Search

Numerous advanced and lightweight signal processing methods have been presented for single-channel speech enhancement (SE). It is imperative to carefully explore how to efficiently combine, integrate, and balance these methods. This paper proposes a more effective and less resource-intensive SE system, focused on the integration and adaptation of several approaches, especially the temporal cepstrum smoothing (TCS). First, a more robust fundamental frequency estimator is employed within TCS, mitigating the performance limitations caused by the inaccuracy of the original estimator. Additionally, a harmonic enhancement mechanism is introduced, effectively recovering the weak harmonic components. By incorporation of the modified TCS in the a posteriori speech presence probability estimation, the unbiased minimum mean square error noise power spectral density estimator can be refined. The modified TCS is also utilized for the a priori signal-to-noise ratio estimation. Moreover, this paper enhances the log-spectral amplitude estimator by applying both super-Gaussian speech priors and speech presence uncertainty for further improvement. Experimental evaluations demonstrate that the proposed method yields an improvement in speech quality while maintaining modest computational and storage requirements. Furthermore, the proposed system exhibits comparable performance to several baseline systems based on lightweight deep neural networks. [ABSTRACT FROM AUTHOR]

Spatial information is important for human perception of speech and sound signals. However, this information is often either distorted or completely neglected in noise reduction because it is challenging, to say the least, to achieve optimal noise reduction and accurate spatial information preservation at the same time. This paper studies the problem of binaural speech enhancement. By jointly diagonalizing the speech and noise correlation matrices, we present a method to construct the noise reduction filter as a linear combination of different eigenvectors, which span a certain subspace of the entire space. A different dimension of the subspace gives a different trade-off between noise reduction and speech/noise spatial information preservation. On the one side, if the dimension is equal to 1, maximum noise reduction is achieved but at the price of significant spatial information distortion. On the other extreme, if the dimension of the subspace is equal to that of the entire space, spatial information is accurately preserved but at the cost of no noise reduction. Therefore, one can achieve different levels of compromises between the amount of noise reduction and the level of speech/noise spatial information preservation by adjusting the dimension of the used subspace. [ABSTRACT FROM AUTHOR]

Reverberation impairs not only the speech quality, but also intelligibility. The weighted-prediction-error (WPE) method, which estimates the late reverberation component based on a multichannel linear predictor, is by far one of the most effective algorithms for dereverberation. Generally, the WPE prediction filter in every short-time-Fourier-transform (STFT) subband has to be long enough to estimate accurately the late reverberation component. As a consequence, WPE is computationally expensive, which makes it difficult to implement into real-time embedded or edge computing devices. Moreover, WPE is sensitive to additive noise and its performance may suffer from dramatic degradation even in environments where the signal-to-noise ratio (SNR) is high. To address these drawbacks, this letter proposes to decompose the multichannel linear prediction filter as a Kronecker product of a temporal (interframe) prediction filter and a spatial filter. An iterative algorithm is then developed to optimize the two filters. In comparison with the original WPE algorithm, the presented method not only exhibits better performance in terms of dereverberation and robustness to additive noise, as there are fewer parameters to estimate for a given number of observation signal samples, but is also computationally more efficient, since the dimensions of the covariance matrices after Kronecker product decomposition are smaller. [ABSTRACT FROM AUTHOR]

The knowledge of speech presence probability (SPP) plays an essential role in noise estimation and speech enhancement. Single channel SPP estimation and centralized multi-channel SPP estimation have been well studied. However, how to estimate SPP in wireless acoustic sensor networks (WASNs) remains a great challenge and few efforts can be found in this topic, particularly for WASN applications with multiple speakers. Accordingly, this paper is devoted to the problem of SPP estimation in WASNs and it presents a distributed model-based SPP estimation method for multi-speaker detection, which does not need any fusion center. A distributed k-means clustering method is first used to cluster the nodes into subnetworks, which detect different speakers. For each node in the subnetwork, the speech and noise power spectral densities are estimated locally by using a model-based method, then a distributed SPP estimator is developed and applied in every subnetwork. A distributed consensus method is used to obtain the distributed clustering and the distributed SPP estimation. Simulation results show that the proposed distributed clustering method can assign nodes into subnetworks based on their noisy observations. Moreover, the proposed distributed SPP estimator achieves robust speech detection performance under different noise conditions. [ABSTRACT FROM AUTHOR]

• Developing a binaural heterophasic MVDR beamforming for speech enhancement. • Rendering the desired speech and residual noise into different perceptual regions. • Implementing the proposed beamformer with DNN assisted statistics estimation. • Listening tests are conducted to evaluate the rendering performance. Beamforming is one of the most effective approaches to distant sound acquisition in complex acoustic environments, where noise, reverberation, and interference coexist; as a result, a significant number of efforts have been devoted to it over the last few decades. However, conventional beamformers produce a monaural output or colinear outputs, which are not optimal from the perception perspective. To take advantage of the human binaural hearing properties, a new type of fixed beamforming methods were developed recently, which attempt not only to attenuate noise but also render the signal of interest and residual noise into different perceptual regions, thereby achieving higher speech intelligibility. This work extends the principle of fixed binaural beamforming and develops a binaural heterophasic minimum variance distortionless response (MVDR) beamformer. A deep neural network (DNN) based noise estimation method is used to assist the implementation of this heterophasic MVDR beamformer, which is advantageous over the traditional one as it renders the desired source signal and residual noise to different perceptual regions, thereby yielding higher intelligibility. In comparison with the fixed binaural heterophasic beamformers, it can take advantage of the statistics of the noise to achieve better array performance. Results of simulations and listening tests validate the properties of the proposed technique. [ABSTRACT FROM AUTHOR]

