An optimized algorithm for binaural beamforming based on convex optimization relaxation technology

Abstract

The present invention provides a binaural beamforming optimization algorithm based on convex optimization relaxation technology. The method includes: receiving binaural beam signals; and optimizing the binaural beam signals by using a relaxed binaural beamforming optimization model. The present invention proposes a more efficient optimization algorithm, which can achieve a higher solution quality; and can simultaneously achieve noise reduction and binaural cue protection under the condition of lower computational complexity.

Description

Technical Field

[0001] The present invention relates to the technical field of acoustic signal processing, and particularly relates to a binaural beamforming optimization algorithm based on convex optimization relaxation technology. Background Art

[0002] Binaural beamforming technology can significantly improve the performance of digital hearing aids, which record the acoustic wave field through multiple microphones worn on both the left and right ears of the user. In order to achieve noise suppression to improve the clarity of the target signal, many beamforming algorithms have been proposed, such as binaural minimum variance distortionless response (BMVDR) beamforming. In addition, in order to enhance the user's sense of space or better perceive the sound field environment, it is necessary to keep the binaural cues of the target signal and the interference signal unchanged after filtering in a noisy acoustic environment. In the field of acoustics, the binaural cues referred to include: interaural level difference (ILD) and interaural phase difference (IPD), that is, the amplitude and phase of the interaural transfer function (ITF) respectively. The binaural linear constraint minimum variance (BLCMV) beamforming algorithm is precisely proposed to simultaneously protect the binaural cues of the target signal and the interference signal.

[0003] Considering that it is unrealistic to accurately estimate the sound field transfer function of each interfering sound source, a relaxed binaural beamformer is proposed to achieve a certain degree of robustness. It uses inequality constraints and equality constraints to protect the binaural cues of the interfering sources and the target source in the sound scene respectively, and the goal is to minimize the output noise power. In addition to the challenge of finding a good trade-off among all these goals, due to the limited computing power of digital hearing aid devices, the computational complexity should be kept as low as possible. The relaxed binaural beamforming (RBB) optimization problem is computationally non-convex and requires an effective algorithm to solve. For the RBB problem, a suboptimal method of successive convex optimization (SCO) is first proposed. This method usually needs to solve multiple convex optimization problems for each frequency point to converge, and the quality of the solution is affected by the iteration depth, resulting in a large computational complexity. The semidefinite convex relaxation (SDCR) method only solves a single convex optimization problem with a lower computational complexity, so its computational speed is much faster than that of the SCO method. Unfortunately, the interaural transfer function error of some interfering signals exceeds the user-defined threshold, that is, the solution obtained (non-rank-one solution) is not a feasible solution to the original problem. Although a hybrid method that comprehensively uses SCO and SDCR has been proposed, it is still a challenge to perfectly balance noise reduction and binaural cue protection at a lower complexity. Summary of the Invention

[0004] The object of the present invention is to provide a binaural beamforming optimization algorithm based on convex optimization relaxation technology. This method proposes a more efficient optimization algorithm, which can make the solution quality higher; and under the condition of lower computational complexity, it can simultaneously achieve noise reduction and binaural cue protection.

[0005] A binaural beamforming optimization algorithm based on convex optimization relaxation technology, comprising:

[0006] Receiving binaural beam signals;

[0007] Processing the binaural beam signals, specifically:

[0008] The arrays on both sides of the hearing aid each consist of M / 2 microphones, where M is an even number;

[0009] The sound field environment contains a target sound source signal located N meters directly in front of the user and K interfering sound source signals located at different horizontal angles on the same circle. The signals received by the microphone array at a certain time-frequency point are

[0010] where s is the target sound source signal, vk is the k-th interfering sound source signal, a ∈ C M is the acoustic transfer function vector of the target sound source signal, b k ∈ C M is the acoustic transfer function vector of the k-th interfering sound source signal, n is the background noise, C is the complex number field, C M represents the M-dimensional complex vector space;

