Echo Cancellation Research Articles

Echo cancellation using Artificial Intelligence technique

To improve the audio quality in telecommunication, echo cancellation was proposed to prevent signals from re-appearing. Generally speaking, the echo canceler first recognizes the original transmitted signal that appears again due to delay in the transmitted or recieved signal. Then, the echo canceler subtracts the echo in the signals. Commonly, echo cancelers are calledacoustic echo cancellation (AEC) in general, whether they aim for acoustic echo, line echo, or both. However, in modern times, new problems have been emerging which require more innovative, more advanced methods to tackle the echo issue. Thus, the paper reviews methods regarding Artificial Intelligence (AI) to improve the current means for echo cancellation. In this paper, we mainly discuss the approaches to echo cancellation, the identification of echo using adaptive AI and the usage of adaptive filters and neural network for echo cancellation. For instance, deep learning approaches can facilitate the issues of AEC (active echo control) and ANC (active noise control),the optimization of LMS algorithm can be achieved by combining Swarm Intelligence (SI), combining adaptive digital filter and recurrent neural network for acoustic echo cancellation etc. Using AI in echo cancellation can help us achieve high-quality, full-duplex telecommunication in modern telephony system.

Velocity-Controlled Kalman Filter for an Improved Echo Cancellation with Continuously Moving Microphones

Velocity-Controlled Kalman Filter for an Improved Echo Cancellation with Continuously Moving Microphones

Novel TransQT Neural Network: A Deep Learning Framework for Acoustic Echo Cancellation in Noisy Double-Talk Scenario

Novel TransQT Neural Network: A Deep Learning Framework for Acoustic Echo Cancellation in Noisy Double-Talk Scenario

Erratum to “End-to-End Deep Learning-Based Adaptation Control for Linear Acoustic Echo Cancellation”

Erratum to “End-to-End Deep Learning-Based Adaptation Control for Linear Acoustic Echo Cancellation”

Stereophonic acoustic echo cancellation based on semi-blind source separation

The non-uniqueness problem embedded in stereophonic acoustic echo cancellation causes performance degradation. Usually, an additional decorrelation preprocessor is required, which is inevitably detrimental to the speech quality. In this paper, we propose a stereophonic acoustic echo cancellation method based on semi-blind source separation. By optimizing a dedicated demixing matrix rooted from stereophonic acoustic echo cancellation, the proposed method is insensitive to the inter-channel correlation between the far-end signals, resulting in better convergence behavior without any decorrelation preprocessing. Experimental results demonstrate the efficacy and robustness of the proposed method in the presence of inter-channel correlation, echo path changes, and double-talk.

On Semi-Blind Source Separation-Based Approaches to Nonlinear Echo Cancellation Based on Bilinear Alternating Optimization

On Semi-Blind Source Separation-Based Approaches to Nonlinear Echo Cancellation Based on Bilinear Alternating Optimization

End-to-End Deep Learning-Based Adaptation Control for Linear Acoustic Echo Cancellation

End-to-End Deep Learning-Based Adaptation Control for Linear Acoustic Echo Cancellation

DsPIC Implementation and Performances Evaluation of Adaptive Filters for System Identification

This paper aims to implement and evaluate performances of adaptive filters applied to linear systems identification on a dsPIC platform. To do so, we considered two different applications: the identification of DC motor transfer function and the acoustic echo cancellation. For the DC motor case, real-time identification is not necessarily required. However, for the acoustic echo cancellation case, real-time identification is crucial, making execution time a key parameter in this scenario. We have implemented three adaptive algorithms: LMS, NLMS, and RLS. The first two are known for their computational simplicity, while RLS is recognized for its convergence speed. Performances evaluation is primarily based on accuracy and computation speed. A comparison is carried out in both cases to determine which algorithm is more suitable for each application. The results obtained showed that RLS is better suited for the DC motor case, while NLMS is more suitable for acoustic echo cancellation, particularly when the impulse response of the echo path is long.

