Microphone array sound source identification method based on generalized sidelobe cancellation and linear discriminant analysis joint optimization
By combining a microphone array with a joint optimization method of generalized sidelobe cancellation and linear discriminant analysis, the problems of training data dependence and recognition difficulties in complex environments for outdoor multi-sound source recognition are solved, and efficient multi-sound source recognition and interference suppression are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-11
- Publication Date
- 2026-03-13
AI Technical Summary
In outdoor long-distance sound source identification tasks, the sound source signals to be identified are weak and are polluted by concurrent interference sources and environmental noise. Existing technologies are difficult to effectively handle multi-sound source identification, especially when the recognition performance degrades due to insufficient training data and complex acoustic environments.
A joint optimization method based on generalized sidelobe cancellation and linear discriminant analysis is adopted. Sound source enhancement and identification are performed through a microphone array. A feedback-sensing generalized sidelobe cancellation structure is constructed for parallel processing. The projection matrix is optimized by combining linear discriminant analysis to achieve effective identification of multiple sound sources.
In outdoor scenarios with limited training data, it significantly improves the accuracy and robustness of multi-sound source recognition, reduces the dependence on large-scale training data, effectively suppresses interference noise and retains key recognition features, and is suitable for complex acoustic environments.
Smart Images
Figure CN121656969A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of microphone array signal enhancement and sound source identification technology, and in particular to a microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis. Background Technology
[0002] Sound source recognition (SSR) technology is widely used in scenarios such as intelligent security and environmental monitoring. Deep learning-based sound source recognition methods have achieved remarkable results in ideal environments with high signal-to-interference-plus-noise ratios. However, in outdoor long-distance sound source recognition tasks, the sound source signals to be identified are relatively weak and are often severely contaminated by concurrent interference sources and environmental noise, leading to a sharp decline in recognition performance.
[0003] A common approach is to introduce a Sound Source Enhancement (SSE) module as a preprocessor before recognition to improve recognition robustness. Common SSE techniques include statistical models, acoustic beamforming, and neural network-based methods. However, numerous studies have shown that independently optimized SSE modules cannot guarantee improved recognition performance. This is mainly due to a mismatch between the optimization objectives of the sound source enhancement and recognition modules, which may lead to the suppression or removal of crucial feature information during the enhancement process. Therefore, joint optimization of the sound source enhancement and recognition modules has become a key strategy for improving recognition performance under low signal-to-noise ratio conditions. The literature (C.-C. Lee, Y. Tsao, H.-M. Wang, et al. D4AM: A general denoising framework for downstream acoustic models[C].Proc. ICLR, Kigali, Rwanda, 2023: 1–17.) discloses a joint optimization framework D4AM, which theoretically can alleviate the above problems. However, due to the use of large-scale neural networks to implement SSE and SSR modules, it has obvious limitations in outdoor multi-source sound recognition scenarios: First, it heavily relies on the scale and quality of training data, making it difficult to obtain sufficient training samples in outdoor scenarios where field experiments are costly and available public datasets are scarce; second, it only uses a single-channel processing framework, which cannot effectively enhance the target sound source and suppress interfering sound sources using spatial information; and third, it mainly targets the enhancement and recognition of a single sound source, making it difficult to handle multiple concurrent sound sources with random arrival directions.
[0004] In summary, for complex outdoor acoustic environments, a joint optimization framework based on microphone arrays needs to be constructed. Its core is to reduce the dependence on training data and to effectively identify multiple concurrent sound sources with random sound sources. Summary of the Invention
[0005] The purpose of this invention is to provide a microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis.
[0006] The technical solution to achieve the purpose of this invention is: a microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis, comprising the following steps:
[0007] Step 1: Acquire microphone array signals, estimate the direction of arrival for multiple concurrent sound sources, and perform time delay alignment preprocessing for each sound source;
[0008] Step 2: Using the time-delay aligned array signal from Step 1, enhance each sound source through a parallel feedback sensing generalized sidelobe cancellation structure; the feedback sensing generalized sidelobe cancellation structure introduces a joint optimization matrix into the lower branch of the generalized sidelobe canceller to receive feedback from the recognition module.
