Robust Estimation of Sound Source Direction with Deterministic Background Noise and Stochastic Source Dynamics Models

Direction of Arrival (DOA), a type of auxiliary information used in acoustic signal processing, is vulnerable to acoustical noise, so we want to male the estimation of DOA in noisy environments, relying on spectral sparseness. The energy of acoustic signals such as speech is wide-band, with individual signals localized in specific but different frequency regions. Our proposal involves filtering out spatial features provisionally from subband frequency components at the dominant frequency of the target signal using particle filtering with a sound source dynamics model. The feasibility of our proposal is confirmed by estimating a sound source direction in noisy conditions, also confirming that frequency selectivity and state estimation using particle filters help improve DOA estimation robustness against noise in noisy conditions.