Robust adaptive algorithm for widely-linear Hammerstein system and its application

The Hammerstein model is a crucial model frequently employed in the field of signal processing. This paper focuses on the widely-linear (WL) complex-valued adaptive filtering and proposes a WL Hammerstein system model. Firstly, a cost function which minimizes the complex correntropic loss criterion is constructed. Subsequently, by extending the alternate direction method of multipliers (ADMM) optimization and non-convex projection method to the WL model, and leveraging the special structure of the autocorrelation matrix in the WL Hammerstein system, a robust recursive adaptive filtering algorithm for the WL Hammerstein system is proposed, along with its simplified version. More importantly, the performance analysis is studied, in which the convergence performance and the mean-square performance are provided. Simulations about system identification and stereophonic acoustic echo cancelation (SAEC) are carried out to validate the theoretical analysis and demonstrate the superior performance of the algorithms in this paper.

Temporal Convolutional Network for Acoustic Echo Cancellation in Double-Talk Scenarios

Temporal Convolutional Network for Acoustic Echo Cancellation in Double-Talk Scenarios

Robust Normalized Subband Adaptive Filter Algorithm Based on Logarithmic and Total Least Squares for Correlated Input Signals and Impulsive Noise

The total least squares normalized subband adaptive filter (TLS-NSAF) algorithm proposed in recent years has shown excellent performance in processing the errors-in-variables (EIV) model of correlated input signals. However, when the system is Interferenced by the impulsive noise, the convergence of TLS-NSAF algorithm will be seriously deteriorated. To address this problem, this paper improves the TLS-NSAF algorithm by using the logarithmic function, and proposes a robust NSAF algorithm based on logarithmic and total least squares method (RNSAF-LTLS). The algorithm performs well in an environment where input signal is a correlated input signal and output signal contains impulsive noise. In addition, there are few studies in the existing literature that apply the TLS method to subbands, and there are few performance analyses associated with it. To address this problem, this paper analyzes the local stability, derives the step size that guarantees the stability of the RNSAF-LTLS algorithm, and calculates the steady-state mean squared deviation (S-MSD). Finally, the RNSAF-LTLS algorithm is verified in system identification (SI) and acoustic echo cancellation (AEC) applications. The simulation results prove the superiority of the proposed algorithm and the correctness of the theoretical analysis.

An improved proportionate arctangent framework based LMS/F algorithm for system identification

This paper proposes an improved proportionate arctangent framework that relies on a least mean square/fourth (IPALMS/F) algorithm for the physical sparse system identification problem. The ALMS/F algorithm has significant robustness against impulsive noise, whereas the improved proportionate concept when utilized with the ALMS/F takes advantage of the sparse characteristic to increase convergence time. Finally, the IPALMS/F algorithm is implemented using a recursive adaptive sparse exponential functional link neural network (RASETFLN) nonlinear filter and is named the RASETFLN-IPALMS/F algorithm which resulted in enhanced performance compared to other existing filters in terms of robustness in an impulsive environment, convergence rate, and steady-state error for system identification and acoustic echo cancellation.

An acoustic echo canceller optimized for hands-free speech telecommunication in large vehicle cabins

Acoustic echo cancelation (AEC) is a system identification problem that has been addressed by various techniques and most commonly by normalized least mean square (NLMS) adaptive algorithms. However, performing a successful AEC in large commercial vehicles has proved complicated due to the size and challenging variations in the acoustic characteristics of their cabins. Here, we present a wideband fully linear time domain NLMS algorithm for AEC that is enhanced by a statistical double-talk detector (DTD) and a voice activity detector (VAD). The proposed solution was tested in four main Volvo truck models, with various cabin geometries, using standard Swedish hearing-in-noise (HINT) sentences in the presence and absence of engine noise. The results show that the proposed solution achieves a high echo return loss enhancement (ERLE) of at least 25 dB with a fast convergence time, fulfilling ITU G.168 requirements. The presented solution was particularly developed to provide a practical compromise between accuracy and computational cost to allow its real-time implementation on commercial digital signal processors (DSPs). A real-time implementation of the solution was coded in C on an ARM Cortex M-7 DSP. The algorithmic latency was measured at less than 26 ms for processing each 50-ms buffer indicating the computational feasibility of the proposed solution for real-time implementation on common DSPs and embedded systems with limited computational and memory resources. MATLAB source codes and related audio files are made available online for reference and further development.

