Sound field optimization method and device of multichannel audio system, equipment and storage medium

Abstract

The invention discloses a sound field optimization method and device for a multichannel audio system, equipment and a storage medium. The sound field optimization method comprises the following steps: initializing the system and acquiring acoustic characteristic data; simulating a propagation path of sound waves based on a sound ray tracking algorithm and identifying measurement data, and constructing a 3D sound field model; calculating a TDOA result of a sound channel signal between different microphone positions based on a GCC-PHAT algorithm, obtaining a listening position, combining the TDOA result to calculate an optimal delay value, and performing corresponding compensation; a target mathematical model is formed, a sound field optimization problem of a plurality of listening positions is converted into a multi-target optimization problem, a Pareto optimal solution is generated based on an NSGA-II algorithm, a solution for balancing performance of different positions is selected, the sound quality of other positions is considered while the optimal effect of the main listening position is ensured, the listening experience of a single position is remarkably improved, and the listening experience of the single position is improved. And sound field optimization in a whole space range is realized, and immersive audio experience is brought to the user.

Description

Technical Field

[0001] The present invention relates to the field of audio technology, and in particular to a sound field optimization method, device, equipment and storage medium for a multi-channel audio system. Background Art

[0002] Traditional multi-channel audio systems face many challenges in practical applications. First, due to the different distances from the speakers to different listening positions, the difference in the arrival time of sound waves can lead to inaccurate sound image positioning, uneven sound field and other problems. Although some high-end systems provide delay compensation functions, they are usually optimized based on a single "sweet listening position" and cannot meet the needs of multiple listening positions. Secondly, the acoustic characteristics of the room have a significant impact on the quality of sound reproduction. Phenomena such as reflection, absorption and diffraction will change the propagation characteristics of sound waves, but existing systems often lack comprehensive consideration of these factors. In addition, different types of audio content such as music, movies, and games have different requirements for the sound field, and fixed optimization parameters are difficult to adapt to diverse usage scenarios. At the same time, users' listening habits and preferences will also affect the optimal sound field settings, but existing systems lack personalized adaptability. Finally, the traditional optimization process usually requires professionals to use complex equipment for measurement and debugging, which not only increases costs, but also limits the popularization and application of the system. These problems seriously restrict the performance of multi-channel audio systems and the improvement of user experience. Summary of the invention

[0003] The present invention provides a sound field optimization method, device, equipment and storage medium for a multi-channel audio system, which effectively solves the problems faced by traditional multi-channel audio systems. It uses a microphone array to collect acoustic data from multiple positions in real time, combines advanced digital signal processing algorithms, automatically calculates and applies optimal delay compensation parameters, introduces spatial sound field modeling technology, and can accurately simulate the propagation characteristics of sound waves in a room. The innovative multi-objective optimization algorithm allows the sound quality of other positions to be taken into account while ensuring the best effect at the main listening position. It not only significantly improves the listening experience of a single position, but also realizes sound field optimization in the entire space, bringing users an unprecedented immersive audio experience.

[0004] A first aspect of the present invention provides a method for optimizing a sound field of a multi-channel audio system, comprising:

[0005] Initialize the multi-channel audio system and obtain acoustic characteristic data of different environments;

[0006] Simulating the propagation path of sound waves in the room based on a ray tracing algorithm and identifying measurement data related to the reflection of the sound rays in the room, and constructing a 3D sound field model covering the listening area of ​​the system based on the measurement data;

[0007] Calculate the TDOA estimation result of the channel signal between different microphone positions based on the GCC-PHAT algorithm, obtain the listening position and calculate the optimal delay value in combination with the TDOA estimation result, and perform corresponding compensation;

[0008] A target mathematical model is constructed to transform the sound field optimization problem of multiple listening positions into a multi-objective optimization problem. The Pareto optimal solution is generated based on the NSGA-II algorithm, and the solution that balances the performance of different positions is selected.

[0009] Preferably, the initializing the multi-channel audio system and acquiring acoustic feature data of different environments includes:

[0010] Deploy multiple microphones within the listening area of ​​the system to form a microphone array;

[0011] Play a logarithmic sweep signal covering the full frequency band to each channel in turn;

[0012] Synchronously recording the room acoustic response of each channel through the microphone array to capture the propagation and reflection of sound in the room;

[0013] The number and positions of the microphones are determined according to environmental characteristics, including the size and shape of the room.

