Audio Signal Processing with Microphone Arrays: Advances and Emerging Challenges (video)

Multi-microphone noise reduction and speaker separation are among the most extensively researched topics in audio signal processing. Traditionally, spatial filters—often referred to as beamformers—have been designed to satisfy specific optimization criteria, such as Minimum Variance Distortionless Response (MVDR), Speech Distortion Weighted Multichannel Wiener Filter (SDW-MWF), and Linearly Constrained Minimum Variance (LCMV). Incorporating acoustic propagation through the use of Relative Transfer Functions (RTFs) for each speaker of interest has proven beneficial. Criteria that leverage speech non-Gaussianity and sparsity have also been proposed, typically within the framework of independent component analysis. Significant efforts are dedicated to estimating the above beamformers' essential building blocks. The multi-microphone audio processing community enthusiastically embraced deep learning. Approaches based on Deep Neural Networks (DNNs) can be broadly categorized into Traditional beamformers controlled by DNNs, Fully DNN-based solutions without explicit beamformer formulations, and Hybrid methods in which selected components—such as weights or building blocks—are learned by the DNN while preserving the underlying beamformer criteria. This webinar will begin with a consolidated overview of beamforming in speech enhancement, emphasizing the role of acoustic propagation. The presenter will then explore the three categories of DNN-based multi-microphone approaches with examples of recent algorithms and conclude by discussing emerging challenges, including robustness, generalizability, and explainability