This paper presents a novel method to estimate the direction-of-arrivals (DOAs) of multiple speakers based on time-frequency sparsity. The acoustic interferences are first suppressed, and then a concave cost function (CCF) is utilized to estimate bin-wise DOAs. The DOAs of sources are subsequently identified by picking peaks in a histogram of bin-wise azimuths. Three aspects distinguish this method from conventional ones. First, a closed-form solution to bin-wise DOA is given by CCF, which replaces the extensively used grid search and enables high computational efficiency. Second, signal enhancement is employed to suppress acoustic interferences on phase spectra. Last, the time-delay weights mitigate the effect of delay outliers. The proposed method is compared with well-established methods in simulated environments. The experimental results confirmed its superiority in computational efficiency and acoustic robustness.
Bibliographic reference. Ying, Dongwen / Zhou, Ruohua / Li, Junfeng / Pan, Jielin / Yan, Yonghong (2014): "Direction-of-arrival estimation of multiple speakers using a planar array", In INTERSPEECH-2014, 2223-2227.