A variable step size total least squares affine-projection-like algorithm: Formula derivation and performance analysis

In the errors-in-variables (EIV) model, the steady state characteristics of the affine-projection-like (APL) algorithm are poor. Hence, the total least squares APL (TLS-APL) algorithm different from the bias-compensated is proposed, which can reduce the adverse influence of input and output noise on weight update. Then, the local extremum of the TLS-APL algorithm is verified, that is, the expected parameter is the optimal value of the iterative estimation. Moreover, the mean, mean square, and steady-state mean square deviation (MSD) performance of the proposed TLS-APL algorithm are deduced by using the Kronecker product and vec operator, respectively. To balance the convergence characteristics and steady-state characteristics, the step size optimization of the TLS-APL algorithm, namely the variable step size TLS-APL (VSS-TLS-APL) algorithm, is derived from the analysis of MSD. Considering the existence of gradient errors, the fitting effects of theoretical steady-state MSD curves and actual steady-state MSD curves under different parameters are analyzed separately, which proves the rationality of theoretical derivation. In addition, the VSS-TLS-APL algorithm with optimized step size is compared with the TLS-APL algorithm with fixed step size in the application of system identification, and the effect of the variable step size (VSS) is verified. The experiments of system identification and acoustic echo cancelation prove that the VSS-TLS-APL algorithm is superior to the existing known algorithms.

An Iterative Wiener Filter Based on a Fourth-Order Tensor Decomposition

This work focuses on linear system identification problems in the framework of the Wiener filter. Specifically, it addresses the challenging identification of systems characterized by impulse responses of long length, which poses significant difficulties due to the existence of large parameter space. The proposed solution targets a dimensionality reduction of the problem by involving the decomposition of a fourth-order tensor, using low-rank approximations in conjunction with the nearest Kronecker product. In addition, the rank of the tensor is controlled and limited to a known value without involving any approximation technique. The final estimate is obtained based on a combination of four (shorter) optimal filters, which are alternatively iterated. As a result, the designed iterative Wiener filter outperforms the traditional counterpart, being more robust to the accuracy of the statistics’ estimates and/or noisy conditions. In addition, simulations performed in the context of acoustic echo cancellation indicate that the proposed iterative Wiener filter that exploits this fourth-order tensor decomposition achieves better performance as compared to some previously developed solutions based on lower decomposition levels. This study could further lead to the development of computationally efficient tensor-based adaptive filtering algorithms.

Recursive least-squares algorithm based on a third-order tensor decomposition for low-rank system identification

A recently developed third-order tensor (TOT) decomposition-based method has proved to be working very well in linear system identification problems that target the estimation of long length impulse responses. This technique exploits the decomposition of the impulse response based on the nearest Kronecker product and low-rank approximations, but it also owns a specific advantage. Its main feature is related to the way of handling the rank of a third-order tensor, which is controlled in terms of the rank of a matrix and, as a result, is limited to small values. In this paper, we further develop a recursive least-squares (RLS) algorithm based on the TOT decomposition technique. The resulting solution combines the estimates provided by three shorter adaptive filters, thus gaining in terms of both performance and complexity. The three component filters are updated in parallel, which represents an important practical advantage, in terms of the modularity of implementation. Also, as compared to the conventional RLS algorithm, a significant reduction of the computational complexity can be achieved for the common setup of the decomposition parameters. Besides, since the length of the adaptive filter highly influences the main performance criteria (e.g., convergence, tracking, and accuracy), the proposed RLS version based on the TOT decomposition outperforms both the conventional algorithm and a previously developed RLS-based solution that involves a second-order decomposition. These performance gains mainly reflect in a faster tracking capability and a better accuracy of the estimate (i.e., lower misalignment). Simulations performed in the framework of echo cancellation support the performance features of the proposed tensorial RLS-based algorithm.

