Robust Speech-Distortion Weighted Interframe Wiener Filters for Single-Channel Noise Reduction

Abstract

In this paper, speech-distortion weighted (SDW) inter-frame Wiener filters (IFWFs) are investigated for singlechannel noise reduction in a filter bank structure. The filters utilize a parameter µ that explicitly sets a trade-off between noise reduction and speech distortion and have traditionally been used in multi-channel applications under the term speechdistortion weighted multichannel Wiener filter (SDW-MWF). The application of these SDW-IFWFs relies on the estimation of interframe correlation (IFC) coefficients and it is shown that the IFC coefficients can be more robustly estimated using a secondary higher resolution filter bank (HRFB). It is then shown how realvalued scalar gains, which are optimal in the primary filter bank, can be applied directly in the HRFB instead of the inter-frame filtering in the primary filter bank, which leads to a more robust noise reduction performance for any value of µ. Computing these gains is also cheaper since matrix inversions are avoided and the primary filter bank is not needed in the actual implementation. Experimental results are given that support the claims, where the proposed methods are compared to relevant reference methods using measures such as the segmental SNR and the objective PESQ.

Robust Constrained MFMVDR Filters for Single-Channel Speech Enhancement Based on Spherical Uncertainty Set

“…Since in practice the noise n(m) is obviously not available, both the normalized noise correlation vector as well as the a-priori SNR need to be estimated from the noisy speech STFT coefficients in order to be able to estimate the normalized speech correlation vector based on (14). Assuming that the normalized speech correlation vector γ x (m) and the normalized noise correlation vector γ n (m) follow multivariate complex Gaussian distributions, the ML estimate of the normalized speech correlation vector γ x (m) was derived in [12] as…”

“…Several approaches were proposed to estimate the normalized speech correlation vector from the noisy speech STFT coefficients. In [12] a maximum-likelihood (ML) estimator was derived based on the assumption that the normalized speech and noise correlation vectors follow multivariate Gaussian distributions, while in [14] it was proposed to estimate the normalized speech correlation vector by applying the Wiener-Khinchin theorem on estimated periodograms in a high frequency-resolution filterbank. In [24], we showed that accurately estimating the normalized speech correlation vector is crucial since even small estimation errors may lead to a degraded performance of the MFMVDR filter, causing speech distortion and unpleasant artifacts in the background noise.…”

Deep Multi-Frame MVDR Filtering for Single-Microphone Speech Enhancement

“…Multi-frame speech enhancement approaches [6,7,8,10] estimate the speech component by applying a finite impulse response filter with N taps to the noisy STFT coefficients, i.e.,…”

“…In contrast to single-frame approaches, multi-frame approaches [6,7,8,9,10] apply a complex-valued filter to the noisy STFT coefficients and are able to take into account the speech correlation across consecutive time frames. Similarly to the minimum-variance-distortionless-response (MVDR) beamformer and the minimum-power-distortionless-response beamformer (MPDR) for multi-microphone speech enhancement [4,11], multi-frame MVDR (MFMVDR) and multi-frame MPDR (MFMPDR) filters have been proposed for single-microphone speech enhancement [6,7,10].…”

On Improvement of Speech Intelligibility and Quality: A Survey of Unsupervised Single Channel Speech Enhancement Algorithms

“…Kristian and Marc [51] investigated speech-distortion weighted inter-frame Wiener filters for the SCSE in a filterbank configuration. The filterbank configuration utilized a regularization parameter as a tradeoff between speech distortion and noise reduction.…”