Vehicle speaker sound field optimization method, device, equipment and storage medium

Abstract

The present application relates to the technical field of sound field optimization, and discloses a sound field optimization method, device, equipment and storage medium for vehicle-mounted loudspeakers. The method comprises: measuring parameters of vehicle-mounted loudspeakers to obtain loudspeaker parameter data and frequency response curves; establishing a digital model of the in-vehicle acoustic environment; creating a virtual loudspeaker array configuration scheme and generating multiple groups of in-vehicle sound field distribution training data; training through a differential neural network to obtain an in-vehicle sound field mapping model; optimizing the sound field for multiple target listening points to obtain MIMO filter coefficients; performing adaptive sound field compensation according to the MIMO filter coefficients and in-vehicle sound field data monitored in real time to obtain target sound field compensation parameters, and generating a real-time optimized in-vehicle sound field according to the target sound field compensation parameters, thereby achieving rapid response and compensation to changes in the in-vehicle sound field, and solving the problem of poor adaptability of traditional fixed parameter optimization methods.

Description

Technical Field

[0001] The present application relates to the technical field of sound field optimization, and in particular to a method, device, equipment and storage medium for optimizing the sound field of a vehicle-mounted speaker. Background Art

[0002] As an important entertainment and information device inside the car, the sound quality of the car audio system directly affects the driving experience of passengers. However, the complexity of the in-car environment poses a huge challenge to the realization of high-quality sound field. Factors such as small cabin space, multiple reflective surfaces, and fixed passenger positions lead to problems such as uneven sound field distribution, uneven frequency response, and inaccurate stereo image positioning.

[0003] Traditional in-car audio optimization methods mainly rely on empirical tuning and fixed equalizer settings, which are difficult to adapt to changes in different car models and riding conditions. Although the introduction of digital signal processing technology has improved the flexibility of sound field optimization, its effect is still limited due to the lack of accurate modeling and real-time adaptation capabilities of the in-car acoustic environment. Summary of the invention

[0004] The present application provides a method, device, equipment and storage medium for optimizing the sound field of a vehicle-mounted speaker, thereby achieving rapid response and compensation for changes in the sound field inside the vehicle, and solving the problem of poor adaptability of traditional fixed parameter optimization methods.

[0005] A first aspect of the present application provides a method for optimizing the sound field of a vehicle-mounted speaker, the method comprising:

[0006] Measure the parameters of the vehicle speakers to obtain the speaker parameter data and frequency response curve;

[0007] Establishing a digital model of the in-vehicle acoustic environment according to the speaker parameter data and the frequency response curve;

[0008] Creating a virtual speaker array configuration scheme based on the in-vehicle acoustic environment digital model, and generating multiple sets of in-vehicle sound field distribution training data according to the virtual speaker array configuration scheme;

[0009] Inputting the plurality of sets of in-vehicle sound field distribution training data into a differential neural network for training to obtain an in-vehicle sound field mapping model;

[0010] Based on the in-car sound field mapping model, the sound field is optimized for multiple target listening points to obtain MIMO filter coefficients;

[0011] Adaptive sound field compensation is performed according to the MIMO filter coefficients and the real-time monitored in-vehicle sound field data to obtain target sound field compensation parameters, and a real-time optimized in-vehicle sound field is generated according to the target sound field compensation parameters.

[0012] A second aspect of the present application provides a vehicle-mounted speaker sound field optimization device, the vehicle-mounted speaker sound field optimization device comprising:

[0013] The measurement module is used to measure the parameters of the vehicle-mounted speaker and obtain the speaker parameter data and frequency response curve;

[0014] An establishing module, used for establishing a digital model of the in-vehicle acoustic environment according to the speaker parameter data and the frequency response curve;

[0015] A creation module, used to create a virtual speaker array configuration scheme based on the in-vehicle acoustic environment digital model, and generate multiple groups of in-vehicle sound field distribution training data according to the virtual speaker array configuration scheme;

[0016] A training module, used for inputting the plurality of sets of in-vehicle sound field distribution training data into a differential neural network for training to obtain an in-vehicle sound field mapping model;

[0017] An optimization module, configured to optimize the sound field for multiple target listening points based on the in-vehicle sound field mapping model to obtain MIMO filter coefficients;

[0018] The compensation module is used to perform adaptive sound field compensation according to the MIMO filter coefficients and the real-time monitored in-vehicle sound field data, obtain target sound field compensation parameters, and generate a real-time optimized in-vehicle sound field according to the target sound field compensation parameters.

[0019] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned vehicle speaker sound field optimization method.

[0020] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned vehicle-mounted speaker sound field optimization method.

[0021] Compared with the prior art, the present application has the following beneficial effects: by accurately measuring the parameters of the vehicle speakers and establishing a digital model of the in-vehicle acoustic environment, accurate basic data is provided for subsequent optimization, which is conducive to improving the accuracy and pertinence of the optimization. The virtual speaker array technology and differential neural network training are used to achieve accurate mapping and prediction of the in-vehicle sound field, overcoming the problem of the difficulty of traditional methods in modeling complex acoustic environments. By introducing multi-objective optimization and particle swarm algorithm, the global optimal MIMO filter coefficients are obtained based on multiple factors such as frequency response flatness, sound pressure level uniformity and stereo image accuracy, thereby improving the overall effect of sound field optimization. Combined with real-time monitoring and adaptive filtering technology, rapid response and compensation to changes in the in-vehicle sound field are achieved, solving the problem of poor adaptability of traditional fixed parameter optimization methods. Through technologies such as sub-band processing and dynamic range compression, while ensuring sound quality, the system's adaptability to different sound sources and environmental noise is improved, enhancing the consistency and comfort of the listening experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0023] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0024] Figure 1 is a flow chart of a method for optimizing the sound field of a vehicle-mounted speaker provided by an embodiment of the present invention;

[0025] Figure 2 is a schematic block diagram of the structure of a vehicle-mounted speaker sound field optimization device provided by an embodiment of the present invention;

[0026] Figure 3 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0029] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0030] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 , an embodiment of the vehicle speaker sound field optimization method in the embodiment of the present application includes:

[0031] Step 100: Measure the parameters of the vehicle-mounted speaker to obtain speaker parameter data and a frequency response curve;

[0032] It is understandable that the execution subject of the present application may be a vehicle-mounted speaker sound field optimization device, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0033] Specifically, a sweep signal with a frequency range of 20Hz to 20kHz is applied to the vehicle speaker. The sweep signal can cover the main frequency bands within the human hearing range, and the original response data of the speaker is measured by the signal. The original response data is subjected to a fast Fourier transform to convert the time domain signal into a frequency domain signal to obtain the frequency response data of the speaker. Based on the frequency response data, the frequency response curve of the speaker between 150Hz and 10kHz is extracted to reflect the response characteristics of the speaker in the medium, low and high frequency ranges. A 1kHz sine signal is applied to the vehicle speaker. Using the voltage and current measurement method, the impedance value of the speaker is calculated by measuring the voltage and current of the speaker at this frequency. Impedance is an important parameter of the speaker and can reveal its electrical performance at different frequencies. Based on the measured impedance data, the Q value, i.e., the quality factor, of the speaker is calculated. The Q value represents the ratio of the bandwidth of the speaker to the center frequency. A high Q value usually indicates that the speaker has a higher resonance peak and has a greater impact on the sound quality. In order to measure the distortion characteristics of the speaker, a multi-band signal with an input power of 1W is applied to the speaker. The output signal is analyzed using a harmonic analyzer to measure the total harmonic distortion. The distortion data can reflect the nonlinear distortion generated by the speaker under different input powers, which helps to evaluate the sound quality of the speaker. By applying 1W input power and measuring the sound pressure level at a distance of 1 meter from the speaker, the sound pressure level values ​​of the speaker at four frequencies of 0.6kHz, 0.8kHz, 1.0kHz and 1.25kHz are obtained. After averaging these values, the sensitivity data of the speaker is obtained. The sensitivity data indicates the sound pressure level that the speaker can output under standard input power, and is usually used to evaluate the efficiency of the speaker. In order to evaluate the power handling capacity of the speaker, a short-term high-power signal is applied to it. The maximum input power of the speaker is measured by a power analyzer to obtain its power handling capacity data, which helps to judge the durability and stability of the speaker under high power conditions and avoid damage caused by power overload in actual use. The frequency response data, impedance data, quality factor, distortion data, sensitivity data and power handling capacity data are normalized to construct the parameter data of the speaker.