A Multi-Stage Acoustic Echo Cancellation Model Based on Adaptive Filters and Deep Neural Networks

The presence of a large amount of echoes significantly impairs the quality and intelligibility of speech during communication. To address this issue, numerous studies and models have been conducted to cancel echo. In this study, we propose a multi-stage acoustic echo cancellation model that utilizes an adaptive filter and a deep neural network. Our model consists of two parts: the Speex algorithm for canceling linear echo, and the multi-scale time-frequency UNet (MSTFUNet) for further echo cancellation. The Speex algorithm takes the far-end reference speech and the near-end microphone signal as inputs, and outputs the signal after linear echo cancellation. MSTFUNet takes the spectra of the far-end reference speech, the near-end microphone signal, and the output of Speex as inputs, and generates the estimated near-end speech spectrum as output. To enhance the performance of the Speex algorithm, we conduct delay estimation and compensation to the far-end reference speech. For MSTFUNet, we employ multi-scale time-frequency processing to extract information from the input spectrum. Additionally, we incorporate an improved time-frequency self-attention to capture time-frequency information. Furthermore, we introduce channel time-frequency attention to alleviate information loss during downsampling and upsampling. In our experiments, we evaluate the performance of our proposed model on both our test set and the blind test set of the Acoustic Echo Cancellation challenge. Our proposed model exhibits superior performance in terms of acoustic echo cancellation and noise reverberation suppression compared to other models.

Sparse Bayesian Least-Mean-Squares Algorithm

Echo seriously damages the speech communication quality of vehicle communication system and seriously interferes with the normal operation of voice navigation, voice dialing, and other functions of the vehicle-mounted human-machine interaction platform. This brief proposes a sparse Bayesian least-mean-squares (SBLMS) algorithm which is designed to exploit the sparseness of echo channel for improving the quality of echo cancellation. The proposed SBLMS algorithm is derived by performing Bayesian inference in an equivalent linear Gaussian observation model with diagonal proportionate matrix. To avoid algorithmic performance degradation at steady-state, a step-size lower bound is derived for the proposed SBLMS algorithm. Also, a reset parameter method is designed to improve the tracking performance of the proposed SBLMS algorithm in a sudden change scenario. Finally, the good performance of the proposed SBLMS algorithm is confirmed by system identification and acoustic echo cancellation.

Hybrid metaheuristic method of ABC kernel filtering for nonlinear acoustic echo cancellation

This research work uses nonlinear kernel adaptive filtering to eliminate echo, a type of voice that is undesirable and communicated over the microphone. Adaptive filters enable the detection of time-varying potentials and the tracking of the signal’s dynamic changes. In this work, a hybrid metaheuristic technique that combines the Artificial Bee Colony (ABC) with the Kernel Adaptive Improved Proportionate and Normalized Least Mean Square (KIPNLMS) filter is developed to significantly boost echo cancellation efficiency. The unique adaptive filter based on sub-filter and proportionate adaptation is combined in the advanced technique, which presents an improved Non-linear Acoustic Echo Cancellation (NAEC) framework. An essential part of kernel techniques is that they translate the information on to a high-dimensional feature space, where linear filtering is performed. In the initial space, the arithmetic operations are carried out by evaluating inner products among pairs of input patterns, also known as kernels. Along with the ABC step size regulation technique, kernel-based adaptive filtering improves Signal to Noise Ratio (SNR) and accelerates convergence. The algorithm's effectiveness is assessed through simulation using signal Echo Return Loss Enhancement (ERLE). The performance of Kernel LMS, Kernal Normalized LMS (KNLMS), Kernal Proportionate NLMS, and KIPNLMS algorithms are compared. In contrast to typical kernel adaptive filters, the suggested ABC-KIPNLMS increases the ERLE more than 10 dB compared to other algorithms.

Target Measurement in Radio Frequency Simulation Using Code-Interrupted Transmitting and Receiving of Pulse Signals

The interrupted transmitting and receiving (ITR) has been utilized in radio frequency simulation to conduct target measurement with pulse signal. Because the ITR results in partly missing of pulse signal, the spectrum aliasing occurs and the target range profile is difficult to be obtained. Therefore, frequency domain filtering and sparse reconstruction are proposed to recover the target range profile. However, frequency domain filtering is ineffective when wideband signal is utilized. The calculation of sparse reconstruction is complex and the reconstruction performance may be poor for some cases of ITR parameters. To solve this problem, the code-interrupted transmitting and receiving (code-ITR) of pulse signal is proposed in this paper. Firstly, the ITR sequences is coded according the duty ratio of each sub-pulse. Secondly, the echo combination and cancellation are processed to eliminate the false peaks in the range profile of ITR echo. By using the code-ITR and echo processing method, the range profile can be obtained for target measurement. Simulations and experiments are performed to verify the effectiveness of the proposed method.