[0014] Preferably, the simulating the propagation path of the sound wave in the room based on the sound ray tracing algorithm and identifying the measurement data related to the sound ray and the reflection in the room includes:

[0015] Get the incident vector and the surface normal vector, calculate the reflection vector to determine the reflection path of the sound wave, and the implementation formula is as follows: P_reflected = P_incident-2(N·P_incident)N,

[0016] Among them, P_incident is the incident vector, N is the surface normal vector, and P_reflected is the reflected vector.

[0017] Preferably, constructing a 3D sound field model covering the listening area of ​​the system according to the measurement data comprises:

[0018] The transfer function from each loudspeaker to each microphone position is determined based on the least squares estimation according to the measurement data: H(ω)=(X^H*X)^-1*X^H*Y,

[0019] Where X is the input signal matrix, Y is the output signal matrix, and H(ω) is the estimated transfer function;

[0020] Obtain point data corresponding to the measurement data of the listening area, construct a 3D sound field model covering the listening area according to the point data, and optimize the acoustic characteristics in the room by implementing a continuous sound field representation in space according to the Kriging interpolation method:

[0021] Z(s_0)=∑[i=1ton]λ_i*Z(s_i),

[0022] Among them, Z(s_0) is the model value of the acoustic characteristics of the point s_0 to be estimated, Z(s_i) is the model value of the acoustic characteristics of the known point s_i, λ_i is the weight, and n is the total number of known measurement points.

[0023] Preferably, the TDOA estimation result of the channel signal between different microphone positions is calculated based on the GCC-PHAT algorithm, comprising:

[0024] Perform fast Fourier transform on each channel signal recorded by each microphone to obtain the spectrum of the two signals;

[0025] Calculate the time difference of arrival TDOA between two signals:

[0026] R_GCC(τ)=IFFT{X_1(ω)X_2^*(ω) / |X_1(ω)X_2^*(ω)|},

[0027] Where R_GCC(τ) is a function of the time delay τ, and X_1(ω) and X_2(ω) are the spectra of the two signals.

[0028] Preferably, the acquiring of the listening position in combination with the TDOA estimation result to calculate the optimal delay value and performing corresponding compensation comprises:

[0029] By calculating the time difference from different speakers to the listening position, a weighted least squares optimization algorithm is used to calculate the optimal compensation delay, so that the difference between the actual arrival time and the target arrival time of all weighted listening positions is minimized:

[0030] minimize∑[i=1toM]w_i*||T_i-(d_i+δ_i)||^2,

[0031] Where M is the number of listening positions, w_i is the weight of the i-th listening position, T_i is the target arrival time at the i-th listening position, d_i is the original delay of the i-th speaker, and δ_i is the compensated delay of the i-th speaker;

[0032] Configure the FIR filter to achieve accurate fractional sample delay, and use Lagrange interpolation to construct the filter coefficients to ensure delay accuracy: h(n)=∏[k=0,k≠ntoN-1](Dk) / (nk),

[0033] Where h(n) is the filter response at the nth sample point, D is the desired fractional delay, and N is the filter order.

[0034] Preferably, the constructing of the target mathematical model to transform the sound field optimization problem of multiple listening positions into a multi-objective optimization problem comprises:

[0035] defining target responses at different listening positions, the target responses including frequency response and phase response;

[0036] Determine multiple objective functions f_1(x), f_2(x), ..., f_k(x) according to the flatness of the balanced frequency response, the linearity of the phase, and the accuracy of the sound image localization at multiple listening positions;

[0037] Minimize multiple objective functions: minimize[f_1(x),f_2(x),...,f_k(x)],

[0038] Constraints: g_i(x)≤0,i=1,2,...,m; h_j(x)=0,j=1,2,...,n;

[0039] Among them, f_1(x), f_2(x)…f_k(x) are the 1st, 2nd…kth objective functions, and g_i(x) and h_j(x) are constraints, namely the maximum gain limit and the maximum movement range limit of the speaker.

[0040] Preferably, the generating of the Pareto optimal solution based on the NSGA-II algorithm and selecting the solution that balances the performance of different positions includes:

[0041] Set the initial parameters and initial population P of the algorithm;

[0042] Perform non-dominated sorting on the initial population P to classify the performance of individuals in the initial population on multiple objective functions and identify individuals on the Pareto frontier;

[0043] The crowding distance is determined by calculating the difference in the objective function values ​​between adjacent individuals, thereby measuring the local density of individuals in the Pareto frontier.