[0011] Assume that the statistical components including the target signal, interference signal, and background noise are independent of each other. Then, the cross-power spectral density matrix CPSDM can be obtained: P = E[yy H = P s + P i + P n ∈ C M×M , the interference-plus-noise cross-power spectral density matrix is P i+n = P i + P n ∈ C M×M , where P s = p s aa H and p s = E[|s| 2 are the cross-power spectral density matrix of the received multi-channel target signal and the power spectral density of the target signal, respectively; are the sum of the cross-power spectral density matrices of all received multi-channel interference signals and the power spectral density of the k-th interference signal, respectively; P n = E[nn H is the cross-power spectral density matrix of the noise. The cross-power spectral density matrix is a statistic that characterizes the correlation between multi-channel microphone signals; (·) H denotes the conjugate transpose;

[0012] Optimize the binaural beam signal using the relaxed binaural beamforming optimization model.

[0013] Optimizing the binaural beam signal using the relaxed binaural beamforming optimization model includes:

[0014] Transform the binaural beam optimization problem into a convex optimization problem;

[0015] Use the semidefinite convex relaxation SDCR technique to obtain a rank-one solution to the convex optimization problem;

[0016] If there is no rank-one solution, generate a suboptimal solution using a Gaussian random process as the solution to the convex optimization problem;

[0017] If a suboptimal solution cannot be generated through the Gaussian random process, construct the convex optimization problem as a second-order cone programming SOCP problem for iterative solution.

[0018] Before optimizing the binaural beam signal, it also includes constructing a relaxed binaural beamforming optimization model, specifically:

[0019] Construct the equality constraint:

[0020]

[0021]

[0022] Construct the inequality constraint:

[0023]

[0024]

[0025] w L ,w R ∈C M They are the binaural beamforming weight vectors of the hearing aids on the left and right sides of the user respectively. Filtering the received microphone signals respectively can obtain the enhanced target signals on both the left and right ears; a L ,a R They are the acoustic transfer functions at the reference positions on the left and right sides respectively; b kL ,b kR They are the acoustic transfer functions of the k-th interference signal at the reference positions on the left and right sides respectively; the left reference position can be the first microphone of the left microphone array, and the right reference position can be the last microphone of the right microphone array; the error between the input and output interaural transfer functions of the k-th interference signal The input-output interaural transfer function of binaural minimum distortionless response beamforming The inequality ensures that the error between the input and output interaural transfer functions of the k-th interference signal does not exceed c corresponding to its binaural minimum distortionless response beamforming k times, that is, ε k ; where c k is the relaxation factor.

[0026] Using the semi-definite convex relaxation SDCR technology to obtain the rank-one solution of the convex optimization problem includes:

[0027] Let Then a new binaural beamformer w can be obtained; transforming the problem through the semi-definite convex relaxation SDCR technology

[0028]

[0029]

[0030] tr(Q k W)≤0,k = 1,...,K

[0031]

[0032] Among them, is the Kronecker product, \(W\in\mathbb{C}\) 2M×2M is an auxiliary variable, \(Q\) k is an intermediate variable. If there exists a rank-one solution, i.e., \(W\) * = \(w\) * \(w\) *H , then the solution of the relaxed binaural beamforming optimization model is obtained.

[0033] If there is no rank-one solution, a sub-optimal solution is generated by using a Gaussian random process as the solution of the convex optimization problem, including:

[0034] If there is no rank-one solution, i.e., \(W\) * \(\neq\) \(w\) * \(w\) *H , a sub-optimal solution that satisfies the inequality constraint is generated by using a Gaussian random process as the solution of the relaxed binaural beamforming optimization model. Assuming there is a solution \((W\) * , \(w\) * ), the feasible solution of the relaxed binaural beamforming optimization model is:

[0035]

[0036] Among them, \(I\in\mathbb{C}\) M×M is the identity matrix, the random variable \(z\) follows a multivariate complex Gaussian distribution \(\mathcal{N}\) C (0, \(W\) * - \(w\) * \(w\) *H ), that is \(z\) is a random error vector used to correct the solution of the SDCR problem Satisfying the binaural beam optimization problem, \(z\) can be split into \(z\) L , \(z\) R two parts, corresponding respectively to

[0037]

[0038] If a sub-optimal solution cannot be generated by using a Gaussian random process, the convex optimization problem is constructed as a second-order cone programming (SOCP) problem for iterative solution, including:

[0039] Construct a second-order cone programming (SOCP) problem for iterative solution:

[0040]

[0041]

[0042]

[0043] wherein, δ k = 1 / (ε k |b kR |), which means taking the real part of the complex number, is the iterative solution of the (l - 1)-th step.

