Acoustic Source Separation Research Articles

End-to-end underwater acoustic source separation model based on EDBG-GALR

In practical work scenarios, it is common for multiple vessels to be present in close proximity within the same water area. In such cases, the acoustic signals emitted by these vessels often overlap, causing interference and reducing the accuracy of vessel identification. This paper proposes an improved GALR end-to-end underwater acoustic source separation algorithm, with ECA-DE (Efficient Channel Attention-Deep Encoder) and Bi-GRU (Bidirectional Gated Recurrent Unit), which consists of an encoder, separation blocks, and a decoder (EDBG-GALR). Addressing the issue of GALR encoder’s limited expressiveness due to its single-layer one-dimensional convolutional layer, we introduce a new deep encoder, ECA-DE, to enhance the encoder’s capability to process temporal signals and improve the efficiency of the separator’s input. Additionally, we integrate Bi-GRU into the GALR separation block’s local modeling module to enhance the local modeling ability of sequence features, thereby reducing the model’s computational and parameter requirements. Experimental results demonstrate that the proposed EDBG-GALR method can effectively separate mixed multi-target signals into multiple single-target signals, achieving a maximum scale-invariant signal-to-noise ratio (SI-SNR) improvement of 3.32 dB and a signal distortion ratio (SDR) improvement of 3.38 dB over baseline methods. These results highlight the practical applicability of EDBG-GALR in complex underwater environments.

Applying matched field array processing and machine learning to computational auditory scene analysis and source separation challenges

Matched field processing (MFP) techniques employing physics-based models of acoustic propagation have been successfully and widely applied to underwater target detection and localization, while machine learning (ML) techniques have enabled detection and extraction of patterns in data. Fusing MFP and ML enables the estimation of Green’s Function solutions to the Acoustic Wave Equation for waveguides from data captured in real, reverberant acoustic environments. These Green’s Function estimates can further enable the robust separation of individual sources, even in the presence of multiple loud, interfering, interposed, and competing noise sources. We first introduce MFP and ML and then discuss their application to Computational Auditory Scene Analysis (CASA) and acoustic source separation. Results from a variety of tests using a binaural headset, as well as different wearable and free-standing microphone arrays are then presented to illustrate the effects of the number and placement of sensors on the residual noise floor after separation. Finally, speculations on the similarities between this proprietary approach and the human auditory system’s use of interaural cross-correlation in formulation of acoustic spatial models will be introduced and ideas for further research proposed.

Sparse Representations With Legendre Kernels for DOA Estimation and Acoustic Source Separation

Recording multiple sound sources in a reverberant environment results in convolutive mixtures. Sound sources can be extracted from microphone array recordings of such mixtures using acoustic source separation techniques. Acoustic source separation using recordings obtained from rigid spherical microphone arrays (RSMA) benefit from the representation of sound fields as series of spherical harmonics. More specifically, RSMAs afford increased flexibility in acoustic beamforming and spatial filtering. We propose a data-driven DOA estimation and acoustic source separation method based on a dictionary-based sparse decomposition of sound fields. The proposed method involves identifying the time-frequency bins with contributions from a single source only and those with sensor noise or diffuse sound field components. The former set of bins is used in DOA estimation and beamforming in the sparse decomposition domain. The latter set is used to calculate the diffuse field covariance matrix used in Wiener post-filtering to improve the source separation performance further. We demonstrate the utility of the proposed method via extensive objective and subjective evaluations.