[0044] Based on the individual fitness and crowding distance, individuals are selected for crossover and mutation to generate a new generation of population Q;

[0045] Merge the initial population P and the newly generated population Q, and select the new population based on the non-dominated sorting and crowding distance;

[0046] Repeat the above steps, update the Pareto frontier, and iterate to ensure the convergence of the Pareto optimal solution.

[0047] As a preferred embodiment, it also includes:

[0048] Based on the adaptive learning mechanism combined with collaborative filtering and online gradient descent algorithms, it can dynamically adjust the optimization strategy according to user habits and environmental changes, further including:

[0049] Obtain the interaction data between the user and the audio content to form a user-item rating matrix R, wherein the interaction data includes the user's ratings of different audio contents, playback history, adjustment settings, and data related to behavioral preferences. In R, the rows represent users, the columns represent audio contents or items, and the values ​​of the elements are the user's ratings or preferences for the items.

[0050] A personalized preference model is constructed based on a collaborative filtering algorithm, so that similarities between users can be discovered and user preferences can be predicted based on similarities. The collaborative filtering algorithm uses matrix decomposition to decompose the user-item rating matrix R into two target low-dimensional matrices to output potential features;

[0051] Among them, R≈P*Q^T, R is the user-item rating matrix, P is the user factor matrix and Q is the item factor matrix.

[0052] As a preferred embodiment, it also includes:

[0053] Based on the adaptive learning mechanism combined with collaborative filtering and online gradient descent algorithms, it can dynamically adjust the optimization strategy according to user habits and environmental changes, further including:

[0054] Collect audio samples from different scenes and annotate each audio sample;

[0055] Extracting target features from the audio sample and using them as inputs of the SVM model, the target features include spectrum features, rhythm features, and time structure features;

[0056] The SVM model is trained using labeled audio samples and extracted features to best distinguish samples from different scene categories;

[0057] Deploy the trained SVM model to the actual audio system. When new audio data arrives, extract the same features, input the features into the SVM model for classification, and predict the scene category to which the audio data belongs based on the input features.

[0058] According to the output prediction result of the SVM model, automatically switch to the corresponding optimization strategy, wherein when it is identified that the user is in a video watching mode, the low-frequency effect is enhanced to provide a better viewing experience; when it is identified that the user is in a music listening mode, the equalizer setting is adjusted to highlight the rhythm of the music;

[0059] The SVM model parameters are continuously updated through online learning to adapt to new data and changes in user preferences.

[0060] As a preferred embodiment, it also includes:

[0061] Adopt parallel computing architecture and intelligent caching strategy for real-time processing and performance optimization, further including:

[0062] Configure multi-core DSP or FPGA for parallel signal processing, and use pipeline structure to distribute different processing tasks to different processor cores to minimize processing delay;

[0063] In task scheduling, the following formula is used to calculate the total processing time T_total:

[0064] T_total=max(T_i+C_i,j / P_j),

[0065] Where T_i is the start time of task i, C_i,j is the execution time of task i on processor j, and P_j is the performance of processor j; and / or,

[0066] Preloading frequently used filter coefficients into fast-access memory based on intelligent caching strategies, using the principle of locality to optimize memory access patterns, i.e. storing frequently accessed data in physically close memory to reduce memory access latency; and / or,

[0067] Dynamically adjusts internal processing sampling rate based on polyphase filter banks as audio content and system load change.

[0068] As a preferred embodiment, it also includes:

[0069] Evaluate multi-channel audio systems, including:

[0070] The target parameters of the system are measured by an audio analysis instrument to perform performance evaluation, wherein the target parameters include frequency response, phase response and group delay parameters, wherein the audio analysis instrument uses a logarithmic swept frequency signal to measure the transfer function of the system;

[0071] The sound quality of a series of audio samples was scored using the MUSHRA test method. The scoring formula is: score = (original score - low anchor score) / (reference score - low anchor score) * 100%, where the low anchor score refers to the subjective score obtained for the audio sample with the worst quality reference in MUSHRA.

[0072] A second aspect of the present invention provides a sound field optimization device for a multi-channel audio system, comprising:

[0073] An initialization module, used to initialize the multi-channel audio system and obtain acoustic feature data of different environments;

[0074] An acquisition module, used to simulate the propagation path of sound waves in the room based on a sound ray tracing algorithm and identify the measurement data related to the reflection of the sound ray in the room, and to construct a 3D sound field model covering the listening area of ​​the system according to the measurement data;

[0075] A calculation module, used to calculate the TDOA estimation result of the channel signal between different microphone positions based on the GCC-PHAT algorithm, obtain the listening position and calculate the optimal delay value in combination with the TDOA estimation result, and perform corresponding compensation;

[0076] The optimization module is used to construct the target mathematical model, transform the sound field optimization problem of multiple listening positions into a multi-objective optimization problem, generate the Pareto optimal solution based on the NSGA-II algorithm, and select the solution that balances the performance of different positions.