[0034] Step 200: establishing a digital model of the in-vehicle acoustic environment according to the speaker parameter data and the frequency response curve;

[0035] Specifically, a 3D laser scan is performed on the interior space of the vehicle to obtain the 3D point cloud data of the interior of the vehicle, record the surface shapes and structural features of each surface in the car, and form accurate spatial information. The 3D point cloud data is imported into the computer-aided design (CAD) software for processing. In the CAD software, the point cloud data is refined and optimized to eliminate noise and irregular points, and an accurate 3D model of the interior of the car is generated to reflect the detailed structure of each surface in the car. The accurate 3D model of the interior of the car is meshed, and the complex car space is divided into multiple regular grid units, so that each unit can be analyzed independently, thereby simplifying the complexity of acoustic simulation calculations. Through meshing, a car interior space grid model composed of multiple small units is generated. According to the acoustic properties of the interior materials, the sound absorption coefficient and reflection coefficient are assigned to each grid unit in the interior space grid model. The sound absorption coefficient and reflection coefficient are key parameters in acoustic simulation, which are used to reflect the absorption and reflection characteristics of different materials to sound waves and form an initial acoustic characteristic model. The measured loudspeaker parameter data and frequency response curve are input into the initial acoustic characteristic model, and a virtual sound source is set at the corresponding position in the model to simulate the sound source characteristics of the loudspeaker, so that the initial acoustic characteristic model becomes an in-vehicle acoustic model containing a sound source, which can analyze the propagation characteristics of the sound waves in the car by calculation. Based on this acoustic model, the finite element analysis method is used to simulate the propagation of sound waves in the car. Finite element analysis is a numerical calculation method that gradually solves the propagation behavior of sound waves in grid units to obtain preliminary in-vehicle sound field distribution results. The sound field distribution results show the intensity, direction and attenuation of sound waves in the car space. In order to verify the accuracy of the simulation results, actual sound field measurements are performed at multiple target positions in the car. The measured sound field data at different positions are obtained by measuring equipment such as sound pressure meters or microphone arrays. The measured sound field data are compared with the preliminary sound field distribution results, and the error value between the two is calculated. The size of the error value directly reflects the matching degree between the initial model and the real in-vehicle acoustic environment. According to the error value, the in-vehicle acoustic model containing the sound source is iteratively optimized. During the optimization process, the sound absorption coefficient, reflection coefficient, and sound source parameters are adjusted to make the simulation results gradually approach the measured results. The error value is recalculated after each iteration, and the model parameters are further adjusted on this basis. This process continues until the error value is less than the preset threshold, and finally a digital model of the in-vehicle acoustic environment that conforms to the actual in-vehicle acoustic characteristics is obtained.

[0036] Step 300: creating a virtual speaker array configuration scheme based on the digital model of the in-vehicle acoustic environment, and generating multiple sets of in-vehicle sound field distribution training data according to the virtual speaker array configuration scheme;

[0037] It should be noted that multiple target listening points are defined in the digital model of the in-car acoustic environment to determine the position of the user sitting in the car. Usually, these target listening points include the driver's seat, the co-pilot seat, and the rear passenger seat. Through the setting, a set of coordinate sets of target listening points is obtained. Based on the target listening point coordinate set, the positions where the speakers can be installed in the car are initialized to obtain an initial virtual speaker array. The array includes speakers installed at multiple positions in the car. The initial goal is to maximize the coverage of all listening points and provide a basic speaker layout for subsequent sound field optimization. The initial virtual speaker array is input into the genetic algorithm for optimization. The genetic algorithm is a global search algorithm that simulates the natural evolution process and is suitable for finding a better configuration scheme. In the optimization process, genetic evolution is carried out with the objective function of maximizing the uniformity of the in-car sound pressure level and the total directivity factor of the system, and the configuration of the virtual speaker is continuously optimized through operations such as crossover, mutation, and selection. The uniformity of sound pressure level is the standard for evaluating the consistency of the distribution of the sound field in the car between various listening points, while the total directivity factor measures the sound wave output characteristics of the speaker array in different directions. Through the continuous iteration of the genetic algorithm, the optimal configuration scheme of a virtual speaker array is obtained, which can make the sound quality effect of each listening point balanced and improved. Based on the virtual speaker array configuration scheme, different frequency response characteristics and directivity parameters are assigned to each virtual speaker to simulate the speaker sound characteristics brought by different installation positions in the car, and a virtual speaker parameter set containing speaker frequency response and directivity is obtained. The virtual speaker parameter set is input into the digital model of the in-car acoustic environment, and the simulation calculation of the in-car sound field is performed to obtain preliminary sound field distribution data, showing the sound pressure level distribution at different target listening points under the current speaker configuration, reflecting the sound field performance of the speaker array in the car. In order to test the performance of the system under different usage conditions, the preliminary sound field distribution data is subjected to frequency scanning and volume level change processing. By changing the frequency and volume combination of the speaker output, multiple groups of sound field distribution data containing different frequencies and volumes are generated. Based on multiple groups of sound field distribution data, the cross-correlation between adjacent virtual speakers is calculated. Cross-correlation refers to the degree of mutual influence between two speakers in the sound field. If the cross-correlation of two speakers is too high, their contributions in the sound field will cancel each other out or produce excessive superposition, affecting the overall sound quality performance. Adjust the position or parameters of speakers with high cross-correlation to reduce mutual interference between them and obtain multiple sets of optimized sound field distribution data. In order to enhance the diversity and robustness of the training data, the optimized sound field distribution data is enhanced. By adding random noise and simulating changes in the in-car environment (such as opening and closing windows, changes in the position of objects in the car, etc.), richer training data is generated to obtain multiple sets of in-car sound field distribution training data.

[0038] Step 400: input multiple groups of in-vehicle sound field distribution training data into a differential neural network for training to obtain an in-vehicle sound field mapping model;

[0039] Specifically, multiple sets of in-vehicle sound field distribution training data are preprocessed, and the data are normalized using the Z-score standardization method to eliminate the influence of different scales of the data, so that it has a standard distribution with a mean of zero and a variance of 1. The wavelet threshold denoising algorithm is used to remove high-frequency noise in the training data, improve the purity and consistency of the data, and obtain a standardized training data set.

[0040] Based on the standardized training data set, the first differential neural network structure is constructed. The structure of the neural network includes an input layer, five differential hidden layers and an output layer. The number of nodes N in the input layer is equal to the number of virtual speakers, and the number of nodes M in the output layer is equal to the number of target listening points multiplied by three to cover multiple dimensional information such as sound pressure level and directionality of each listening point. In the five differential hidden layers, the initial number of nodes in each layer is set to 2N in order to provide sufficient nonlinear capabilities so that the network can fully fit the complex sound field distribution in the car. In the process of constructing the network, the weight W of each differential hidden layer node is initialized using the He initialization method. The specific formula is W=sqrt(2 / n in )*random.randn(n in ,n out ), where n in is the number of input nodes, n outis the number of output nodes. The He initialization method is suitable for deep neural networks with ReLU activation function, which can effectively alleviate the problem of gradient disappearance, thereby accelerating the convergence process of the network. The standardized training data set is divided into training set, validation set and test set in a ratio of 8:1:1. The training set is used to learn the model parameters, the validation set is used to evaluate the performance of the model during the training process to avoid overfitting, and the test set is used to evaluate the performance of the final model. The training set data is input into the differential neural network, and the Adam optimizer and the mean square error (MSE) loss function are used for preliminary training. During the training process, the learning rate is set to 0.001, the batch size is 64, and 100 epochs are trained to obtain the initial sound field mapping model. The Adam optimizer combines the advantages of momentum and adaptive learning rate, and is suitable for most deep learning tasks, so that the model can converge quickly and stably. After obtaining the initial sound field mapping model, in order to improve the efficiency and performance of the network, the contribution Cj of each hidden layer node is calculated based on the output results of the model. The calculation formula of contribution Cj is Cj=sum(|∂L / ∂Wj|) / sum(|Wj|), where L is the loss function and Wj is the weight of node j. The lower the contribution of a node, the smaller its impact on the overall output of the model. Pruning is used to reduce the complexity of the model. Nodes with contributions lower than the threshold of 0.01 are pruned to obtain the optimized network structure. In the network structure after pruning optimization, in order to improve the network's fitting ability and processing ability for complex acoustic scenes, the top 20% of the hidden layers in terms of contribution are identified, and new hidden layers are inserted between these layers to enhance the network's representation ability. The setting formula for the number of nodes Nnew in the newly added layer is Nnew=1.5*(Nprev+Nnext) / 2, where Nprev and Nnext are the number of nodes in two adjacent layers, respectively. This setting ensures that the newly added layer has an appropriate number of nodes to achieve more effective feature extraction and mapping. Select appropriate activation functions for the newly added nodes, use the orthogonal projection operator method, select the most relevant S-shaped function from {Sigmoid, Tanh, ReLU, LeakyReLU, ELU}, and the selection criterion is to minimize the projection error. Select the most appropriate activation function for different nodes to make full use of the advantages of different activation functions in different situations and improve the network's expressiveness and training efficiency. After completing the activation function selection, fine-tune the network weights and use L2 regularization to prevent overfitting to obtain the second differential neural network structure. L2 regularization effectively reduces the overfitting phenomenon of the model on complex data by applying a penalty term to the weights, making the model more concise and with stronger generalization ability. Re-input the standardized training data set into the optimized second differential neural network structure, and use the back propagation algorithm with early stopping strategy for iterative training.After each epoch, the mean square error (MSE) is calculated on the validation set. Training is stopped when the MSE changes less than 0.1% for five consecutive iterations or when the maximum number of iterations reaches 10,000. The early stopping strategy effectively prevents overtraining of the model, thereby reducing the risk of overfitting and making the model more robust in practical applications. The model performance is evaluated on the test set to determine the effect of the network in actual in-car sound field mapping and obtain the final in-car sound field mapping model.