[0009] Step 3: Input the enhanced signal obtained in Step 2 into the linear discriminant analysis module, and calculate the projection matrix for sound source identification by solving an optimization problem that maximizes inter-class divergence and minimizes intra-class divergence;
[0010] Step 4: Based on the projection matrix described in Step 3, construct an optimization objective for updating the joint optimization matrix described in Step 2, and iteratively execute Step 2 to Step 4 alternately until the set convergence condition is met;
[0011] Step 5: In the identification stage, the sound source category of the test sample is determined by using the optimized joint optimization matrix and projection matrix.
[0012] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis.
[0013] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis.
[0014] A computer program product includes a computer program that, when executed by a processor, implements the aforementioned microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis.
[0015] Compared with existing technologies, the significant advantages of this invention are as follows: 1) This invention constructs a joint optimization framework for sound source enhancement and recognition based on a microphone array, which effectively solves the optimization target mismatch problem existing in traditional cascaded schemes, enabling the front-end enhancement module to retain the discriminative features key to recognition while suppressing interference noise; 2) The joint optimization method proposed in this invention adopts a fully linear framework, has a closed-form solution and clear physical meaning, significantly reduces the dependence on large-scale training data, and is suitable for outdoor sound source recognition scenarios with limited training samples; 3) This invention, through a parallel and innovative feedback-sensing generalized sidelobe cancellation structure, can simultaneously process multiple concurrent sound sources with random arrival directions, achieving efficient multi-source recognition in complex outdoor scenarios. Attached Figure Description
[0016] Figure 1 This is a flowchart of the microphone array sound source identification method based on the joint optimization of generalized sidelobe cancellation and linear discriminant analysis according to the present invention.
[0017] Figure 2 This is a flowchart illustrating the microphone array sound source identification method based on the joint optimization of generalized sidelobe cancellation and linear discriminant analysis according to the present invention.
[0018] Figure 3 This is a convergence characteristic curve during the training phase of the embodiment.
[0019] Figure 4 This is a box plot of the recognition performance when the arrival directions of the three concurrent sound sources are random, as shown in the example. Detailed Implementation
[0020] The present invention will now be further described with reference to the accompanying drawings.
[0021] This invention can achieve effective interference suppression and robust identification of multiple sound sources under limited training data. Based on array signal processing theory, this invention innovatively constructs a feedback-sensing generalized sidelobe cancellation structure and uses discriminant feedback indicators provided by linear discriminant analysis to guide its parameter optimization process, effectively solving the problem of mismatch between the optimization targets of front-end enhancement and back-end identification, while ensuring good recognition performance.
[0022] Combination Figure 1 and Figure 2 A microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis includes the following steps:
[0023] Step 1: Acquire array signals. For K sound sources (including C categories) in the acoustic scene, use a microphone array to acquire the target acoustic signals in the outdoor environment, and complete preprocessing tasks such as signal sampling, multi-source direction of arrival estimation, and time delay alignment.
[0024] Step 2: Construct feedback-sensing generalized sidelobe cancellation structures. Using the time-delay aligned array signal obtained in Step 1, parallel feedback-sensing generalized sidelobe cancellation structures are constructed for each sound source.
[0025] The feedback-sensing generalized sidelobe cancellation structure is implemented by introducing a joint optimization matrix at the blocking matrix output of a traditional generalized sidelobe canceller.
[0026] The output of the feedback sensing generalized sidelobe cancellation structure used to enhance the k-th sound source is: Where q is the fixed beamformer, B is the blocking matrix, G is the joint optimization matrix, and f k For an adaptive noise cancellation filter, x k This is the array signal after time delay alignment for the k-th sound source.
