Sparse Iterative Beamforming Using Spherical Microphone Arrays for Low-Latency Direction of Arrival Estimation in Reverberant Environments

Acoustic direction of arrival estimation methods allows positional information about sound sources to be transmitted over a network using minimal bandwidth. For these purposes, methods that prioritize low computational overhead and consistent accuracy under non-ideal conditions are preferred. The estimation method introduced in this paper uses a set of steered beams to estimate directional energy at sparsely distributed orientations around a spherical microphone array. By iteratively adjusting beam orientations based on the orientation of maximum energy, an accurate orientation estimate of a sound source may be produced with minimal computational cost. Incorporating conditions based on temporal smoothing and diffuse energy estimation further refines this process. Testing under simulated conditions indicates favorable accuracy under reverberation and source discrimination when compared with several other contemporary localization methods. Outcomes include an average localization error of less than 10 degrees under 2 s of reverberation lime (T60) and the potential to separate up to four sound sources under the same conditions. Results from testing in a laboratory environment demonstrate potential for integration into real-time frameworks.