[0041] Step 500: Optimize the sound field for multiple target listening points based on the in-vehicle sound field mapping model to obtain MIMO filter coefficients;

[0042] Specifically, the output of the in-car sound field mapping model is analyzed to extract the sound field characteristics of multiple target listening points, including frequency response, phase response and sound pressure level, to form a target listening point sound field feature set, which is used to describe the acoustic performance of each listening point at different frequencies. A multi-objective optimization function is constructed based on the sound field feature set of the target listening point, which contains three sub-objectives: frequency response flatness, sound pressure level uniformity and stereo image accuracy. The frequency response flatness aims to ensure that the output frequency of the in-car audio system at each listening point is balanced, avoid certain frequency components being too strong or too weak, and provide a more natural sound experience; sound pressure level uniformity requires that the sound pressure difference between each target listening point is as small as possible to ensure that all passengers can enjoy a consistent volume level; and the accuracy of the stereo image refers to the ability to obtain accurate sound source localization at different listening points, so that the sound of the left and right channels can truly present the sense of space hierarchy and direction. By comprehensively considering these three sub-objectives, a comprehensive optimization objective function is obtained. The comprehensive optimization objective function is input into the particle swarm optimization algorithm for solution. In the particle swarm optimization algorithm, each particle represents a set of candidate MIMO filter coefficients, and the particle swarm is initialized to obtain the initial particle swarm. The particle swarm optimization algorithm is a global search algorithm based on swarm intelligence, which is suitable for solving complex nonlinear optimization problems. By updating the position of particles in multidimensional space, the global optimal solution is effectively explored and approached. Each particle corresponds to a set of candidate filter coefficients, which represents the sound field optimization scheme. The goal of the particle swarm optimization algorithm is to find a set of filter coefficients that minimizes the comprehensive optimization objective function and achieve the best in-car sound field optimization. For each particle in the initial particle swarm, the in-car sound field mapping model is used to perform forward propagation calculations to obtain the predicted sound field distribution of each target listening point. The forward propagation process simulates the influence of the speaker on the in-car sound field under the current filter coefficient conditions, and predicts the sound pressure level and phase response characteristics of each listening point under different frequencies and input conditions. Based on the predicted sound field distribution data, the fitness value of each particle is calculated. The fitness value is used to measure the degree of compliance of the particle with the optimization objective function under the current conditions. The higher the fitness value, the better the effect of the group of filter coefficients in optimizing the sound field. According to the fitness value, the individual optimal solution of each particle and the global optimal solution of the entire particle swarm are updated to obtain the updated particle swarm. The individual optimal solution records the best position found by each particle in each iteration, while the global optimal solution is the best solution in the entire particle swarm. In order to improve the diversity of the particle swarm and prevent it from falling into the local optimal solution, the crossover and mutation operations are applied to the updated particle swarm to generate a new generation of particle swarm. The crossover operation combines the features of different particles to generate new particles with the excellent features of the parent generation, while the mutation operation increases the diversity of the group by randomly perturbing some particles to improve the ability to explore unknown areas.After generating a new generation of particle swarm, the steps of forward propagation calculation and fitness calculation are performed again. By repeatedly iterating this series of operations, the particle swarm is continuously updated to optimize the value of the comprehensive objective function. The whole process will continue until the preset number of iterations is reached or the convergence condition is met. The termination condition of the iterative process is set to the change in fitness value below a certain threshold or the maximum number of iterations is reached, so as to ensure that the particle swarm can find the best MIMO filter coefficients when computing resources allow.

[0043] Step 600 , adaptive sound field compensation is performed according to the MIMO filter coefficients and the real-time monitored in-vehicle sound field data to obtain target sound field compensation parameters, and a real-time optimized in-vehicle sound field is generated according to the target sound field compensation parameters.

[0044] Specifically, multiple MEMS microphones are installed at key positions in the car. MEMS microphones are small in size and highly sensitive, and are suitable for installation in the car for sound field monitoring. Through these microphones, real-time sound field data in the car is collected. The real-time sound field data is preprocessed, including denoising and spectrum analysis. The denoising process can eliminate random noise and interference signals in the environment, thereby ensuring the accuracy of the collected data, while the spectrum analysis is used to convert the sound field signal into the frequency domain, extract its frequency characteristics, and obtain real-time characteristic information of the sound field in the car. The MIMO filter coefficients and the real-time in-car sound field characteristics are input into the adaptive filter for processing, and the adaptive filter uses the minimum mean square error algorithm to update the filter coefficients in real time. The minimum mean square error algorithm is an adaptive algorithm that dynamically adjusts the filter parameters by minimizing the error between the output signal and the ideal target, so that the filter can always maintain the best compensation effect in the constantly changing in-car sound field environment. After adjustment by the LMS algorithm, the filter coefficients that are preliminarily updated are obtained, which are used to preliminarily adjust the sound field distribution in the car. Based on the preliminarily adjusted filter coefficients, the predicted sound field distribution of the target listening point is calculated, and compared with the ideal sound field distribution to obtain the sound field error vector. The sound field error vector represents the difference between the actual sound field and the ideal sound field, and is a quantitative description of the quality of the in-car sound field compensation. In order to minimize these differences, the sound field error vector is input into the Wiener filter for processing. The Wiener filter is a filter that can optimally estimate the noise in the signal. By processing the error vector, a compensation signal is obtained, which effectively reduces the sound field error at the target listening point. The compensation signal is decomposed in the frequency domain and decomposed into multiple sub-band signals. Each sub-band signal represents the acoustic characteristics of a different frequency band. By adjusting the gain and correcting the phase of each sub-band signal, the sound characteristics of each frequency band are finely controlled to ensure that the sound field is optimally optimized in each frequency range. After adjustment, an optimized sub-band compensation signal is obtained. All optimized sub-band compensation signals are synthesized to obtain an overall time domain compensation signal. The time domain compensation signal is superimposed with the original audio signal to obtain a compensated audio signal, which includes real-time compensation for the uneven characteristics in the in-car sound field to improve the overall listening effect. The compensated audio signal is subjected to dynamic range compression and frequency response equalization. Dynamic range compression is used to reduce the impact of instantaneous large peaks in the audio signal, thereby protecting the speakers and improving the clarity of the sound, while frequency response equalization is used to adjust the volume of each frequency band to make the sound more balanced and natural. After these two steps, the target sound field compensation parameters are obtained. The target sound field compensation parameters are applied to the digital signal processor of the car audio system to process the output audio signal in real time.The digital signal processor can apply compensation parameters in real time in an efficient manner, ensuring that the in-car sound field always remains in its best condition in a changing environment. The optimized in-car sound field can provide a more consistent and natural sound performance, allowing every passenger to experience high-quality sound effects no matter where they sit in the car.

[0045] In the embodiment of the present application, accurate basic data is provided for subsequent optimization by accurately measuring the parameters of the vehicle speakers and establishing a digital model of the in-vehicle acoustic environment, which is conducive to improving the accuracy and pertinence of the optimization. By using virtual speaker array technology and differential neural network training, accurate mapping and prediction of the in-vehicle sound field are achieved, overcoming the problem of traditional methods having difficulty in modeling complex acoustic environments. By introducing multi-objective optimization and particle swarm algorithm, the global optimal MIMO filter coefficients are obtained based on multiple factors such as frequency response flatness, sound pressure level uniformity, and stereo image accuracy, thereby improving the overall effect of sound field optimization. Combined with real-time monitoring and adaptive filtering technology, rapid response and compensation to changes in the in-vehicle sound field are achieved, solving the problem of poor adaptability of traditional fixed parameter optimization methods. Through technologies such as sub-band processing and dynamic range compression, while ensuring sound quality, the system's adaptability to different sound sources and environmental noise is improved, enhancing the consistency and comfort of the listening experience.

[0046] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0047] Apply a sweep frequency signal of 20 Hz-20 kHz to the vehicle speaker to obtain original response data, and perform fast Fourier transform on the original response data to obtain frequency response data;

[0048] Calculate the frequency response curve of the vehicle speaker in the range of 150Hz-10kHz based on the frequency response data;

[0049] Apply a 1kHz sinusoidal signal to the car speaker, calculate the impedance value of the speaker by voltage and current measurement method to obtain impedance data, and calculate the Q value of the car speaker based on the impedance data to obtain the quality factor;

[0050] Apply a multi-band signal with 1W input power to the car speaker, measure the total harmonic distortion with a harmonic analyzer, and obtain distortion data;

[0051] Under the condition of 1W / 1M, the sound pressure level of the vehicle speaker is measured, and the average value of the four frequency points of 0.6kHz, 0.8kHz, 1.0kHz and 1.25kHz is taken to obtain the sensitivity data;

[0052] Apply a short-term high-power signal to the car speaker, measure the maximum input power of the car speaker with a power analyzer, and obtain the power handling capacity data;

[0053] The frequency response data, impedance data, quality factor, distortion data, sensitivity data and power handling capacity data are normalized to construct the speaker parameter data.

