Speech Enhancement in Distributed Microphone Arrays Using Polynomial Eigenvalue Decomposition

d'Olne, Emilie; Neo, Vincent W.; Naylor, Patrick A.

doi:10.23919/eusipco55093.2022.9909555

Cited by 5 publications

(3 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Likewise, the estimate {q m+1 (e jΩ ), λm+1 (e jΩ )} is extracted from R(m+1) (e jΩ ). Using (11) and the Bauer-Fike theorem [34], we find that the accuracy of the extracted (m + 1)st eigenvalue is upperbounded as…”

Section: B Perturbation Analysis and Error Propagationmentioning

confidence: 99%

“…One of the most popular decompositions is the eigenvalue decomposition (EVD). Due to its complexity even with efficient implementations [6][7][8], a partial or reduced EVD can be considered useful for low rank applications, such as in speech enhancement where a large number of microphones may record only a very limited number of speakers [9][10][11]. In the narrowband case, the power method in conjunction with Hotelling's deflation approach [12] is well suited for factorising rank-deficient matrices, where the number of eigenvalues and eigenvectors to be determined is smaller than the dimension of the matrix; hence this paper aims to extend this utility to the broadband case.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Low-Rank Para-Hermitian Matrix EVD via Polynomial Power Method with Deflation

Khattak,

Proudler,

Weiss

2023

2023 IEEE 9th International Workshop on Computational Advances in Multi-Sensor Adaptive Processing (CAMSAP)

View full text Add to dashboard Cite

The power method in conjunction with deflation provides an economical approach to compute an eigenvalue decomposition (EVD) of a low-rank Hermitian matrix, which typically appears as a covariance matrix in narrowband sensor array processing. In this paper, we extend this idea to the broadband case, where a polynomial para-Hermitian matrix needs to be diagonalised. For the low-rank case, we combine a polynomial equivalent of the power method with a deflation approach to subsequently extract eigenpairs. We present perturbation analysis and simulation results based on an ensemble of low-rank randomized para-Hermitian matrices. The proposed approach demonstrates higher accuracy, faster execution time, and lower implementation cost than state-of-the-art algorithms.

show abstract

Section: B Perturbation Analysis and Error Propagationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Low-Rank Para-Hermitian Matrix EVD via Polynomial Power Method with Deflation

Khattak,

Proudler,

Weiss

2023

2023 IEEE 9th International Workshop on Computational Advances in Multi-Sensor Adaptive Processing (CAMSAP)

View full text Add to dashboard Cite

show abstract

“…The topic of polynomial eigenvalue decomposition (PEVD) has recently gained traction in the signal processing literature. Applications were found in multichannel enhancement for arbitrarily-shaped arrays [1], spherical microphones [2], or distributed microphone networks [3]; in channel identification [4]; in DOA estimation with polynomial MUSIC [5][6][7]; in voice activity detection [8]; or in beamforming with a broadband MVDR beamformer [9]. These methods rely on the estimation of a space-time covariance matrix capturing signal correlations in space, time, and frequency, thereby allowing true broadband processing [1].…”

Section: Introductionmentioning

confidence: 99%

Frame-Based Space-Time Covariance Matrix Estimation for Polynomial Eigenvalue Decomposition-Based Speech Enhancement

d'Olne

Neo

Naylor

2022

2022 International Workshop on Acoustic Signal Enhancement (IWAENC)

View full text Add to dashboard Cite

Recent work in speech enhancement has proposed a polynomial eigenvalue decomposition (PEVD) method, yielding significant intelligibility and noise-reduction improvements without introducing distortions in the enhanced signal [1]. The method relies on the estimation of a space-time covariance matrix, performed in batch mode such that a sufficiently long portion of the noisy signal is used to derive an accurate estimate. However, in applications where the scene is nonstationary, this approach is unable to adapt to changes in the acoustic scenario. This paper thus proposes a frame-based procedure for the estimation of space-time covariance matrices and investigates its impact on subsequent PEVD speech enhancement. The method is found to yield spatial filters and speech enhancement improvements comparable to the batch method in [1], showing potential for real-time processing.

show abstract