[0044] A binaural beamforming optimization system based on convex optimization relaxation technology, comprising:

[0045] A signal receiving module, configured to receive binaural beam signals;

[0046] A data processing module, configured to optimize the binaural beam signals by using a relaxed binaural beamforming optimization model.

[0047] A computer device, comprising a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the above-mentioned binaural beamforming optimization algorithm based on convex optimization relaxation technology is implemented.

[0048] A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned binaural beamforming optimization algorithm based on convex optimization relaxation technology is implemented.

[0049] The present invention proposes a binaural beamforming optimization algorithm based on convex optimization relaxation technology. This algorithm comprehensively utilizes three algorithms: semi-definite convex relaxation, Gaussian random process, and second-order cone programming. On the premise of ensuring the quality of the solution, the algorithm performance is greatly improved: a sub-optimal solution is generated through the Gaussian random process, avoiding the solution of the original problem through second-order cone programming as much as possible, greatly reducing the computational complexity, and improving the system processing performance. Convex approximation is also performed through second-order cone programming. After a finite number of iterations, the algorithm will eventually converge to the local optimal solution, and thus achieve an optimal trade-off between noise suppression and binaural cue protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings here are incorporated into the specification and form a part of this specification, marking the embodiments conforming to the present invention, and are used together with the specification to explain the principles of the present invention.

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0052] Figure 1 is the flowchart of the present invention;

[0053] Figure 2This is the schematic diagram of the digital hearing aid of the present invention. Specific embodiments

[0054] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0055] It should be noted that all directional indications (such as up, down, left, right, front, back...) in the embodiments of the present invention are only used to explain the relative position relationship and movement conditions between components in a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indications will also change accordingly.

[0056] In addition, the descriptions involving "first", "second", etc. in the present invention are only for descriptive purposes, and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In addition, the technical solutions between various embodiments can be combined with each other, but it must be based on the ability of those of ordinary skill in the art to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the protection scope required by the present invention.

[0057] Existing algorithms either have low computational efficiency or the solutions obtained may have infeasible solutions (that is, the input-output binaural cue distortion error of the interference signal exceeds the threshold set by the user). The present invention can make the solution quality higher, and can simultaneously achieve noise reduction and binaural cue protection under the condition of lower computational complexity.

[0058] Embodiment 1

[0059] A binaural beamforming optimization algorithm based on convex optimization relaxation technology, referring to Figure 1 , includes:

[0060] S100 receives binaural beam signals;

[0061] Process the binaural beam signals, specifically:

[0062] The arrays on both the left and right sides of the hearing aid each consist of M / 2 microphones, where M is an even number;

[0063] The sound field environment contains a target sound source signal located N meters directly in front of the user and K interference sound source signals located at different horizontal angles on the same circle. The signals received by the microphone array at a certain time-frequency point are

[0064] In the formula, s is the target sound source signal, v k is the k-th interfering sound source signal, a ∈ C M is the acoustic transfer function vector of the target sound source signal, b k ∈ C M is the acoustic transfer function vector of the k-th interfering sound source signal, n is the background noise, C is the complex number field, C M represents the M-dimensional complex vector space;