[0054] Specifically, a frequency sweep signal with a frequency range of 20Hz to 20kHz is applied to the vehicle-mounted speaker. A frequency sweep signal is a signal with a gradually increasing frequency. By applying a frequency sweep signal, the response of the speaker in the full frequency range is comprehensively tested. During the frequency sweep process, the original response data of the speaker is recorded, including the change in sound pressure level and the corresponding phase response. The original time domain response data is subjected to a fast Fourier transform, the time domain signal is converted into a frequency domain signal, the amplitude and phase of each frequency component in the signal are extracted, and the frequency response data of the speaker is obtained. Through the frequency response data, the output characteristics of the speaker at different frequencies are analyzed, for example, determining at which frequency points there are resonance peaks or output attenuation. Based on the obtained frequency response data, the frequency response curve of the vehicle-mounted speaker in the frequency range of 150Hz to 10kHz is calculated. By analyzing the response curve in this frequency band, it is determined whether the frequency response of the speaker is flat enough, thereby determining its performance in daily music playback. For example, the frequency response curve calculation formula is used:

[0055] ;

[0056] in, Indicates the frequency The frequency response at is at frequency The output power at is the input power. By calculating , and obtain the frequency response curve in the range of 150Hz to 10kHz, so as to evaluate whether the output performance of the speaker at different frequencies meets the expected standards. A 1kHz sinusoidal signal is applied to the speaker, and the impedance value of the speaker is calculated by the voltage and current measurement method. In this process, the applied voltage is measured. and current , then use the impedance formula:

[0057] ;

[0058] in, represents the impedance of the speaker, is the voltage applied to the speaker, is the current flowing through the speaker. The quality factor (Q value) of the speaker is calculated based on the impedance data. The Q value is an important parameter that reflects the resonance characteristics of the speaker and is usually calculated using the following formula:

[0059] ;

[0060] in, represents the resonant frequency of the loudspeaker, and The corresponding impedance drops to the resonance peak The higher the quality factor, the more obvious the resonance of the speaker, which will affect the sound quality, especially at low frequencies, which can easily cause muddy sound. A multi-band signal with an input power of 1W is applied to the speaker, and the output signal is analyzed using a harmonic analyzer to measure the total harmonic distortion (THD). Harmonic distortion refers to the energy of multiple harmonic components contained in the speaker output, and is usually described by the following formula:

[0061] ;

[0062] in, is the amplitude of the fundamental frequency signal, and are the amplitudes of each harmonic. THD reflects the nonlinear distortion generated by the speaker when reproducing sound. The higher the distortion, the worse the sound quality. For the sensitivity test of the speaker, the sound pressure level is measured at a distance of 1 meter from the speaker with a power input of 1W. Four frequencies of 0.6kHz, 0.8kHz, 1.0kHz and 1.25kHz are selected, and their average sound pressure levels are calculated to obtain the sensitivity data of the speaker. Sensitivity reflects the sound pressure output capacity of the speaker under a given input power, usually in decibels (dB). The higher the sensitivity, the greater the sound output of the speaker under a smaller power. In order to evaluate the performance of the speaker under high power conditions, a short-term high-power signal is applied to the speaker, and its maximum withstand capacity is measured by a power analyzer. The data of the maximum input power is used to determine the maximum electrical power that the speaker can withstand in a short period of time without causing damage. The frequency response data, impedance data, quality factor, distortion data, sensitivity data and maximum power withstand capacity data are normalized to eliminate the differences in the order of magnitude of different physical quantities for unified analysis and comparison. The normalized data constitutes the parameter data set of the speaker.

[0063] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0064] Perform 3D laser scanning on the interior space of the vehicle to obtain 3D point cloud data, and then input the 3D point cloud data into computer-aided design software for processing to obtain an accurate 3D model of the interior of the vehicle;

[0065] Mesh the precise three-dimensional model of the interior of the vehicle to obtain a vehicle interior space mesh model consisting of multiple mesh units;

[0066] According to the acoustic characteristics of the materials inside the vehicle, the sound absorption coefficient and the reflection coefficient are assigned to each grid unit in the grid model of the vehicle interior space to obtain an initial acoustic characteristic model;

[0067] Inputting the loudspeaker parameter data and the frequency response curve into the initial acoustic characteristic model, setting a virtual sound source at a corresponding position, and obtaining an in-vehicle acoustic model including the sound source;

[0068] Conduct finite element analysis on the in-car acoustic model including the sound source, simulate the propagation characteristics of sound waves in the car, and obtain preliminary sound field distribution results;

[0069] Perform actual sound field measurements at multiple target locations in the vehicle to obtain measured sound field data, compare the measured sound field data with the preliminary sound field distribution results, and calculate the error value;

[0070] The in-vehicle acoustic model including the sound source is iteratively optimized based on the error value until the error value is less than a preset threshold, thereby obtaining a digital model of the in-vehicle acoustic environment.

[0071] Specifically, the interior space of the vehicle is subjected to a three-dimensional laser scan. 3D laser scanning is a method of obtaining geometric information of the surface of an object through a laser beam. The detailed geometric structure of the interior of the vehicle is obtained by scanning, and the result is presented in the form of three-dimensional point cloud data. The three-dimensional point cloud data consists of a large number of points with spatial coordinate (x, y, z) information, which describe the structure of the interior of the vehicle, including surface details such as seats, dashboards, and doors. The point cloud data is input into computer-aided design (CAD) software for processing, and the CAD software converts the original point cloud data into an operable three-dimensional geometric model. After processing, the generated accurate three-dimensional model of the interior of the vehicle can reflect the real geometric characteristics of the vehicle, and eliminates the noise and redundant points generated by reflection or occlusion during the scanning process. The accurate three-dimensional model of the interior of the vehicle is meshed. The complex geometry is divided into small, regular units, so that the acoustic characteristics inside each unit are easier to simulate and calculate. The grid unit is divided into small cubes or tetrahedrons, each unit representing a part of the vehicle compartment space. Through meshing, a grid model of the interior space of the vehicle composed of multiple grid units is obtained. According to the acoustic properties of the materials in each part of the car, the corresponding sound absorption coefficient and reflection coefficient are assigned to each grid unit in the grid model of the car space. The acoustic characteristic parameters depend on the physical properties of the material, such as the fabric of the seat, the plastic of the dashboard, the window glass, etc. Different materials have different absorption and reflection capabilities for sound waves. ) represents the ratio of the material to the incident sound energy, and its value range is 0 to 1, where 1 represents complete absorption and 0 represents complete reflection. The reflection coefficient (r) is negatively correlated with the sound absorption coefficient, indicating the degree to which the sound wave is reflected by the material. Determine the reflection coefficient. Thus, the initial acoustic characteristic model is obtained, which describes the reaction of each grid unit in the car to the sound wave and is the basis for acoustic simulation. The speaker parameter data and frequency response curve are input into the initial acoustic characteristic model, and a virtual sound source is set at the corresponding position in the car to simulate the sound radiation of the speaker in the car. The speaker parameters include its frequency response characteristics, directivity and sound pressure level, etc. These parameters determine the sound wave output characteristics of the speaker at different frequencies. By setting up a virtual sound source, the working state of the real speaker is simulated in the digital model to form an in-car acoustic model containing a sound source. Finite element analysis is performed on the in-car acoustic model containing a sound source to simulate the propagation characteristics of sound waves in the car. Finite element analysis is a numerical calculation method suitable for solving problems with complex boundary conditions and geometric shapes. By discretizing the wave equation of the sound field distribution into the solution of a finite number of grid units, the propagation path, reflection and absorption of sound waves in the car are simulated. The propagation behavior of sound waves is described by the wave equation:

[0072] ;

[0073] in, is the Laplace operator, which is used to describe the change of sound pressure in space. is the sound pressure at the spatial position and time The value at is the speed of sound. By solving this equation, the sound pressure distribution at different positions in the car is obtained, and finally the preliminary sound field distribution results are obtained, which describe the sound pressure level and phase of each area inside the car under the excitation of the speaker. In order to verify the accuracy of the simulation results, actual sound field measurements are performed at multiple target positions in the car, such as installing microphones at the ears of the driver, front passengers and rear passengers. Through measurement, the measured sound field data is obtained. The measured sound field data is compared with the preliminary sound field distribution results, and the error value between the two is calculated. The size of the error is calculated by the following formula:

[0074] ;

[0075] in, represents the error value, is the number of measurement points, and Respectively represent The simulated sound pressure and the measured sound pressure at each measurement point. The size of the error value directly reflects the accuracy of the model. The smaller the error, the closer the simulation result is to the actual situation. Based on the calculated error value, the in-vehicle acoustic model containing the sound source is iteratively optimized. By adjusting the acoustic parameters of the mesh model, the simulated sound field distribution is more consistent with the actual measured sound field data. For example, the sound absorption coefficient or reflection coefficient is adjusted to better simulate the acoustic properties of the material, or the output characteristics of the virtual sound source are readjusted according to the actual measurement data. After each optimization iteration, finite element analysis is performed again and a new error value is calculated. This process is repeated until the error value is less than the preset threshold, and finally an accurate digital model of the in-vehicle acoustic environment is obtained.

[0076] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0077] A plurality of target listening points are defined in the digital model of the in-vehicle acoustic environment to obtain a target listening point coordinate set, and based on the target listening point coordinate set, positions where loudspeakers can be installed in the vehicle are initialized to obtain an initial virtual loudspeaker array;

[0078] The initial virtual loudspeaker array is input into the genetic algorithm for optimization, with maximizing the uniformity of the sound pressure level and the total directivity factor of the system as the objective function, and the virtual loudspeaker array configuration scheme is obtained;

[0079] Based on the virtual speaker array configuration scheme, different frequency response characteristics and directivity parameters are assigned to each virtual speaker to obtain a virtual speaker parameter set, and the virtual speaker parameter set is input into the digital model of the in-vehicle acoustic environment to perform sound field simulation calculations to obtain preliminary sound field distribution data;

[0080] Performing frequency scanning and volume level changes on the preliminary sound field distribution data to generate multiple groups of sound field distribution data containing different frequency and volume combinations;

[0081] Based on multiple sets of sound field distribution data, the cross-correlations between adjacent virtual loudspeakers are calculated, and loudspeakers with high cross-correlations are adjusted to obtain multiple sets of optimized sound field distribution data;

[0082] The optimized multiple sets of sound field distribution data are subjected to data enhancement processing, and multiple sets of in-vehicle sound field distribution training data are obtained by adding random noise and simulating changes in the in-vehicle environment.