Performance mechanisms and stability criteria for highly sub-wavelength, broadband nonreciprocal non-local acoustic metamaterials

Active acoustic metamaterials have received considerable attention recently due in large part to their ability to generate nonreciprocal wave transmission. We have shown previously how the spatial separation of acoustic sensors and sources can be leveraged to generate highly sub-wavelength, broadband nonreciprocal behavior in an acoustic waveguide. We showed that by transmitting pressure signals from one location along the waveguide to a loudspeaker at another location, we could achieve a peak isolation of 35 dB with a Q10dB factor of just over 0.25. Here, we discuss the mechanisms by which this nonreciprocity is achieved and the conditions required for the system to be stable. We show how conclusions regarding the performance and stability characteristics of an idealized plane wave acoustic model of our system must be adjusted when accounting for the controlling electronics and loudspeaker dynamics used in the experimental testing.Active acoustic metamaterials have received considerable attention recently due in large part to their ability to generate nonreciprocal wave transmission. We have shown previously how the spatial separation of acoustic sensors and sources can be leveraged to generate highly sub-wavelength, broadband nonreciprocal behavior in an acoustic waveguide. We showed that by transmitting pressure signals from one location along the waveguide to a loudspeaker at another location, we could achieve a peak isolation of 35 dB with a Q10dB factor of just over 0.25. Here, we discuss the mechanisms by which this nonreciprocity is achieved and the conditions required for the system to be stable. We show how conclusions regarding the performance and stability characteristics of an idealized plane wave acoustic model of our system must be adjusted when accounting for the controlling electronics and loudspeaker dynamics used in the experimental testing.

Integration of Neural Networks and Probabilistic Spatial Models for Acoustic Blind Source Separation

We formulate a generic framework for blind source separation (BSS), which allows integrating data-driven spectro-temporal methods, such as deep clustering and deep attractor networks, with physically motivated probabilistic spatial methods, such as complex angular central Gaussian mixture models. The integrated model exploits the complementary strengths of the two approaches to BSS: the strong modeling power of neural networks, which, however, is based on supervised learning, and the ease of unsupervised learning of the spatial mixture models whose few parameters can be estimated on as little as a single segment of a real mixture of speech. Experiments are carried out on both artificially mixed speech and true recordings of speech mixtures. The experiments verify that the integrated models consistently outperform the individual components. We further extend the models to cope with noisy, reverberant speech and introduce a cross-domain teacher–student training where the mixture model serves as the teacher to provide training targets for the student neural network.

A Novel Separation Technique of Flexural Loading-Induced Acoustic Emission Sources in Railway Prestressed Concrete Sleepers

This study determines the effectiveness and suitability of acoustic emission in order to monitor the structural health condition of railway pre-stressed concrete sleepers since the acoustic emission sensing provides effective means for detecting cracking events and quantifying the crack density. The aim of this study is two-fold. First, we establish a novel acoustic emission source separation that offers the possibility for analyzing the content of each source separately. This study is thus the first to demonstrate that these sources can be separated from each other by employing a classification algorithm based on decision trees. Hence, a technically sound decision can be made regarding the severity of the damage and appropriate measures can be taken. Second, the paper presents a new method for compressing the essential acoustic emission information of the loaded structure for further post-processing. Data compression of 70 % with respect to the full set of features is achieved through the elimination of those acoustic emission signatures that yield the largest mutual correlation with other signatures. The reduced set of acoustic emission signatures (only two uncorrelated features) still possesses the essential information of structural health of the structure of interest at a cost of a relatively small drop in classification accuracy from around 96% to 84%. The proposed method is validated experimentally on full-scale pre-stressed concrete sleepers under positive and negative bending moment configurations under static three-point bending tests, which correspond to anticipated failure modes at rail seats by industry.

Acoustic emission signal source separation for a flank wear estimation of drilling tools

The knowledge of the tool wear condition is essential for the dimensional accuracy of the workpiece. Commonly used systems for wear monitoring are usually based on the piezoelectric force measurement. However, in industry these systems are difficult to integrate. A suitable to integrate sensor type is the acoustic emission (AE) sensor. In this paper the main focus will be on the flank wear of drilling tools. For the investigation of the emitted frequency of the flank wear different analogy experiments needs to be realized. With the help of machine learning algorithms the recorded data will be classified.