[0065] Assuming that the statistical components including the target signal, the interference signal and the background noise are independent of each other, then the cross-power spectral density matrix CPSDM can be obtained: P = E[yy H = P s + P i + P n ∈ C M×M , the interference-plus-noise cross-power spectral density matrix is P i+n = P i + P n ∈ C M×M , where, P s = p s aa H and p s = E[|s| 2 are the cross-power spectral density matrix of the received multi-channel target signal and the power spectral density of the target signal respectively; p vk = E[|v k | 2 are the sum of the cross-power spectral density matrices of all received multi-channel interference signals and the power spectral density of the k-th interference signal respectively; P n = E[nn H is the cross-power spectral density matrix of the noise, and the cross-power spectral density matrix is a statistic characterizing the correlation between multi-channel microphone signals; (·) H represents the conjugate transpose;

[0066] S200 optimizes the binaural beam signal by using a relaxed binaural beamforming optimization model.

[0067] Digital hearing aids can exchange the signals captured by the left and right microphone arrays through a wireless link. These signals generally contain the position information of each sound source signal in the sound field environment. After designing a binaural beamformer to process the microphone signals, not only can the clarity of the target speech be improved, but also the binaural cues of the sound source can be protected, enhancing the sense of space of the hearing-impaired, refer to Figure 2 .

[0068] The S200 optimizes the binaural beam signal by using a relaxed binaural beamforming optimization model, which includes:

[0069] S210 transforms the binaural beam optimization problem into a convex optimization problem;

[0070] S220 uses the semidefinite convex relaxation SDCR technique to obtain a rank-one solution to the convex optimization problem;

[0071] S230, if there is no rank-one solution, uses a Gaussian random process to generate a suboptimal solution as the solution to the convex optimization problem;

[0072] S240, if a suboptimal solution cannot be generated through the Gaussian random process, constructs the convex optimization problem as a second-order cone programming SOCP problem for iterative solution.

[0073] Before S200 optimizes the binaural beam signal, it also includes constructing a relaxed binaural beamforming optimization model, specifically:

[0074] Construct the equality constraint:

[0075]

[0076]

[0077] Construct the inequality constraint:

[0078]

[0079]

[0080] w L ,w R ∈C M are the binaural beamforming weight vectors of the hearing aids on the left and right sides of the user respectively. Filtering the received microphone signals respectively can obtain the enhanced target signals on both the left and right ears; a L ,a R are the acoustic transfer functions at the reference positions on the left and right sides respectively; b kL ,b kR are the acoustic transfer functions of the k-th interfering signal at the reference positions on the left and right sides respectively. The left reference position can be the first microphone of the left microphone array, and the right reference position can be the last microphone of the right microphone array; the error between the input and output inter-aural transfer functions of the k-th interfering signal The input-output inter-aural transfer function of the binaural minimum distortionless response beamforming The inequality ensures that the error between the input and output inter-aural transfer functions of the k-th interfering signal does not exceed c k times its corresponding value for the binaural minimum distortionless response beamforming, that is, ε k; where c k is the relaxation factor.

[0081] Establish a relaxed binaural beamforming optimization model. First, protect the binaural cues of the target signal through equality constraints; second, considering the case of estimation errors in the acoustic transfer function vectors of interference signals in practical applications, in order to improve the robustness of the system, protect the binaural cues of interference signals through inequality constraints, that is, make the absolute value of the interaural transfer function error between input and output not exceed the user-preset threshold; finally, the objective function is to minimize the output noise power.

[0082] S220 uses the semi-definite convex relaxation SDCR technique to obtain a rank-one solution to the convex optimization problem, including:

[0083] Convert the non-convex problem with inequality constraints into a convex optimization problem for solution, let Convert the problem through the semi-definite convex relaxation SDCR technique

[0084]

[0085]

[0086] tr(Q k W) ≤ 0, k = 1,..., K

[0087]

[0088] where is the Kronecker product, W ∈ C 2M×2M is the auxiliary variable. If there exists a rank-one solution, that is, W * = w * w *H , then the solution of the relaxed binaural beamforming optimization model is obtained.