[0083] Specifically, multiple target listening points are defined in the digital model of the in-car acoustic environment, including the ear positions of each passenger in the car, such as the positions of the driver, the front passenger seat, and the rear passengers, covering all auditory areas in the car. By determining the target listening points, the coordinate set of the target listening points is obtained, which is recorded as , where each triplet represents the three-dimensional coordinates of a listening point. Based on the coordinate set of the target listening point, the positions where the speakers can be installed in the car are initialized to obtain an initial virtual speaker array. The initialized virtual speaker array ensures that the preliminary sound field covers each target listening point by setting speakers at appropriate positions in the car. The initial virtual speaker array is input into the genetic algorithm for optimization, with the objective function of maximizing the uniformity of the sound pressure level and the total directivity factor of the system. The genetic algorithm is an optimization method based on the biological evolution mechanism, which simulates the selection of nature. The crossover and mutation process gradually approaches the optimal solution. The objective function contains two sub-objectives, the uniformity of the sound pressure level and the total directivity factor of the system, denoted as ,in Indicates the uniformity of sound pressure level, measuring the difference in sound pressure between listening points; It represents the total directivity factor of the system, which measures the speaker array's ability to control the direction of sound propagation; and are weight factors, which are used to adjust the relative importance of the two sub-goals. During the optimization process, each particle represents a speaker configuration scheme. Through iterative updates and fitness evaluation, the genetic algorithm gradually finds the virtual speaker array configuration scheme that maximizes the objective function. Based on the optimized virtual speaker array configuration scheme, different frequency response characteristics and directivity parameters are assigned to each virtual speaker to obtain a virtual speaker parameter set. The frequency response characteristics of each speaker determine its output characteristics at different frequencies, while the directivity parameters determine the radiation ability of sound in different directions. The virtual speaker parameter set is recorded as ,in It is The frequency response function of a loudspeaker is Indicates frequency; Indicates The speakers are in the direction angle These parameters are input into the digital model of the acoustic environment in the car, and the sound field is simulated and calculated to obtain preliminary sound field distribution data, which describes the sound pressure level and phase response characteristics at various locations in the car. In order to verify and improve the robustness of the model, the preliminary sound field distribution data is subjected to frequency scanning and volume level changes to generate multiple sets of sound field distribution data containing different frequency and volume combinations. Frequency scanning is to observe the changes of the sound field in different frequency bands by changing the input frequency of the speaker, while the volume level change is to evaluate the balance of the sound field at different volumes by adjusting the amplitude of the input signal. For example, the sound pressure distribution at different frequencies is obtained by frequency scanning, which is expressed as ,in Indicates the sound pressure level, is the frequency, is the volume level. These multiple sets of sound field distribution data can help understand the performance of the in-car sound field at different frequencies and volumes. Based on the generated multiple sets of sound field distribution data, the cross-correlation between adjacent virtual speakers is calculated. Cross-correlation refers to the degree of mutual influence of sound waves between two speakers in space, which is expressed by the cross-correlation function as follows:

[0084] ;

[0085] in, Indicates and The cross-correlation between the loudspeakers, and Represents the two speakers at time The sound pressure signal at is the integration time range. If the cross-correlation is too high, it means that the sounds of the two speakers will seriously interfere in the space, affecting the overall sound field effect. Adjust the position or parameters of the speakers with high cross-correlation to obtain multiple sets of optimized sound field distribution data. Perform data enhancement processing on the optimized multiple sets of sound field distribution data to increase the generalization ability of the model, so that good sound quality performance can be maintained under different environmental conditions. In the data enhancement processing, random noise is added to simulate the environmental noise encountered in the car during actual driving, such as engine noise, road noise, etc.; simulate the changes in the in-car environment, such as window switches, changes in seat positions, etc., these factors will affect the distribution of the sound field. Through data enhancement, more diverse sound field distribution training data can be obtained, thereby improving the adaptability of the sound field optimization model under different environmental conditions.

[0086] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0087] Preprocess multiple sets of in-vehicle sound field distribution training data, including normalizing the data using the Z-score standardization method and applying the wavelet threshold denoising algorithm to remove high-frequency noise to obtain a standardized training data set;

[0088] Constructing a first differential neural network structure, including an input layer, five differential hidden layers and an output layer, wherein the number of input layer nodes N is equal to the number of virtual speakers, the number of output layer nodes M is equal to the number of target listening points multiplied by three, and the number of nodes in each differential hidden layer is initially set to 2N;

[0089] Initialize the weight W of each differential hidden layer node using the He initialization method, that is, W=sqrt(2 / n in )*random.randn(n in ,n out ), where nin is the number of input nodes, n out is the number of output nodes;

[0090] The standardized training data set is divided into training set, validation set and test set in a ratio of 8:1:1, and the training set data is input into the differential neural network. The Adam optimizer and mean square error loss function are used for preliminary training. The learning rate is set to 0.001, the batch size is 64, and 100 epochs are trained to obtain the initial sound field mapping model.

[0091] Based on the output results of the initial sound field mapping model, the contribution Cj of each hidden layer node j is calculated. The contribution calculation formula is Cj=sum(|∂L / ∂Wj|) / sum(|Wj|), where L is the loss function and Wj is the weight of node j. The nodes whose contribution Cj is lower than the threshold of 0.01 are pruned to obtain the optimized network structure.

[0092] In the optimized network structure, the top 20% contribution layers are identified, and new hidden layers are inserted between these layers. The number of nodes in the newly added layer, Nnew, is set to Nnew=1.5*(Nprev+Nnext) / 2, where Nprev and Nnext are the number of nodes in two adjacent layers respectively;

[0093] Use the orthogonal projection operator method to select the activation function for the newly added nodes, and select the most relevant S-shaped function from {Sigmoid, Tanh, ReLU, LeakyReLU, ELU}. The selection criterion is to minimize the projection error; fine-tune the network weights, use L2 regularization to prevent overfitting, and obtain the second differential neural network structure;

[0094] The standardized training data set is re-input into the second differential neural network structure, and iterative training is performed using the back-propagation algorithm with an early stopping strategy. The mean square error (MSE) is calculated on the validation set after each epoch. Training is stopped when the MSE change is less than 0.1% for five consecutive iterations or when the maximum number of iterations reaches 10,000. Finally, the model performance is evaluated on the test set to obtain the in-car sound field mapping model.

[0095] Specifically, multiple sets of in-car sound field distribution training data are preprocessed. The data is normalized using the Z-score standardization method, and the mean of the data is converted to 0 and the standard deviation is 1, so that the data is processed under the same dimension, which helps to speed up the convergence of the neural network. The calculation formula of Z-score standardization is:

[0096] ;

[0097] in, is the normalized data, is the original data, is the mean of the data, is the standard deviation of the data. Through this standardization process, the differences in dimensions of different features are eliminated, so that the influence of each feature on model training remains consistent. The wavelet threshold denoising algorithm is applied to denoise the data, and the high-frequency noise in the data is removed by decomposing the data and eliminating the high-frequency noise components. A first differential neural network structure is constructed, which includes an input layer, five differential hidden layers, and an output layer. The number of nodes in the input layer is is equal to the number of virtual speakers, while the number of nodes in the output layer is equal to the number of target listening points multiplied by three (representing three acoustic characteristics, such as sound pressure level, phase, and frequency response). The number of nodes in each differential hidden layer is initially set to , thus ensuring that the network has sufficient complexity to fit the complex distribution characteristics of the in-car sound field. During the initialization of the network weights, the weight of each hidden layer node Use the He initialization method for initialization, the formula is:

[0098] ;

[0099] in, represents the weight matrix, is the number of input nodes, is the number of output nodes, Represents a random number matrix that obeys a standard normal distribution. The He initialization method is suitable for deep networks with ReLU activation functions, which can effectively prevent the gradient from disappearing, making the training more stable. After building the neural network, the standardized training data set is divided into training set, validation set, and test set in a ratio of 8:1:1, which are used for model training, validation, and final performance evaluation, respectively. The training set data is then input into the differential neural network for preliminary training. The Adam optimizer and mean square error (MSE) loss function are used during the training process. The Adam optimizer combines the advantages of momentum and adaptive learning rate to effectively speed up the convergence of the network. The loss function MSE is used to measure the error between the model output and the actual data. The calculation formula of MSE is:

[0100] ;

[0101] in, is the loss value, is the actual value, is the predicted value, is the number of samples. The learning rate of the initial training is set to 0.001, the batch size is 64, and 100 epochs are trained. After this process, the initial sound field mapping model is obtained. After obtaining the initial model, in order to optimize the network structure, the number of nodes in each hidden layer is calculated. Contribution Contribution It is used to measure the influence of the node on the final output. The calculation formula is:

[0102] ;

[0103] in, is the loss function Weight of the node The partial derivative of represents the sensitivity of the node to the loss function; is the absolute value and weight of the nodes, indicating the importance of the nodes. For nodes whose contribution is lower than the threshold of 0.01, they are pruned to obtain the optimized network structure. The pruning process helps to reduce the complexity of the network, reduce the risk of overfitting, and improve computational efficiency. In the optimized network structure, the top 20% of the contribution layers are identified, and new hidden layers are inserted between these layers to enhance the network's expressiveness. Number of nodes in the newly added layer Determined by the following formula:

[0104] ;

[0105] in, and Represents the number of nodes in two adjacent layers. By inserting a new hidden layer, the nonlinear representation ability of the network is enhanced to better fit the complex sound field characteristics. In order to select a suitable activation function, the orthogonal projection operator method is used to select the most relevant S-shaped function from {Sigmoid, Tanh, ReLU, LeakyReLU, ELU}. The selection criterion is to minimize the projection error to ensure that the activation function of the newly added node can effectively improve the performance of the network. After completing the activation function selection, the network weights are fine-tuned and L2 regularization is used to prevent overfitting. L2 regularization suppresses excessive weights by adding a penalty term of the square of the weight to the loss function and improves the generalization ability of the model. The standardized training data set is re-input into the optimized second differential neural network structure, and iterative training is performed using the back propagation algorithm with an early stopping strategy. The early stopping strategy is an effective means to prevent overfitting. After each epoch, the mean square error (MSE) on the validation set is calculated. When the MSE changes less than 0.196 for 5 consecutive iterations or reaches the maximum number of iterations of 10,000, the training is stopped. This strategy prevents overfitting of the training data while ensuring model convergence. The performance of the model is evaluated on the test set to measure its predictive ability on unseen data and obtain the final in-car sound field mapping model.

