Multi-Resolution Spectrums Analysis-Based DOA Estimation Using Energy and Sparsity Joint Mask

First Order Ambisonics has attracted significant attention for direction-of-arrival estimation. Combining with the time-frequency analysis techniques, existing methods could realize accurate localization results in most of acoustic scenarios. However, these methods encounter challenges such as energy leakage due to the uncertainty of linear time-frequency transformation, and insufficient localization cues in high-reverberant noisy acoustic environments. This paper addresses these challenges by integrating the localization cues from spectrums with varying time-frequency resolutions. Specifically, by applying analysis window with different sizes, the spectrums with non-redundant localization cues can be generated. Then, a joint mask is designed by combining Hoyer sparsity and inter-channel energy measurements to assess the localization contribution at each point within spectrums. The accurate direct-of-arrival estimation can be achieved by applying the masked DOA cues of T-F points from spectrums with different resolutions. Objective evaluations are conducted in both simulation and actual recording environments. Corresponding results prove that the proposed method could exhibit a superior localization performance than several existing localization estimators.