New analytical calculation and estimation of TDOA for underdetermined BSS in noisy environments

We have proposed a new algorithm for sparseness-based underdetermined blind source separation (BSS) that can cope with diffused noise environments. This algorithm includes a technique for estimating the time-difference-of-arrival (TDOA) parameter separately in individual frequency bins for each source. In this paper, we propose methods that integrate the frequency-bin-wise TDOA parameter to estimate the TDOA of each source. The accuracy of TDOA estimation with the proposed approach is shown experimentally in comparison with a conventional approach. The separation performance and calculation time of the proposed approach is also examined.