UNISeC: Inspection, Separation, and Classification of Underwater Acoustic Noise Point Sources

Advancements in oceanic research have resulted in a plethora of activities such as undersea oil/gas exploration, environmental monitoring, sonar-based coastal surveillance, which have each increased the acoustic noise levels in the ocean and have raised concerns in the scientific community about the effect of human-generated sounds on marine life. Knowledge of the statistical characteristics of noise sources and their spatial distribution is paramount for understanding the impact on marine life as well as for regulating and policing such activities. Furthermore, studies have shown that assuming the underwater noise probability density function to be Gaussian, exponential, or Weibull is often not valid; therefore, statistically profiling the sources of the ambient noise is also essential to improve the performance of acoustic communication systems in the harsh underwater environment. In this paper, a novel solution based on the blind source separation method is proposed to enable separation of underwater acoustic noise point sources in the presence of channel propagation multipath. The proposed Underwater Noise Inspection, Separation, and Classification (UNISeC) system performs several pre- and postprocessing steps forming a novel gray-box model. Assuming there is no prior information on the noise sources, UNISeC estimates the number of such sources as well as characterizes and classifies them via a recursive pilot-aided probing method while minimizing the environmental acoustic contamination. A correlation-based characterization as well as power spectral density based classification approaches are investigated to verify the proposed method. Several scenarios are also presented and evaluated in detail via simulations.

Time difference of arrival-based acoustic source localization and separation technique for sparse arrays

An acoustic source localization and separation technique intended for sparse arrays is suggested in this paper. Unlike conventional beamforming approaches, the proposed technique is based on time difference of arrival (TDOA) information. This effectively eliminates the spatial aliasing problem due to overly large inter-element spacing of sensors. This method begins with estimation of the TDOAs between at least 4 sensors for 3-dimensional sound fields, with the aid of the generalized crosscorrelation-phase transformation (GCC-PHAT) algorithm and the subspace-based time delay estimation algorithm. Next, a constrained least squares (CLS) algorithm is applied to locate the source based on the above-estimated TDOAs. Once the source location is found, the source signal is extracted by using the Tikhonov regularization (TIKR) algorithm and minimum variance distortionless response (MVDR) beamformer. Regularization parameter is selected with a special search procedure to best preserve the audio quality in the inverse solution process. Experiments and listening tests are undertaken to validate the proposed TDOA-based localization and separation technique.

Localization and separation of acoustic sources by using a 2.5-dimensional circular microphone array.

Circular microphone arrays (CMAs) are sufficient in many immersive audio applications because azimuthal angles of sources are considered more important than the elevation angles in those occasions. However, the fact that CMAs do not resolve the elevation angle well can be a limitation for some applications which involves three-dimensional sound images. This paper proposes a 2.5-dimensional (2.5-D) CMA comprised of a CMA and a vertical logarithmic-spacing linear array (LLA) on the top. In the localization stage, two delay-and-sum beamformers are applied to the CMA and the LLA, respectively. The direction of arrival (DOA) is estimated from the product of two array output signals. In the separation stage, Tikhonov regularization and convex optimization are employed to extract the source amplitudes on the basis of the estimated DOA. The extracted signals from two arrays are further processed by the normalized least-mean-square algorithm with the internal iteration to yield the source signal with improved quality. To validate the 2.5-D CMA experimentally, a three-dimensionally printed circular array comprised of a 24-element CMA and an eight-element LLA is constructed. Objective perceptual evaluation of speech quality test and a subjective listening test are also undertaken.

An efficient equivariant adaptive separation via independence algorithm for acoustical source separation and identification

To balance the convergence rate and steady-state error of blind source separation (BSS) algorithms, an efficient equivariant adaptive separation via independence (Efficient EASI) algorithm is proposed based on separating indicator, which was derived from the convergence condition of EASI, and can be used to evaluate the separation degree of separated signals. Furthermore, a nonlinear monotone increasing function between suitable step sizes and separating indicator is constructed to adaptively adjust step sizes, and forgetting factor is employed to weaken effects of data at the initial stage. Numerical case studies and experimental studies on a test bed with shell structures are provided to validate the efficiency improvement of the proposed method. This study can benefit for vibration & acoustic monitoring and control, and machinery condition monitoring and fault diagnosis.

