Multiple speech source localization using ambiguity suppression

We propose a novel omnidirectional direction of arrival (DOA) estimation method for multiple speech source with a compact equilateral triangular microphone array. The time difference of arrival is extracted from the mixture ratios for broadband speech signal. We apply expanding technique to estimate the location of signal source with all the possible incident angles and obtain possibility histogram of DOA omnidirectionally. Then ambiguity suppression method using binary masking is introduced to eliminate unreliable candidates and combine all in the same coordinate system. And a Von Mises mixture model based clustering method is used to estimate the final speech source direction. The simulation results show that the proposed method has robustness in noisy and reverberant environment, and can be applied to the underdetermine case that the number of speech sources is greater than the number of sensors.