[0089] S230 If there is no rank-one solution, use the Gaussian random process to generate a sub-optimal solution as the solution to the convex optimization problem, including:

[0090] If there is no rank-one solution, that is, W * ≠ w * w *H , generate a sub-optimal solution that satisfies the inequality constraints through the Gaussian random process as the solution of the relaxed binaural beamforming optimization model. Assume there is a solution (W * , w * ), then the feasible solution of the relaxed binaural beamforming optimization model is

[0091]

[0092] where I ∈ C M×Mis the identity matrix, and the random variable z follows a multivariate complex Gaussian distribution Ν C (0, W * -w * w *H ), that is z is a random error vector used to correct the solution of the SDCR problem To satisfy the binaural beam optimization problem, z can be split into z L , z R two parts, respectively corresponding to

[0093] If the suboptimal solution cannot be generated by the Gaussian random process, the convex optimization problem is constructed as a second-order cone programming SOCP problem for iterative solution, including:

[0094] Construct a second-order cone programming SOCP problem for iterative solution:

[0095]

[0096]

[0097]

[0098] where δ k = 1 / (ε k |b kR |), means taking the real part of the complex number, is the iterative solution of the (l - 1)-th step.

[0099] The non-convex inequality constraint is converted into a cone constraint, so the convex problem can still be solved by the CVX tool; the convex approximation is carried out by using the alternating optimization technique, and then the optimal approximate solution of the original problem can be obtained. First, the solution of MVDR is used as the initial input to solve the SOCP problem, and then the optimal solution of each time is used as the input of the next time in turn, and the solution process is repeated until the algorithm converges; after a finite number of iterative solutions for each frequency point, the algorithm will converge to the local optimal solution At this point, each step of the solution is always feasible for the relaxed binaural beamforming optimization model.

[0100] Example 2

[0101] A binaural beamforming optimization system based on the convex optimization relaxation technique, including:

[0102] A signal receiving module for receiving binaural beam signals;

[0103] A data processing module for optimizing binaural beam signals by using a relaxed binaural beamforming optimization model.

[0104] Example 3

[0105] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements any one of the binaural beamforming optimization algorithms based on convex optimization relaxation technology.

[0106] Embodiment 4

[0107] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements any one of the binaural beamforming optimization algorithms based on convex optimization relaxation technology.

[0108] The present invention proposes a binaural beamforming optimization algorithm based on convex optimization relaxation technology. This algorithm comprehensively utilizes three algorithms: semi-definite convex relaxation, Gaussian random process, and second-order cone programming, and greatly improves the algorithm performance while ensuring the quality of the solution: generating a sub-optimal solution through the Gaussian random process, avoiding the solution of the original problem through second-order cone programming as much as possible, greatly reducing the computational complexity, and improving the system processing performance. It also performs convex approximation through second-order cone programming, and after a finite number of iterations, the algorithm will finally converge at the local optimal solution, thereby achieving an optimal trade-off between noise suppression and binaural cue protection.

[0109] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather will conform to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. An optimized binaural beamforming algorithm based on convex optimization relaxation technology, characterized in that, Comprising: Receiving a binaural beam signal; Processing the binaural beam signal, specifically: The arrays on both the left and right sides of the hearing aid each consist of M / 2 microphones, where M is an even number; The sound field environment contains a target sound source signal located N meters directly in front of the user and K interfering sound source signals located at different horizontal angles on the same circle. The signal received by the microphone array at a certain time-frequency point is ; In the formula, is the target sound source signal, is the k-th interfering sound source signal, is the acoustic transfer function vector of the target sound source signal, is the acoustic transfer function vector of the k-th interfering sound source signal, and n is the background noise, is the complex domain, and C M represents an M-dimensional complex vector space; Assuming that the statistical components including the target signal, interference signal, and background noise are mutually independent, the cross-power spectral density matrix CPSDM can be obtained: , and the cross-power spectral density matrix of interference plus noise is , where and are the cross-power spectral density matrix of the received multi-channel target signals and the power spectral density of the target signal, respectively; , are the sum of the cross-power spectral density matrices of all received multi-channel interference signals and the power spectral density of the k-th interference signal, respectively; is the cross-power spectral density matrix of the noise. The cross-power spectral density matrix is a statistic that characterizes the correlation between multi-channel microphone signals; represents the conjugate transpose; Optimizing the binaural beam signal using a relaxed binaural beamforming optimization model; The optimizing the binaural beam signal using a relaxed binaural beamforming optimization model includes: Converting the binaural beam optimization problem into a convex optimization problem; Using the semi-definite convex relaxation SDCR technique to obtain a rank-one solution to the convex optimization problem; If there is no rank-one solution, using a Gaussian random process to generate a sub-optimal solution as the solution to the convex optimization problem; If a sub-optimal solution cannot be generated through the Gaussian random process, constructing the convex optimization problem as a second-order cone programming SOCP problem for iterative solution.

