Joint Learning of Covariance Estimation and White Noise Gain for Robust MVDR Beamforming

2026-06-23

AI Summary

This work addresses the sensitivity of minimum variance distortionless response (MVDR) beamforming to microphone self-noise and array mismatch in real-world scenarios, where conventional approaches relying on fixed white noise gain (WNG) thresholds or diagonal loading struggle to adapt to time-varying acoustic conditions. To overcome this limitation, the paper proposes a data-driven, end-to-end robust MVDR framework that jointly learns time-frequency noise masks and frequency-dependent WNG thresholds for the first time. By integrating a differentiable MVDR layer, the method enables joint optimization of covariance matrix estimation and beamforming without manual parameter tuning. Evaluated across diverse acoustic environments, the proposed approach significantly improves speech quality and intelligibility, consistently outperforming traditional MVDR variants with fixed WNG constraints.

Abstract

The minimum variance distortionless response (MVDR) beamformer is widely used for multichannel speech enhancement due to strong noise suppression while preserving target signals. In practice, its performance is sensitive to microphone self-noise and array mismatches. Existing approaches typically rely on fixed, manually tuned WNG thresholds or diagonal loading, leading to suboptimal performance under unknown or time-varying acoustic conditions. This paper proposes a data-driven MVDR framework that adaptively estimates the WNG constraint using a deep neural network. The network jointly predicts a time-frequency noise mask for covariance estimation and a frequency-dependent WNG threshold, enabling dynamic robustness-directivity control. A differentiable robust MVDR layer is integrated into the framework, allowing end-to-end optimization. Experiments demonstrate consistent improvements in speech quality and intelligibility over conventional fixed-WNG MVDR methods.

Problem

• MVDR beamforming
• white noise gain
• covariance estimation
• robustness
• array mismatch

Innovation

• joint learning
• covariance estimation
• white noise gain
• robust MVDR
• differentiable beamforming