In this paper, we introduce a new class of optimal rectangular filtering matrices for single-channel speech enhancement. The new class of filters exploits the fact that the dimension of the signal subspace is lower than that of the full space. By doing this, extra degrees of freedom in the filters, that are otherwise reserved for preserving the signal subspace, can be used for achieving an improved output signal-to-noise ratio (SNR). Moreover, the filters allow for explicit control of the tradeoff between noise reduction and speech distortion via the chosen rank of the signal subspace. An interesting aspect is that the framework in which the filters are derived unifies the ideas of optimal filtering and subspace methods. A number of different optimal filter designs are derived in this framework, and the properties and performance of these are studied using both synthetic, periodic signals and real signals. The results show a number of interesting things. Firstly, they show how speech distortion can be traded for noise reduction and vice versa in a seamless manner. Moreover, the introduced filter designs are capable of achieving both the upper and lower bounds for the output SNR via the choice of a single parameter. [ABSTRACT FROM AUTHOR]

Binaural noise reduction with a stereophonic (or simply stereo) setup has become a very important problem as stereo sound systems and devices are being more and more deployed in modern voice communications. This problem is very challenging since it requires not only the reduction of the noise at the stereo inputs, but also the preservation of the spatial information embodied in the two channels so that after noise reduction the listener can still localize the sound source from the binaural outputs. As a result, simply applying a traditional single-channel noise reduction technique to each channel individually may not work as the spatial effects may be destroyed. In this paper, we present a new formulation of the binaural noise reduction problem in stereo systems. We first form a complex signal from the stereo inputs with one channel being its real part and the other being its imaginary part. By doing so, the binaural noise reduction problem can be processed by a single-channel widely linear filter. The widely linear estimation theory is then used to derive optimal noise reduction filters that can fully take advantage of the noncircularity of the complex speech signal to achieve noise reduction while preserving the desired signal (speech) and spatial information. With this new formulation, the Wiener, minimum variance distortionless response (MVDR), maximum signal-to-noise ratio (SNR), and tradeoff filters are derived. Experiments are provided to justify the effectiveness of these filters. [ABSTRACT FROM AUTHOR]

Noise reduction, which aims at estimating a clean speech from noisy observations, has attracted a considerable amount of research and engineering attention over the past few decades. In the single-channel scenario, an estimate of the clean speech can be obtained by passing the noisy signal picked up by the microphone through a linear filter/transformation. The core issue, then, is how to find an optimal filter/transformation such that, after the filtering process, the signal-to-noise ratio (SNR) is improved but the desired speech signal is not noticeably distorted. Most of the existing optimal filters (such as the Wiener filter and subspace transformation) are formulated from the mean-square error (MSE) criterion. However, with the MSE formulation, many desired properties of the optimal noise-reduction filters such as the SNR behavior cannot be seen. In this paper, we present a new criterion based on the Pearson correlation coefficient (PCC). We show that in the context of noise reduction the squared PCC (SPCC) has many appealing properties and can be used as an optimization cost function to derive many optimal and suboptimal noise-reduction filters. The clear advantage of using the SPCC over the MSE is that the noise-reduction performance (in terms of the SNR improvement and speech distortion) of the resulting optimal filters can be easily analyzed. This shows that, as far as noise reduction is concerned, the SPCC-based cost function serves as a more natural criterion to optimize as compared to the MSE. [ABSTRACT FROM AUTHOR]

This paper investigates the problem of single-channel noise reduction in the time domain. The objective is to find a lower dimensional filter that can yield a noise reduction performance as close as possible to or even better than that obtained by the full-rank solution. This is achieved in three steps. First, we transform the observation signal vector sequence, through a semi-orthogonal matrix, into a sequence of transformed signal vectors with a reduced dimension. Second, a reduced-rank filter is applied to get an estimate of the clean speech in the transformed domain. Third, the estimate of the clean speech in the time domain is obtained by an inverse semi-orthogonal transformation. The focus of this paper is on the derivation of semi-orthogonal transformations under certain estimation criteria in the first step and the design of the reduced-rank optimal filters that can be used in the second step. We show how noise reduction using the principle of rank reduction can be cast as an optimal filtering problem, and how different semi-orthogonal transformations affect the noise reduction performance. Simulations are performed under various conditions to validate the deduced filters for noise reduction. [ABSTRACT FROM AUTHOR]

Abstract: Conventional multichannel noise reduction techniques are formulated by splitting the processed microphone observations into two terms: filtered noise-free speech and residual additive noise. The first term is treated as desired signal while the second is a nuisance. Then, the objective has typically been to reduce the nuisance while keeping the filtered speech as similar as possible to the clean speech. It turns out that this treatment of the overall filtered speech as the desired signal is inappropriate as will become clear soon. In this paper, we present a new study of the multichannel time-domain noise reduction filters. We decompose the noise-free microphone array observations into two components where the first is correlated with the target signal and perfectly coherent across the sensors while the second consists of residual interference. Then, well-known time-domain filters including the minimum variance distortionless response (MVDR), the space–time (ST) prediction, the maximum signal-to-noise ratio (SNR), the linearly constrained minimum variance (LCMV), the multichannel tradeoff, and Wiener filters are derived. Besides, the analytical performance evaluation of these time-domain filters is provided and new insights into their functioning are presented. Numerical results are finally given to corroborate our study. [Copyright &y& Elsevier]