Far-field Azimuth localization based on time delay of arrival estimation

This paper addresses the issue of the Far-Field Sound Source Localization (SSL) based on Time Delay of Arrival (TDOA) Estimation. We propose the use of sound sources´ excitation characteristics for TDOA estimation, and Azimuth localization. Our approach localizes multi-sources using only two observed mixtures captured by two microphones. The method has been performed on real stereo mixtures, recorded in a normalized 5.1 Dolby studio. The localization performance is compared to other techniques in terms of objective metrics, essentially Localization Success Rate (LSR), and Average Localization Error (ALE). Experiments show that the proposed technique is straightforward and promises the most accurate results with a precision of 3 degrees.