[0106] In a specific embodiment, the process of executing step 500 may specifically include the following steps:

[0107] M1: Analyze the output of the in-car sound field mapping model, extract the sound field characteristics of multiple target listening points, including frequency response, phase response and sound pressure level, and obtain the sound field feature set of the target listening points;

[0108] M2: Based on the sound field feature set of the target listening point, a multi-objective optimization function is constructed, including three sub-objectives: frequency response flatness, sound pressure level uniformity, and stereo image accuracy, to obtain a comprehensive optimization objective function;

[0109] M3: Input the comprehensive optimization objective function into the particle swarm optimization algorithm, initialize the particle swarm, each particle represents a set of candidate MIMO filter coefficients, and obtain the initial particle swarm;

[0110] M4: For each particle in the initial particle group, the in-car sound field mapping model is used to perform forward propagation calculation to obtain the predicted sound field distribution of each target listening point;

[0111] M5: Based on the predicted sound field distribution and the comprehensive optimization objective function, the fitness value of each particle is calculated, and the individual optimal solution and the global optimal solution are updated to obtain the updated particle swarm;

[0112] M6: Apply crossover and mutation operations to the updated particle swarm to generate a new generation of particle swarm, and repeat steps M4 and M5 until a preset number of iterations or convergence conditions are reached to obtain the MIMO filter coefficients.

[0113] Specifically, the output of the in-car sound field mapping model is analyzed to extract the sound field characteristics of multiple target listening points. The target listening points include the driver's seat, the front passenger seat and the rear passenger seat. The sound field characteristics of these points include frequency response, phase response and sound pressure level. The acoustic characteristics characterize the changes in sound pressure at different frequencies, the changes in the cabinet position, and the changes in the overall volume level of the listening point. By analyzing the output of the in-car sound field mapping model, the sound field feature set of each target listening point at different frequencies is obtained, which is recorded as ,in Indicates the target listening point The frequency response of represents the phase response, Represents the sound pressure level. Based on the sound field feature set of the target listening point, a multi-objective optimization function is constructed to optimize the performance of the in-car sound field. The multi-objective optimization function contains three sub-objectives, namely frequency response flatness, sound pressure level uniformity, and stereo image accuracy. Frequency response flatness measures the flatness of the sound pressure curve of each listening point in the entire frequency range. Ideally, the frequency response should be as flat as possible to ensure that sounds of different frequencies have a balanced output. Its calculation formula is expressed as:

[0114] ;

[0115] in, is the loss function of frequency response flatness, For the The frequency response of each listening point, is the average frequency response of all listening points, is the total number of target listening points. The purpose of sound pressure level uniformity is to make the sound pressure level of all listening points as consistent as possible to avoid excessive or weak sound in some areas. It is calculated using the following formula:

[0116] ;

[0117] in, is the loss function of sound pressure level uniformity, For the The sound pressure level at each listening point, is the average sound pressure level of all listening points. Stereo image accuracy measures the balance between the left and right channels, ensuring that the directionality of the sound is correctly felt at each listening point. The comprehensive optimization objective function combines these three sub-objectives in a weighted manner:

[0118] ;

[0119] in, is the comprehensive optimization objective function, is the weight coefficient of each sub-goal, which is used to adjust the relative importance of different goals. is the loss function of stereo image accuracy. The comprehensive optimization objective function is input into the particle swarm optimization algorithm for solution. The particle swarm optimization algorithm is an optimization method based on swarm intelligence. It simulates the foraging behavior of bird flocks to find the global optimal solution in the search space. In particle swarm optimization, each particle represents a candidate solution, and each particle represents a set of candidate MIIMO (multiple-input multiple-output) filter coefficients used to control the signal output of each speaker. Initialize the particle swarm, and the initial position and velocity of each particle are randomly distributed to obtain the initial particle swarm. The initial position of the particle represents a set of initial MIMO filter coefficients, denoted as ,in is the number of particles. For each particle in the initial particle swarm, the in-car sound field mapping model is used to perform forward propagation calculations to obtain the predicted sound field distribution of each target listening point. The current filter coefficients of the particles are applied to the digital model of the in-car acoustic environment to simulate the propagation of the sound output by the loudspeaker in the car, and the frequency response, phase response and sound pressure level of each listening point are calculated. Based on the predicted sound field distribution and the comprehensive optimization objective function, the fitness value of each particle is calculated. The fitness value is used to measure the performance of the current particle solution in multi-objective optimization. The lower the fitness value, the closer the filter coefficients of this group are to meet the requirements of the target sound field distribution. For each particle, its individual optimal solution is updated, that is, the best solution obtained by the current particle so far. At the same time, the global optimal solution is updated in the entire particle swarm, that is, the optimal solution found in all particles. Through the update, an updated particle swarm is obtained. In order to improve the diversity of the particle swarm and avoid falling into the local optimal solution, the crossover and mutation operations are applied to the updated particle swarm to generate a new generation of particle swarms. The crossover operation generates new particles with parent characteristics by exchanging the feature vectors of different particles, while the mutation operation increases the diversity of the group by randomly perturbing the positions of particles. Repeat the steps of forward propagation calculation and fitness calculation to continuously update the position and speed of the particle group and optimize the value of the comprehensive objective function. The whole process is repeated until the preset number of iterations is reached or the convergence condition is met. Usually, the convergence condition is set to the fitness value change being less than a certain threshold or reaching the maximum number of iterations to ensure that the optimal MIMO filter coefficients are found under the constraints of computing resources.

[0120] In a specific embodiment, the process of executing step 600 may specifically include the following steps:

[0121] Install multiple MEMS microphones at key locations in the car to collect real-time sound field data, and pre-process the real-time sound field data, including denoising and spectrum analysis, to obtain real-time in-car sound field characteristics;

[0122] The MIMO filter coefficients and the real-time in-vehicle sound field characteristics are input into the adaptive filter, and the filter coefficients are updated in real time using the minimum mean square error algorithm to obtain the preliminary adjusted filter coefficients;

[0123] Based on the initially adjusted filter coefficients, the predicted sound field distribution of the target listening point is calculated and compared with the ideal sound field distribution to obtain a sound field error vector;

[0124] The sound field error vector is input into the Wiener filter, the error is optimally estimated, and a compensation signal is obtained;

[0125] Decomposing the compensation signal in the frequency domain to obtain multiple sub-band signals, and performing gain adjustment and phase correction on each sub-band signal separately to obtain an optimized sub-band compensation signal;

[0126] The optimized sub-band compensation signals are synthesized to obtain a time domain compensation signal, and the time domain compensation signal is superimposed on the original audio signal to obtain a compensated audio signal;

[0127] The compensated audio signal is subjected to dynamic range compression and frequency response equalization processing to obtain target sound field compensation parameters, which are then applied to the digital signal processor of the vehicle audio system to process the output audio in real time to obtain a real-time optimized in-vehicle sound field.

[0128] Specifically, multiple MEMS microphones are installed at key locations in the car to accurately collect real-time sound field data in the car. The real-time sound field data is preprocessed, and the preprocessing includes two steps: denoising and spectrum analysis. The purpose of denoising is to remove unnecessary signals such as environmental noise, wind noise in the car, engine noise, etc., which is achieved through low-pass filters or other noise reduction algorithms, such as using wavelet threshold denoising methods to weaken high-frequency noise. The denoised data is subjected to spectrum analysis, and the time domain signal is converted into a frequency domain signal through fast Fourier transform to obtain the amplitude and phase information at each frequency to form real-time in-car sound field characteristics. Spectral analysis reveals the changes in the sound field at different frequencies. The MIMO (multiple input multiple output) filter coefficients and real-time sound field characteristics are input into the adaptive filter for processing. The adaptive filter uses the minimum mean square error (LMS) algorithm to update the filter coefficients in real time. The goal of the LMS algorithm is to minimize the mean square error between the filter output signal and the ideal reference signal, and to achieve real-time optimization by continuously adjusting the filter weights. The weight update formula of the LMS algorithm is:

[0129] ;

[0130] in, It is The filter weights for the iterations, is the learning rate, which controls the step size of the update, represents the current error signal, is the input signal. Through the iterative process, the filter coefficients that are preliminarily adjusted are obtained to dynamically compensate for the changes in acoustic characteristics at different locations in the car. Based on the preliminarily adjusted filter coefficients, the predicted sound field distribution of the target listening point is calculated. By applying the adjusted filter coefficients to the in-car acoustic model, the sound field characteristics such as sound pressure level and frequency response at different listening points are predicted. The predicted sound field characteristics are compared with the ideal sound field distribution to obtain a sound field error vector, which is used to represent the difference between the current sound field and the ideal sound field. The sound field error vector is expressed as:

[0131] ;