[0077] A third aspect of the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the above-mentioned sound field optimization method for a multi-channel audio system when executed by the processor.

[0078] A fourth aspect of the present invention 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 sound field optimization method for a multi-channel audio system. With respect to the prior art, the present invention has the following beneficial effects:

[0079] The present invention proposes a sound field optimization method for a multi-channel audio system, which successfully solves many challenges faced by traditional multi-channel audio systems. The system first uses a high-precision microphone array and advanced digital signal processing technology to accurately analyze the acoustic characteristics of the room, including reflection, absorption and diffraction. Based on these analysis results, the system uses algorithms such as ray tracing and Kriging interpolation to construct an accurate 3D spatial sound field model. The innovative multi-objective optimization algorithm can simultaneously consider the sound field performance of multiple listening positions, while ensuring the best effect of the main position, taking into account the sound quality of other positions. The adaptive learning mechanism of the system, combined with algorithms such as collaborative filtering and online gradient descent, can dynamically adjust the optimization strategy according to user habits and environmental changes. In terms of real-time processing, the system adopts a parallel computing architecture and an intelligent caching strategy to ensure low latency and efficient signal processing. The user interface design incorporates advanced interactive technologies such as 3D visualization and voice control, which greatly improves the user experience. Through rigorous objective measurement and subjective evaluation, the system shows excellent performance, achieving accurate sound and image positioning and balanced frequency response at multiple positions at the same time. In addition, the open architecture design of the system provides strong support for future expansion and integration. Overall, the system represents a major leap forward in home audio technology, bringing users an unprecedented immersive listening experience.

[0080] The present invention transforms the sound field optimization problem of multiple listening positions into a multi-objective optimization problem, and can simultaneously consider multiple objectives such as the flatness of the frequency response, the linearity of the phase, and the accuracy of the sound image positioning, so as to obtain a solution that performs well in all key performance indicators. This comprehensive optimization method helps to improve the overall listening experience, so that listeners at different positions can enjoy high-quality audio output; the collaborative filtering algorithm is used to build a personalized preference model, which can predict the user's preferences for different audio content based on the user's historical behavior and preferences, which not only improves the intelligence level of the system, but also enhances the user's satisfaction, because the system can automatically adjust the settings to match the user's personal taste; by analyzing the audio content features and using the SVM classifier for scene recognition, the system can This dynamic adjustment mechanism ensures the best auditory effect in different application scenarios; the parallel computing architecture and intelligent caching strategy can effectively reduce processing delays and optimize memory access patterns, thereby improving the system's real-time processing capabilities and response speed; audio analysis instrument measurements and MUSHRA test method sound quality scoring can objectively evaluate system performance and provide a basis for further optimization. This scientific evaluation method helps to continuously improve system design and ensure long-term high standards of sound quality performance. The design of the entire solution is always centered around improving user experience, whether it is achieving better listening effects through multi-objective optimization or meeting users' personalized needs through personalized models and adaptive learning mechanisms. It reflects the user-centric design concept. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 is a schematic flow chart of a sound field optimization method for a multi-channel audio system in this embodiment;

[0082] Figure 2 This is a schematic diagram of the steps of initializing the multi-channel audio system and acquiring acoustic feature data of different environments in S1 of this embodiment;

[0083] Figure 3 FIG. 4 is a schematic diagram of an embodiment of a computer device in this embodiment. DETAILED DESCRIPTION

[0084] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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.

[0085] Embodiment 1

[0086] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1-2 , an embodiment of the present invention provides a sound field optimization method for a multi-channel audio system, comprising:

[0087] S1: Initialize the multi-channel audio system and obtain acoustic feature data of different environments, specifically including the following steps:

[0088] S11: deploying multiple microphones in the listening area of ​​the system to form a microphone array, that is, strategically arranging multiple high-precision microphones in the listening area to form a microphone array; wherein the number and positions of the microphones are determined according to environmental characteristics, and the environmental characteristics include the size and shape of the room.

