Robust direction‐of‐arrival estimation by understanding global acoustic scene.

Abstract

Direction‐of‐arrival (DOA) is an important clue in acoustic signal processing. It is, however, difficult to accurately estimate DOAs in the presence of acoustic interference such as background noises and reverberation. The author has proposed a robust DOA finder with an environmental noise model, which describes spectral characteristics of background noises. Knowledge on acoustic interferences helps to make DOA estimation robust, but is difficult to be estimated exactly. The method works well only using the dominant subband components, in which the target signal is distinguished compared with interferences, in some noisy conditions. The noise model is updated time by time based on the DOA estimates to cope with nonstationary noises. Better DOA estimation gives a more accurate noise model and vice versa. In contrast, even a little error starts to cause negative feedback in DOA estimation and finally it goes to the fatal error. This paper proposes to introduce a switching scheme, which evaluates the reliabilities of DOA estimates in a stochastic manner and judge whether the noise model should be updated or not into the previously proposed DOA estimator. It is confirmed that the DOA estimator with the proposed scheme can improve its noise robustness even under nonstationary noises.