Explainable DNN-Based Beamformer With Postfilter

• Published in: IEEE Transactions on Audio Speech and Language Processing Vol. 33; pp. 3070 - 3084
• Main Authors: Cohen, Adi, Wong, Daniel, Lee, Jung-Suk, Gannot, Sharon
• Format: Journal Article
• Language: English
• Published: IEEE 2025
• Subjects: Array signal processing; Beamforming; deep neural networks for spatial filtering; Microphones; Neural networks; Noise; postfilter; Spatial filters; Speech enhancement; Time-frequency analysis; Training; Vectors; Wiener filters
• ISSN: 2998-4173; 2998-4173
• DOI: 10.1109/TASLPRO.2025.3581110

This paper introduces an explainable DNN-based beamformer with a postfilter (ExNet-BF+PF) for multichannel signal processing. Our approach combines the U-Net network with a beamformer structure to address this problem. The method involves a two-stage processing pipeline. In the first stage, time-invariant weights are applied to construct a multichannel spatial filter, namely a beamformer. In the second stage, a time-varying single-channel post-filter is applied at the beamformer output. Additionally, we incorporate an attention mechanism inspired by its successful application in noisy and reverberant environments to improve speech enhancement further. The proposed scheme obviates the necessity for prior knowledge of the speaker's activity, which is required in classical beamforming designs. Furthermore, our study fills a gap in the existing literature by conducting a thorough spatial analysis of the network's performance. Specifically, we examine how the network utilizes spatial information during processing. This analysis yields valuable insights into the network's functionality, thereby enhancing our understanding of its overall performance. A thorough experimental study analyzes the performance of the proposed method in various scenarios and compares both the results and computational resource consumption with several baseline methods, including the classical Minimum Variance Distortionless Response (MVDR) beamformer, Deep Neural Network (DNN)-based spatial filters that preserve a beamforming structure, and deep end-to-end schemes.