on each frequency bin and by fusing the narrowband components. A machine learning method based on a support vector machine (SVM) is used for selecting only the narrowband components that positively contribute to the broadband fusion. We investigate the direction of arrival (DOA) estimation problem using a uniform linear array (ULA). The skewness measure of response power function is used as input feature for the supervised SVM learning. Simulations demonstrate the effectiveness of the WMVDR in an outdoor noisy environment.