[0132] in, is the error vector, Indicated in The sound field error of the target listening point is is the number of target listening points. In order to compensate for the sound field error, the sound field error vector is input into the Wiener filter for optimal estimation. The Wiener filter is a linear filter that achieves optimal estimation of the signal by minimizing the mean square value of the error. After Wiener filtering, a compensation signal is obtained, which is used to reduce the error between the current sound field and the ideal sound field and improve the sound field distribution in the car. The compensation signal is decomposed in the frequency domain and decomposed into multiple sub-band signals using a bandpass filter. Each sub-band signal represents the sound information of a different frequency band. By performing gain adjustment and phase correction on the sub-band signals respectively, the different frequency bands of the sound field can be controlled more finely. The gain adjustment is to ensure that the sound pressure level of each frequency band reaches the ideal level, while the phase correction is to ensure that the sound waves of different frequency bands maintain a suitable phase relationship during propagation, thereby avoiding phase interference. The calculation formula for gain adjustment is expressed as:

[0133] ;

[0134] in, Indicates The gain of each sub-band is is the target sound pressure level, is the currently measured sound pressure level. , adjust the current sound pressure level to the target level. Synthesize the optimized sub-band compensation signals to obtain the overall time domain compensation signal. By superimposing it with the original audio signal, the compensated audio signal is obtained. In order to optimize the compensated audio signal, dynamic range compression and frequency response equalization are performed on it. The purpose of dynamic range compression is to reduce the instantaneous large changes in the signal, prevent the speaker from being damaged by excessive sound pressure, and improve the overall clarity of the audio. The formula for dynamic range compression is:

[0135] ;

[0136] in, is the compressed output signal, is the original signal, is the set threshold, A function that represents the sign of a signal and is used to preserve the positive and negative directions of the original signal. Dynamic range compression ensures that the speaker will not exceed its capacity due to excessive input by compressing the part above the threshold. Frequency response equalization adjusts the different frequency components of the audio signal to make the overall sound more balanced and natural. For example, the low and high frequencies are moderately enhanced to compensate for the attenuation of high or low frequencies in the car due to material properties. Frequency response equalization is to ensure that the overall performance of the sound meets the best auditory experience. After processing, the target sound field compensation parameters are obtained. The target sound field compensation parameters are applied to the digital signal processor in the car audio system to process the output audio signal in real time to achieve the ideal sound field distribution. The compensation parameters can be dynamically adjusted according to the real-time changes in the sound field in the car, thereby ensuring that the sound field in the car always maintains the best auditory effect.

[0137] The above describes the vehicle speaker sound field optimization method in the embodiment of the present application. The following describes the vehicle speaker sound field optimization device in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the vehicle speaker sound field optimization device includes:

[0138] The measuring module 11 is used to measure the parameters of the vehicle-mounted speaker to obtain the speaker parameter data and frequency response curve;

[0139] Establishing module 12, used for establishing a digital model of the in-vehicle acoustic environment according to the speaker parameter data and the frequency response curve;

[0140] A creation module 13 is used to create a virtual speaker array configuration scheme based on the digital model of the in-vehicle acoustic environment, and generate multiple groups of in-vehicle sound field distribution training data according to the virtual speaker array configuration scheme;

[0141] The training module 14 is used to input multiple sets of in-vehicle sound field distribution training data into the differential neural network for training to obtain an in-vehicle sound field mapping model;

[0142] An optimization module 15 is used to optimize the sound field of multiple target listening points based on the in-vehicle sound field mapping model to obtain MIMO filter coefficients;

[0143] The compensation module 16 is used to perform adaptive sound field compensation according to the MIMO filter coefficients and the real-time monitored in-vehicle sound field data, obtain target sound field compensation parameters, and generate a real-time optimized in-vehicle sound field according to the target sound field compensation parameters.

[0144] Through the collaboration of the above components, accurate parameter measurement of the vehicle speakers and establishment of a digital model of the in-vehicle acoustic environment provide accurate basic data for subsequent optimization, which is conducive to improving the accuracy and pertinence of the optimization. The virtual speaker array technology and differential neural network training are used to achieve accurate mapping and prediction of the in-vehicle sound field, overcoming the difficulty of traditional methods in modeling complex acoustic environments. The introduction of multi-objective optimization and particle swarm algorithm, based on multiple factors such as frequency response flatness, sound pressure level uniformity and stereo image accuracy, obtains the global optimal MIMO filter coefficients, and improves the overall effect of sound field optimization. Combined with real-time monitoring and adaptive filtering technology, it achieves rapid response and compensation to changes in the in-vehicle sound field, solving the problem of poor adaptability of traditional fixed parameter optimization methods. Through technologies such as sub-band processing and dynamic range compression, while ensuring sound quality, the system's adaptability to different sound sources and environmental noise is improved, enhancing the consistency and comfort of the listening experience.

[0145] See also Figure 3 , Figure 3 This is a schematic block diagram of the structure of an electronic device 300 provided in an embodiment of the present application. The electronic device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected via a device bus 203, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0146] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned vehicle speaker sound field optimization methods.

[0147] The processor 301 is used to provide computing and control capabilities to support the operation of the entire electronic device 300 .

[0148] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned vehicle speaker sound field optimization methods.

[0149] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a partial structure related to the present application scheme, and does not constitute a limitation on the electronic device 300 involved in the present application scheme. The specific electronic device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0150] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0151] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the electronic device 300 described above can refer to the corresponding process of the aforementioned vehicle speaker sound field optimization method, which will not be repeated here.

[0152] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the vehicle speaker sound field optimization method provided in the embodiment of the present application.

[0153] The computer-readable storage medium may be an internal storage unit of the electronic device 300 in the aforementioned embodiment, such as a hard disk or memory of the electronic device 300. The computer-readable storage medium may also be an external storage device of the electronic device 300, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped with the electronic device 300.

[0154] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0155] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0156] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for optimizing the sound field of a vehicle-mounted speaker, characterized in that: The method comprises: Measure the parameters of the vehicle speakers to obtain the speaker parameter data and frequency response curve; The digital model of the in-vehicle acoustic environment is established according to the loudspeaker parameter data and the frequency response curve; specifically comprising: performing 3D laser scanning on the in-vehicle space to obtain three-dimensional point cloud data, and inputting the three-dimensional point cloud data into computer-aided design software for processing to obtain an accurate three-dimensional model of the interior of the vehicle compartment; meshing the accurate three-dimensional model of the interior of the vehicle compartment to obtain an in-vehicle space mesh model composed of a plurality of mesh units; assigning a sound absorption coefficient and a reflection coefficient to each mesh unit in the in-vehicle space mesh model according to the acoustic characteristics of the in-vehicle material to obtain an initial acoustic characteristic model; and The frequency response curve is input into the initial acoustic characteristic model, and a virtual sound source is set at a corresponding position to obtain an in-vehicle acoustic model including the sound source; a finite element analysis is performed on the in-vehicle acoustic model including the sound source to simulate the propagation characteristics of sound waves in the vehicle compartment to obtain a preliminary sound field distribution result; actual sound field measurements are performed at multiple target positions in the vehicle to obtain measured sound field data, and the measured sound field data is compared with the preliminary sound field distribution result to calculate an error value; the in-vehicle acoustic model including the sound source is iteratively optimized based on the error value until the error value is less than a preset threshold value to obtain a digital model of the in-vehicle acoustic environment; Creating a virtual speaker array configuration scheme based on the in-vehicle acoustic environment digital model, and generating multiple sets of in-vehicle sound field distribution training data according to the virtual speaker array configuration scheme; Inputting the plurality of sets of in-vehicle sound field distribution training data into a differential neural network for training to obtain an in-vehicle sound field mapping model; Based on the in-car sound field mapping model, the sound field is optimized for multiple target listening points to obtain MIMO filter coefficients; Adaptive sound field compensation is performed according to the MIMO filter coefficients and the real-time monitored in-vehicle sound field data to obtain target sound field compensation parameters, and a real-time optimized in-vehicle sound field is generated according to the target sound field compensation parameters.

2. The method for optimizing the sound field of a vehicle-mounted speaker according to claim 1, characterized in that: The parameter measurement of the vehicle-mounted speaker to obtain speaker parameter data and a frequency response curve includes: Applying a sweep frequency signal of 20 Hz-20 kHz to the vehicle speaker to obtain original response data, and performing fast Fourier transform on the original response data to obtain frequency response data; Calculate the frequency response curve of the vehicle speaker in the range of 150 Hz-10 kHz based on the frequency response data; Applying a 1 kHz sinusoidal signal to the vehicle speaker, calculating the impedance value of the speaker by a voltage and current measurement method to obtain impedance data, and calculating the Q value of the vehicle speaker based on the impedance data to obtain a quality factor; Applying a multi-band signal with an input power of 1 W to the vehicle-mounted speaker, measuring the total harmonic distortion by a harmonic analyzer, and obtaining distortion data; Under the condition of 1W / 1M, the sound pressure level of the vehicle-mounted speaker is measured, and the average value of the four frequency points of 0.6kHz, 0.8kHz, 1.0kHz and 1.25kHz is taken to obtain the sensitivity data; Applying a short-time high-power signal to the vehicle-mounted speaker, measuring the maximum input power of the vehicle-mounted speaker by a power analyzer, and obtaining power handling capacity data; The frequency response data, the impedance data, the quality factor, the distortion data, the sensitivity data and the power handling capacity data are normalized to construct speaker parameter data.