[0027] Step 3: Construct a linear discriminant analysis model. The enhanced signal s from Step 2... k The data is divided into segments of length L to form a training sample matrix. Let represent the i-th training sample in the c-th sound source. The goal of linear discriminant analysis is to solve for the projection matrix V, which satisfies the following formula:
[0028]
[0029]
[0030]
[0031] Where S w S represents the within-class divergence. b N represents the inter-class divergence; C represents the total number of sound source categories, and N represents the total number of sound source categories. c μ represents the total number of training samples for the c-th type of sound source; c Let μ be the centroid of the c-th type of sound source sample, and μ be the centroid of all samples. The solution to the projection matrix V satisfies the generalized eigenvalue decomposition problem, and is a matrix V. The matrix formed by the eigenvectors corresponding to the Q largest eigenvalues.
[0032] The projection matrix of the linear discriminant analysis is determined by maximizing the ratio of inter-class divergence to intra-class divergence for all sound source categories.
[0033] Step 4: Joint Optimization Parameter Update. Based on the projection matrix described in Step 3, construct an optimization objective for updating the joint optimization matrix. The corresponding cost function and the solution to the joint optimization matrix G are:
[0034]
[0035]
[0036] in This represents the array signal of the i-th training sample from the c-th sound source before enhancement. f k The generalized inverse of . Steps 2 through 4 are executed iteratively until the convergence condition shown in the following equation is met:
[0037]
[0038] in Here, p represents the cost function of the algorithm described in this invention; p represents the number of iterations. .
[0039] In step 4, the update process of the joint optimization matrix aims to minimize the intra-class divergence of the projected enhanced signal. The iterative method involves alternately updating the joint optimization matrix and the projection matrix, and the convergence of the joint optimization training process is determined by monitoring changes in the cost function.
[0040] Step 5: Identification Stage Classification. After the parameters have been alternately optimized to convergence, the matrices G and V are fixed, and the test sample X is classified. test Calculate its Euclidean distance from the centroid of each category in the projection space. The test sample is assigned to the category of the nearest centroid.
[0041] The present invention will be further described below with reference to embodiments:
[0042] Example
[0043] Combination Figure 1 and Figure 2 A microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis includes the following steps:
[0044] Step 1: Acquire Array Signals. Acquire target acoustic signals from the outdoor environment using a microphone array, and perform preprocessing tasks such as signal sampling and time delay alignment. In this embodiment, a 4-element planar array with a side length of 2 meters is used, with a sampling frequency of 4 kHz. The targets are categorized into three sound source types: helicopters, tracked vehicles, and quadcopter drones. The arrival directions of the sound sources are randomly set to ensure that the azimuth angle between any two sound sources is greater than 30°.
[0045] Step 2: Construct a feedback-sensing generalized sidelobe cancellation structure. Using the time-delay aligned array signal obtained in Step 1, construct a parallel feedback-sensing generalized sidelobe cancellation structure for each sound source, where the fixed beamforming q = [0.25, 0.25, 0.25, 0.25]. T The blocking matrix is in standard difference form.
[0046] Step 3: Construct a linear discriminant analysis model. Divide the enhanced signal from Step 2 into segments of 0.5s length, L = 2000, to form a training sample matrix. Use linear discriminant analysis to solve for the two-dimensional projection matrix V.
[0047] Step 4: Joint Optimization Parameter Update. Based on the projection matrix V described in Step 3, update the joint optimization matrix G, and perform alternating optimization until the cost function converges. The convergence characteristics are as follows: Figure 3 As shown, after about 36 iterations, the cost function value decreased to 0.1113 and tended to stabilize.
[0048] Step 5: Classification in the Recognition Stage. Under three concurrent sound source scenarios, 10,000 Monte Carlo trials were performed. In each trial, three test samples were randomly selected from the three sound source test sets, each belonging to a different category. Three azimuth angles with an interval of more than 30° were randomly selected to construct the microphone array signal for one random trial. In the 10,000 trials with random arrival directions from the three concurrent sound sources, the mean values of recognition accuracy, precision, recall, and F1 score were 0.9033, 0.9131, 0.9033, and 0.9041, respectively. The recognition performance is as follows: Figure 4 As shown in the box plot.