Nonlinear factor analysis and its application to acoustical source separation and identification

Acoustical signals of mechanical systems can provide original information of operating conditions, and thus benefit for machinery condition monitoring and fault diagnosis. However, acoustical signals measured by sensors are mixed signals of all the sources, and normally it is impossible to be directly used for acoustical source identification or feature extraction. Therefore, this paper presents nonlinear factor analysis (NLFA) and applies it to acoustical source separation and identification of mechanical systems. The effects by numbers of hidden neurons and mixed signals on separation performances of NLFA are comparatively studied. Furthermore, acoustical signals from a test bed with shell structures are separated and identified by NLFA and correlation analysis, and the effectiveness of NLFA on acoustical signals is validated by both numerical case studies and an experimental case study. This work can benefit for machinery noise monitoring, reduction and control, and also provide pure source information for machinery condition monitoring or fault diagnosis.

Wireless Sensor Array Network DoA Estimation from Compressed Array Data via Joint Sparse Representation.

A compressive sensing joint sparse representation direction of arrival estimation (CSJSR-DoA) approach is proposed for wireless sensor array networks (WSAN). By exploiting the joint spatial and spectral correlations of acoustic sensor array data, the CSJSR-DoA approach provides reliable DoA estimation using randomly-sampled acoustic sensor data. Since random sampling is performed at remote sensor arrays, less data need to be transmitted over lossy wireless channels to the fusion center (FC), and the expensive source coding operation at sensor nodes can be avoided. To investigate the spatial sparsity, an upper bound of the coherence of incoming sensor signals is derived assuming a linear sensor array configuration. This bound provides a theoretical constraint on the angular separation of acoustic sources to ensure the spatial sparsity of the received acoustic sensor array signals. The Cramr–Rao bound of the CSJSR-DoA estimator that quantifies the theoretical DoA estimation performance is also derived. The potential performance of the CSJSR-DoA approach is validated using both simulations and field experiments on a prototype WSAN platform. Compared to existing compressive sensing-based DoA estimation methods, the CSJSR-DoA approach shows significant performance improvement.

Localization and separation of acoustic sources by using a 2.5-dimensional circular microphone array

Localization and separation of acoustic sources by using a 2.5-dimensional circular microphone array

Blind Acoustic Source Separation Via System Identification for Leak Detection in Pipelines

The main motivation for this paper is to improve an acoustic leak detection system for pipelines by using blind source separation. In this setup hundreds of microphones are used to continuously monitor a pipeline. We propose to use a source separation scheme to eliminate overlapping sounds in the measured signals making is easier to detect and locate acoustic events in the measured data. To separate the sources, a large scale system identification problem results. In this paper we present one way that the identification problem can be made more computationally efficient. First, the blind source separation problem is parameterized as a channel estimation problem. Due to the presence of echoes, the channel impulse responses are very long, but are sparse in the sense that they are zero for a significant portion of the response. Then this sparsity is exploited for reducing the computational complexity of the identification problem. Our method is tested on a small scale test pipeline.

Underdetermined reverberant acoustic source separation using weighted full-rank nonnegative tensor models.

In this paper, a fusion of K models of full-rank weighted nonnegative tensor factor two-dimensional deconvolution (K-wNTF2D) is proposed to separate the acoustic sources that have been mixed in an underdetermined reverberant environment. The model is adapted in an unsupervised manner under the hybrid framework of the generalized expectation maximization and multiplicative update algorithms. The derivation of the algorithm and the development of proposed full-rank K-wNTF2D will be shown. The algorithm also encodes a set of variable sparsity parameters derived from Gibbs distribution into the K-wNTF2D model. This optimizes each sub-model in K-wNTF2D with the required sparsity to model the time-varying variances of the sources in the spectrogram. In addition, an initialization method is proposed to initialize the parameters in the K-wNTF2D. Experimental results on the underdetermined reverberant mixing environment have shown that the proposed algorithm is effective at separating the mixture with an average signal-to-distortion ratio of 3 dB.