3. The method for optimizing the sound field of a vehicle-mounted speaker according to claim 1, characterized in that: The step of creating a virtual speaker array configuration scheme based on the in-vehicle acoustic environment digital model, and generating multiple sets of in-vehicle sound field distribution training data according to the virtual speaker array configuration scheme includes: Defining a plurality of target listening points in the digital model of the in-vehicle acoustic environment to obtain a target listening point coordinate set, and initializing positions where loudspeakers can be installed in the vehicle based on the target listening point coordinate set to obtain an initial virtual loudspeaker array; Inputting the initial virtual loudspeaker array into a genetic algorithm for optimization, taking maximization of sound pressure level uniformity and system total directivity factor as objective functions, and obtaining a virtual loudspeaker array configuration scheme; Based on the virtual speaker array configuration scheme, different frequency response characteristics and directivity parameters are assigned to each virtual speaker to obtain a virtual speaker parameter set, and the virtual speaker parameter set is input into the digital model of the in-vehicle acoustic environment to perform sound field simulation calculation to obtain preliminary sound field distribution data; Performing frequency scanning and volume level changes on the preliminary sound field distribution data to generate multiple groups of sound field distribution data containing different frequency and volume combinations; Based on the multiple groups of sound field distribution data, calculating the cross-correlation between adjacent virtual loudspeakers, and adjusting loudspeakers with high cross-correlation to obtain multiple groups of optimized sound field distribution data; The optimized multiple groups of sound field distribution data are subjected to data enhancement processing, and multiple groups of in-vehicle sound field distribution training data are obtained by adding random noise and simulating changes in the in-vehicle environment.

4. The method for optimizing the sound field of a vehicle-mounted speaker according to claim 1, characterized in that: The step of inputting the plurality of sets of in-vehicle sound field distribution training data into a differential neural network for training to obtain an in-vehicle sound field mapping model comprises: Preprocessing the multiple sets of in-vehicle sound field distribution training data, including normalizing the data using a Z-score normalization method and removing high-frequency noise using a wavelet threshold denoising algorithm to obtain a standardized training data set; Constructing a first differential neural network structure, including an input layer, five differential hidden layers and an output layer, wherein the number of input layer nodes N is equal to the number of virtual speakers, the number of output layer nodes M is equal to the number of target listening points multiplied by three, and the number of nodes in each differential hidden layer is initially set to 2N; Initialize the weight W of each differential hidden layer node using the He initialization method, that is, W=sqrt(2 / n in )*random.randn(n in ,n out ), where n in is the number of input nodes, n out is the number of output nodes; The standardized training data set is divided into a training set, a validation set and a test set in a ratio of 8:1:1, and the training set data is input into the differential neural network, and preliminary training is performed using the Adam optimizer and the mean square error loss function, with the learning rate set to 0.001, the batch size to 64, and 100 epochs of training to obtain an initial sound field mapping model; Based on the output result of the initial sound field mapping model, the contribution Cj of each hidden layer node j is calculated, and the contribution calculation formula is Cj=sum(|∂L / ∂Wj|) / sum(|Wj|), where L is the loss function and Wj is the weight of node j; the nodes whose contribution Cj is lower than the threshold value 0.01 are pruned to obtain the optimized network structure; In the optimized network structure, the top 20% contribution layers are identified, and new hidden layers are inserted between these layers. The number of nodes Nnew in the newly added layer is set to Nnew=1.5*(Nprev+Nnext) / 2, where Nprev and Nnext are the number of nodes in two adjacent layers respectively; Use the orthogonal projection operator method to select the activation function for the newly added nodes, and select the most relevant S-shaped function from {Sigmoid, Tanh, ReLU, LeakyReLU, ELU}. The selection criterion is to minimize the projection error; fine-tune the network weights, use L2 regularization to prevent overfitting, and obtain the second differential neural network structure; The standardized training data set is re-input into the second differential neural network structure, and iterative training is performed using the back propagation algorithm with an early stopping strategy. The mean square error (MSE) is calculated on the validation set after each epoch, and the training is stopped when the MSE change is less than 0.1% for 5 consecutive iterations or the maximum number of iterations is 10,000. Finally, the model performance is evaluated on the test set to obtain the in-car sound field mapping model.

5. The method for optimizing the sound field of a vehicle-mounted speaker according to claim 1, characterized in that: The sound field optimization of multiple target listening points based on the in-vehicle sound field mapping model to obtain MIMO filter coefficients includes: M1: Analyze the output of the in-car sound field mapping model, extract the sound field characteristics of multiple target listening points, including frequency response, phase response and sound pressure level, and obtain a sound field feature set of the target listening points; M2: Based on the target listening point sound field feature set, a multi-objective optimization function is constructed, including three sub-objectives: frequency response flatness, sound pressure level uniformity, and stereo image accuracy, to obtain a comprehensive optimization objective function; M3: Input the comprehensive optimization objective function into the particle swarm optimization algorithm, initialize the particle swarm, each particle represents a set of candidate MIMO filter coefficients, and obtain an initial particle swarm; M4: For each particle in the initial particle group, use the in-vehicle sound field mapping model to perform forward propagation calculation to obtain the predicted sound field distribution of each target listening point; M5: Based on the predicted sound field distribution and the comprehensive optimization objective function, the fitness value of each particle is calculated, and the individual optimal solution and the global optimal solution are updated to obtain an updated particle swarm; M6: applying crossover and mutation operations to the updated particle swarm to generate a new generation of particle swarm, and repeatedly performing steps M4 and M5 until a preset number of iterations or convergence conditions are reached to obtain MIMO filter coefficients.

6. The method for optimizing the sound field of a vehicle-mounted speaker according to claim 1, characterized in that: The method of performing adaptive sound field compensation according to the MIMO filter coefficient and the real-time monitored in-vehicle sound field data to obtain target sound field compensation parameters, and generating a real-time optimized in-vehicle sound field according to the target sound field compensation parameters, comprises: Install multiple MEMS microphones at key locations in the vehicle to collect real-time sound field data, and pre-process the real-time sound field data, including denoising and spectrum analysis, to obtain real-time in-vehicle sound field characteristics; Inputting the MIMO filter coefficients and the real-time in-vehicle sound field characteristics into an adaptive filter, and using a minimum mean square error algorithm to update the filter coefficients in real time to obtain preliminarily adjusted filter coefficients; Based on the preliminarily adjusted filter coefficients, the predicted sound field distribution of the target listening point is calculated, and compared with the ideal sound field distribution to obtain a sound field error vector; Inputting the sound field error vector into a Wiener filter, performing optimal estimation on the error, and obtaining a compensation signal; Decomposing the compensation signal in the frequency domain to obtain a plurality of sub-band signals, and performing gain adjustment and phase correction on each sub-band signal separately to obtain an optimized sub-band compensation signal; Synthesizing the optimized sub-band compensation signals to obtain a time domain compensation signal, and superimposing the time domain compensation signal with the original audio signal to obtain a compensated audio signal; The compensated audio signal is subjected to dynamic range compression and frequency response equalization processing to obtain target sound field compensation parameters, and the target sound field compensation parameters are applied to a digital signal processor of the vehicle audio system to perform real-time processing on the output audio to obtain a real-time optimized in-vehicle sound field.

7. A vehicle-mounted speaker sound field optimization device, characterized in that: The vehicle-mounted speaker sound field optimization device comprises: The measurement module is used to measure the parameters of the vehicle-mounted speaker and obtain the speaker parameter data and frequency response curve; The establishment module is used to establish a digital model of the in-vehicle acoustic environment according to the speaker parameter data and the frequency response curve; specifically comprising: performing 3D laser scanning on the in-vehicle space to obtain three-dimensional point cloud data, and inputting the three-dimensional point cloud data into computer-aided design software for processing to obtain an accurate three-dimensional model of the interior of the vehicle compartment; meshing the accurate three-dimensional model of the interior of the vehicle compartment to obtain an in-vehicle space mesh model composed of a plurality of mesh units; assigning a sound absorption coefficient and a reflection coefficient to each mesh unit in the in-vehicle space mesh model according to the acoustic properties of the in-vehicle material to obtain an initial acoustic characteristic model; and The digital data and the frequency response curve are input into the initial acoustic characteristic model, and a virtual sound source is set at a corresponding position to obtain an in-vehicle acoustic model including the sound source; a finite element analysis is performed on the in-vehicle acoustic model including the sound source to simulate the propagation characteristics of sound waves in the vehicle compartment to obtain a preliminary sound field distribution result; actual sound field measurements are performed at multiple target positions in the vehicle to obtain measured sound field data, and the measured sound field data is compared with the preliminary sound field distribution result to calculate an error value; the in-vehicle acoustic model including the sound source is iteratively optimized based on the error value until the error value is less than a preset threshold value to obtain a digital model of the in-vehicle acoustic environment; A creation module, used to create a virtual speaker array configuration scheme based on the in-vehicle acoustic environment digital model, and generate multiple groups of in-vehicle sound field distribution training data according to the virtual speaker array configuration scheme; A training module, used for inputting the plurality of sets of in-vehicle sound field distribution training data into a differential neural network for training to obtain an in-vehicle sound field mapping model; An optimization module, configured to optimize the sound field for multiple target listening points based on the in-vehicle sound field mapping model to obtain MIMO filter coefficients; The compensation module is used to perform adaptive sound field compensation according to the MIMO filter coefficients and the real-time monitored in-vehicle sound field data, obtain target sound field compensation parameters, and generate a real-time optimized in-vehicle sound field according to the target sound field compensation parameters.

8. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the electronic device to execute the vehicle speaker sound field optimization method according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the vehicle-mounted speaker sound field optimization method according to any one of claims 1 to 6 is implemented.

Images