[0049] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis, characterized in that, Includes the following steps: Step 1: Acquire microphone array signals, estimate the direction of arrival for multiple concurrent sound sources, and perform time delay alignment preprocessing for each sound source; Step 2: Using the time-delay aligned array signal from Step 1, enhance each sound source through a parallel feedback sensing generalized sidelobe cancellation structure; the feedback sensing generalized sidelobe cancellation structure introduces a joint optimization matrix into the lower branch of the generalized sidelobe canceller to receive feedback from the recognition module. Step 3: Input the enhanced signal obtained in Step 2 into the linear discriminant analysis module, and calculate the projection matrix for sound source identification by solving an optimization problem that maximizes inter-class divergence and minimizes intra-class divergence; Step 4: Based on the projection matrix described in Step 3, construct an optimization objective for updating the joint optimization matrix described in Step 2, and iteratively execute Step 2 to Step 4 alternately until the set convergence condition is met; Step 5: In the identification stage, the sound source category of the test sample is determined by using the optimized joint optimization matrix and projection matrix.
2. The microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis according to claim 1, characterized in that, In step 1, for K sound sources in the acoustic scene, the target acoustic signal in the outdoor environment is acquired using a microphone array, and signal sampling, multi-source arrival direction estimation, and time delay alignment are completed.
3. The microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis according to claim 1, characterized in that, In step 2, the output of the feedback sensing generalized sidelobe cancellation structure used to enhance the k-th sound source is: Where q is the fixed beamformer, B is the blocking matrix, G is the joint optimization matrix, and f k For an adaptive noise cancellation filter, x k This is the array signal after time delay alignment for the k-th sound source.
4. The microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis according to claim 3, characterized in that, Step 3 specifically involves: enhancing the signal s from step 2. k The data is divided into segments of length L to form a training sample matrix. This represents the i-th training sample from the c-th sound source; the goal of linear discriminant analysis is to solve for the projection matrix V, which satisfies the following formula: ； ； ； Where S w S represents the within-class divergence. b N represents the inter-class divergence; C represents the total number of sound source categories, and N represents the total number of sound source categories. c This represents the total number of training samples for the c-th type of sound source; Let μ be the centroid of the c-th type of sound source sample, and μ be the centroid of all samples; the solution of the projection matrix V satisfies the generalized eigenvalue decomposition problem, and is a matrix V. The matrix formed by the eigenvectors corresponding to the Q largest eigenvalues.
5. The microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis according to claim 1, characterized in that, The update process of the joint optimization matrix in step 4 aims to minimize the intra-class divergence of the projected enhanced signal. The iterative method updates the joint optimization matrix and the projection matrix alternately, and determines whether the joint optimization training process has converged by monitoring the change of the cost function.
6. The microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis according to claim 4, characterized in that, Step 4 specifically involves: Based on the projection matrix described in step 3, an optimization objective for updating the joint optimization matrix is constructed, and the corresponding cost function and the solution of the joint optimization matrix G are: ； ； in This represents the array signal of the i-th training sample from the c-th sound source before enhancement. f k The generalized inverse; Steps 2 through 4 are executed iteratively until the convergence condition shown in the following equation is met: ； in Here, p represents the cost function of the algorithm described in this invention; p represents the number of iterations. .
7. The microphone array sound source identification method based on joint optimization of generalized sidelobe cancellation and linear discriminant analysis according to claim 6, characterized in that, Step 5 specifically involves: after the parameters have converged through alternating optimization, fixing matrices G and V, and then testing the test sample X. test Calculate its Euclidean distance from the centroid of each category in the projection space. The test sample is assigned to the category of the nearest centroid.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1-7.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-7.
10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-7.