Robust Time-Delay Estimation for Speaker Localization Using Mutual Information Among Multiple Microphone Signals

Time-delay estimation algorithms for speaker localization usually suffer from adverse effects of background noise and reverberation. The multichannel cross-correlation coefficient (MCCC) algorithm exploits spatial redundancy among multiple microphone signals to boost the robustness of the time-delay estimator. The MCCC algorithm, however, does not completely exploit the useful redundancy among the signals received at an array of microphones for the time-delay estimation. This issue is investigated in this paper from an information theory perspective. An equivalent time-delay estimation algorithm is derived to disclose the mechanism and limitation of the MCCC algorithm. Two new time-delay estimation algorithms are proposed on the basis of the mutual information among an array of microphones. The first algorithm fully utilizes the mutual information between all the different microphone signals to enhance its robustness to reverberation. The second algorithm emphasizes the monotone factors in these mutual information functions to promote its robustness to noise and reverberation. The effectiveness of the new time-delay estimators is demonstrated in noisy and reverberant environments.