DOA estimation in underdetermined stereophonic mixture by kernel density estimator

An estimation method of Direction-of-arrival (DOA) of multiple speech sources from a stereophonic mixture in an underdetermined case is proposed. Based on source sparseness in time-frequency domain, an improved algorithm is proposed via reliable spectrogram components selection and the kernel density estimator approach based on a probabilistic model in phase difference observation. The performance of the proposed method is assessed by real data. Results obtained by our method show outperform the others in direction angle resolution and robustness against noise.