2. The binaural beamforming optimization algorithm based on the convex optimization relaxation technique according to claim 1, wherein Before optimizing the binaural beam signal, it further includes constructing a relaxed binaural beamforming optimization model, specifically: Constructing equality constraints: ； ； Constructing inequality constraints: ； ； are the binaural beamforming weight vectors on the left and right sides of the user for the hearing device, respectively. Filtering the received microphone signals respectively can obtain enhanced target signals on both the left and right ears; are the acoustic transfer functions at the left and right reference positions respectively; are the acoustic transfer functions of the k-th interference signal at the left and right reference positions respectively. The left reference position can be the first microphone of the left microphone array, and the right reference position can be the last microphone of the right microphone array; the error between the input and output interaural transfer functions of the k-th interference signal , the input-output interaural transfer function of the binaural minimum distortionless response beamforming ; the inequality ensures that the error between the input and output interaural transfer functions of the k-th interference signal does not exceed its corresponding one for the binaural minimum distortionless response beamforming times, that is ; where is the relaxation factor.

3. An optimization algorithm for binaural beamforming based on convex optimization relaxation technology according to claim 2, characterized in that, The using the semi-definite convex relaxation SDCR technique to obtain a rank-one solution to the convex optimization problem includes: Let , then a new binaural beamformer can be obtained ; transform the problem through the semi-definite convex relaxation SDCR technique ； ； ； ； Among them, , , is the Kronecker product, is the auxiliary variable, and Q k is the intermediate variable. If there exists a rank-one solution, i.e., , then the solution of the relaxed binaural beamforming optimization model is obtained.

4. An optimization algorithm for binaural beamforming based on convex optimization relaxation technology according to claim 2, characterized in that The if there is no rank-one solution, using a Gaussian random process to generate a sub-optimal solution as the solution to the convex optimization problem includes: If there is no rank-one solution, that is , a suboptimal solution that satisfies the inequality constraints is generated through a Gaussian random process as the solution to the relaxed binaural beamforming optimization model. Assuming there is a solution , the feasible solution to the relaxed binaural beamforming optimization model is: ; Among them, is the identity matrix, and the random variable follows a multivariate complex Gaussian distribution , that is , is the random error vector, which is used to correct the solution of the SDCR problem and satisfies the binaural beamforming optimization problem, can be split into two parts, corresponding to respectively.

5. The optimized binaural beamforming algorithm based on the convex optimization relaxation technique according to claim 2, characterized in that The if a sub-optimal solution cannot be generated through the Gaussian random process, constructing the convex optimization problem as a second-order cone programming SOCP problem for iterative solution includes: Constructing a second-order cone programming SOCP problem for iterative solution: ； ； ； Among them, , means taking the real part of a complex number, is the iteration solution of the step.

6. A binaural beamforming optimization system based on convex optimization relaxation technology, which is applied to a binaural beamforming optimization algorithm based on convex optimization relaxation technology according to any one of claims 1-5, and is characterized in that Comprising: A signal receiving module for receiving a binaural beam signal; A data processing module for optimizing the binaural beam signal using a relaxed binaural beamforming optimization model.

7. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements a binaural beamforming optimization algorithm based on convex optimization relaxation technology according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a binaural beamforming optimization algorithm based on convex optimization relaxation technology according to any one of claims 1 to 5.

Images