A Compact Noise Covariance Matrix Model for MVDR Beamforming

Abstract

Acoustic beamforming is routinely used to improve the SNR of the received signal in applications such as hearing aids, robot audition, augmented reality, teleconferencing, source localisation and source tracking. The beamformer can be made adaptive by using an estimate of the time-varying noise covariance matrix in the spectral domain to determine an optimised beam pattern in each frequency bin that is specific to the acoustic environment and that can respond to temporal changes in it. However, robust estimation of the noise covariance matrix remains a challenging task especially in non-stationary acoustic environments. This paper presents a compact model of the signal covariance matrix that is defined by a small number of parameters whose values can be reliably estimated. The model leads to a robust estimate of the noise covariance matrix which can, in turn, be used to construct a beamformer. The performance of beamformers designed using this approach is evaluated for a spherical microphone array under a range of conditions using both simulated and measured room impulse responses. The proposed approach demonstrates consistent gains in intelligibility and perceptual quality metrics compared to the static and adaptive beamformers used as baselines.

Postfilter for Dual Channel Speech Enhancement Using Coherence and Statistical Model-Based Noise Estimation

“…We compared the performance of the proposed postfilter to those of the postfilter using the TBRR-based multichannel noise PSD estimator [ 21 ], which is denoted as TBRR , and the one adopting the SPP-based single-channel noise estimator introduced in Section 2.2 [ 36 ], which is denoted as Single-SPP . Although there have been several recent research studies on better spatial filtering [ 9 , 10 ], not much effort has been devoted to improve the postfilters for the spatial filtering recently, except for the deep learning-based approaches. Deep learning-based postfilters using single-channel [ 24 , 25 ] or multichannel information [ 23 , 27 , 28 ] have been proposed, but these approaches often require high computational complexity and large training datasets.…”

“…Over the past decades, there has been a growing demand for speech enhancement using microphone arrays in speech processing applications such as automatic speech recognition, mobile communications, and hearing aids [ 1 , 2 , 3 , 4 ]. Multichannel speech enhancement aims to reduce the additive noise and improve the quality of the speech signals obtained by multiple microphones placed in a variety of acoustic environments [ 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 ]. In many multichannel speech enhancement systems, beamforming algorithms, such as the minimum-variance distortionless-response (MVDR) beamformer [ 11 ] and the general transfer function generalized sidelobe canceler (TF-GSC) [ 12 , 13 ], have been employed to extract a desired signal, exploiting spatial information on the location of the sound sources.…”

Subspace Hybrid Beamforming for Head-Worn Microphone Arrays

“…In [24], the SCM is decomposed into isotropic, identity and plane-wave (PW) components followed by removal of any (near-) target PWs from the modelled SCM to avoid signal cancellation. Although some signal-dependent beamformers are shown [24] to perform better than signal-independent beamformers, such evaluations are typically limited to simulations or simple-scene real-recordings.…”

“…In [24], the SCM is decomposed into isotropic, identity and plane-wave (PW) components followed by removal of any (near-) target PWs from the modelled SCM to avoid signal cancellation. Although some signal-dependent beamformers are shown [24] to perform better than signal-independent beamformers, such evaluations are typically limited to simulations or simple-scene real-recordings. These typically avoid the real-world complexities of the cocktail party problem with wearable arrays such as rapid dynamics of the noise field due to head rotation or the presence of non-isotropic noise fields.…”

Unbiased estimators of two second-order moments of the covariance matrix

“…When the sample size is small, the performance of traditional signal estimation algorithms will rapidly decrease as the number of dimensions increases. The main reason is that the signal estimation algorithm often involves the covariance matrix between array elements [5], [6]. In practice, the covariance matrix is often unknown [7], and it needs to be estimated with limited samples [8].…”