Abstract
A probabilistic time difference of arrival (TDOA) estimation method is extended for multiple pairs of microphones to collaboratively perform source localization. To deal with reverberation, we treat the reverberant signal as a separate type of noise, and the TDOA estimation result outperforms existing approaches such as an interaural phase difference (IPD) method [22], the GCC-PHAT [12], and the adopted method based on probabilistic modeling [21]. Then, the task of sound source localization using arbitrarily many pairs of microphones is formulated as a maximum likelihood problem, and the x-y coordinate of the sound source is estimated by gradient descent minimization of the cost function. Simulation in a reverberant room shows that, when using two pairs of microphones for source localization, large errors tend to occur when the source and the microphones are on the same line. Nevertheless, by adding a third pair of microphones, the proposed algorithm is able to reduce the maximum mean square localization error by more than 20%. And the relationship with different γ and 〖RT〗_60 is tested. The simulation shows that, no matter what 〖RT〗_60 is, γ=0.5 has the better performance than γ=0.1 and 0.3 in our system. However, γ=0.7 has the lower position error when 〖RT〗_60 is 0.45 s, and γ=0.9 has the lower position error when 〖RT〗_60 is 0.6 s.