Speech Enhancement With Robust Beamforming for Spatially Overlapped and Distributed Sources

Most of the existing Beamforming methods are based on assumptions that sources all point and angular separation between direction arrival (DOA) source interference is large enough to assure good performance. In this paper, we consider a tough scenario where target simultaneously spatially distributed overlapped. To improve performance in scenario, propose two approaches: first approach exploits non-Gaussianity as well spectrogram sparsity output microphone array; second generalized with overlapped groups Beampattern. The proposed criteria solved by linearized preconditioned alternating method multipliers (LPADMM) high accuracy computational efficiency. Numerical simulations real data experiments show advantages approaches compared previously for signal enhancement.