The Doppler Effect based acoustic source separation for a wayside train bearing monitoring system

Wayside acoustic condition monitoring and fault diagnosis for train bearings depend on acquired acoustic signals, which consist of mixed signals from different train bearings with obvious Doppler distortion as well as background noises. This study proposes a novel scheme to overcome the difficulties, especially the multi-source problem in wayside acoustic diagnosis system. In the method, a time–frequency data fusion (TFDF) strategy is applied to weaken the Heisenberg׳s uncertainty limit for a signal׳s time–frequency distribution (TFD) of high resolution. Due to the Doppler Effect, the signals from different bearings have different time centers even with the same frequency. A Doppler feature matching search (DFMS) algorithm is then put forward to locate the time centers of different bearings in the TFD spectrogram. With the determined time centers, time–frequency filters (TFF) are designed with thresholds to separate the acoustic signals in the time–frequency domain. Then the inverse STFT (ISTFT) is taken and the signals are recovered and filtered aiming at each sound source. Subsequently, a dynamical resampling method is utilized to remove the Doppler Effect. Finally, accurate diagnosis for train bearing faults can be achieved by applying conventional spectrum analysis techniques to the resampled data. The performance of the proposed method is verified by both simulated and experimental cases. It shows that it is effective to detect and diagnose multiple defective bearings even though they produce multi-source acoustic signals.

Mixed-signal VLSI independent component analyzer for hearing aid applications.

We present a mixed-signal architecture for implementation of independent component analysis designed for the task of blind source separation of acoustic sources interfacing miniature microphone array. The matrix-vector multiplication is implemented through integration of switched current sources controlled by the pulse-width modulated signals. The proposed architecture implementing 3×3 static ICA in 0.5μm CMOS technology occupies chip area of 0.49 mm(2) with the power consumption of 80μW at 5 V supply voltage.

Deconvolution-based acoustic source localization and separation algorithms

In this paper, localization and separation of acoustic sources are examined. Depending on the number of sources in relation to the array channels, the problem is investigated in terms of underdetermined and overdetermined configurations. In the underdetermined configuration, virtual monopole sources are assumed in uniformly spaced angles. The problem is then formulated into compressive sampling (CS) problem which can be solved by using the linearly constrained -norm convex (CVX) optimization. The solution yields the directions of real sources and the source signal spectrum, which enables localization and reconstruction of sources at one shot. In the underdetermined configuration, source localization and signal separation is carried out in two steps. First, the directions of arrival (DOA) are estimated with Minimum Variance Distortionless Response (MVDR) or Multiple Signal Classification (MUSIC). Next, Tikhonov regularization (TIKR) is utilized to recover the source spectrum. In the localization problem for both configurations, Neyman-Pearson detector is employed to determine thresholds for source detection. Numerical and experimental results show that the proposed methods produce improved speech quality in terms of mean opinion score (MOS) in perceptual evaluation of speech quality (PESQ) test.

Acoustic source separation for the detection of coronary artery sounds

Coronary artery disease (CAD) is the leading cause of death in the United States, being responsible for more than 20% of all deaths in the country. This is in large part due to the difficulty of diagnostic screening for CAD. Phonoangiography seeks to detect CAD via the acoustic signature associated with turbulent flow near an abnormally constricted, or stenosed, region. However, the usefulness of the technique is severely hindered by the low strength of the CAD signal compared to the background noise within the chest. In this work, acoustic finite element analysis (FEA) was performed on physiologically accurate chest geometries to demonstrate the feasibility of an original acoustic source separation methodology for isolating coronary sounds. This approach is based upon pseudoinversion of mixing matrices determined through a combination of experiment and computation. This allows calculation of the sound emitted by the coronary arteries based upon measurements of the acoustic velocity on the chest surface. This work demonstrates the feasibility of such a technique computationally and examines the vulnerability of the proposed approach to measurement errors.