[0089] S12: Play a full-band logarithmic sweep signal to each channel in turn. The system generates a sweep signal covering the full frequency band of 20Hz-20kHz. The logarithmic sweep method is used to ensure uniform analysis of the entire frequency band. Logarithmic sweep means that the frequency change is not linear, but changes according to a logarithmic scale, which helps to better analyze the low and high frequency parts;

[0090] S13: Synchronously recording the room acoustic response of each channel through the microphone array to capture the propagation and reflection of sound in the room; when measuring the room response, playing the sweep signal through each channel speaker in turn, and the microphone array synchronously recording the room acoustic response of each channel, and repeating the measurement multiple times to improve data reliability.

[0091] During implementation, the system deployed 16 microphones in a 7m×5m×3m living room, forming a 4×4 grid array. For a 5.1-channel system, each channel plays a 30-second logarithmic sweep signal from 20Hz to 20kHz. Each microphone records the response of all channels at the same time, and the measurement is repeated 3 times to ensure the accuracy of the data.

[0092] S2: Based on the ray tracing algorithm, simulate the propagation path of sound waves in the room and identify the measurement data related to the sound lines and reflections in the room, and build a 3D sound field model covering the listening area of ​​the system according to the measurement data; the specific implementation steps are as follows: 1. Analysis of room acoustic characteristics: Use the ray tracing algorithm to simulate the propagation path of sound waves in the room and calculate the position and reflection characteristics of the main reflection surfaces. 2. Transfer function estimation: Based on the measurement data, estimate the transfer function from each speaker to each microphone position. These functions describe how sound propagates from one point to another. Use a parameterized model to describe the transfer function for subsequent optimization. 3. Sound field interpolation: Use the data from the measurement points to build a sound field model for the entire listening area. Use the Kriging interpolation method to achieve continuous sound field representation in space. This method can estimate the value of the point to be estimated based on the value of the known point, and can accurately simulate and predict the sound field distribution in the room.

[0093] During implementation, the system simulated 10,000 sound rays using a ray tracing algorithm and identified the four main reflective surfaces in the room. By analyzing the measurement data of 16 microphone positions, the system estimated the transfer function from the five channels to each microphone, for a total of 80 transfer functions. Using these data points, the system constructed a 3D sound field model covering the entire listening area with a resolution of 10cm×10cm×10cm.

[0094] Specifically, in step S2, simulating the propagation path of the sound wave in the room based on the sound ray tracing algorithm and identifying the measurement data related to the sound ray and the reflection in the room includes:

[0095] Get the incident vector and the surface normal vector, calculate the reflection vector to determine the reflection path of the sound wave, and the implementation formula is as follows: P_reflected = P_incident-2(N·P_incident)N,

[0096] Among them, P_incident is the incident vector, which refers to the direction of propagation of the sound wave before it hits the surface, and N is the surface normal vector. The surface normal vector is a vector perpendicular to the surface and its direction points to the outside of the surface. In acoustics, the normal vector is used to determine how the sound wave interacts with the surface. P_reflected is the reflection vector, which refers to the direction of propagation of the sound wave after it hits the surface. According to the law of reflection, the reflection vector can be calculated from the incident vector and the surface normal vector. This formula is a standard method for calculating the reflection vector. Among them, N·P_incident represents the projection of the incident vector in the direction of the normal vector. The result obtained by multiplying it by 2 is the distance from the incident point to the reflection point, and then multiplying it by the normal vector N to get the reflection vector. The above scheme uses the sound ray tracing algorithm to simulate the propagation path of the sound wave in the room. For each sound ray, its incident vector, surface normal vector and reflection vector are calculated to determine how the sound wave interacts with the surface of the room. It can identify the main reflection surfaces in the room and their positions and reflection characteristics, which is conducive to predicting the acoustic effects in the room, such as echo and reverberation.

[0097] Assume that there is an incident vector P_incident = (1,0,0) (indicating that the sound wave propagates along the positive direction of the x-axis), and a surface normal vector N = (0,1,0) (indicating that the surface is perpendicular to the y-axis). According to the above formula, we can calculate the reflected vector P_reflected:

[0098] P_reflected=(1,0,0)-2((0,1,0)·(1,0,0))(0,1,0),

[0099] P_reflected=(1,0,0)-2(0)(0,1,0),

[0100] P_reflected = (1,0,0). This means that the sound wave is still propagating in the positive x-direction after hitting the surface, because no change in direction has occurred.

[0101] Specifically, in step S2, constructing a 3D sound field model covering the system listening area according to the measurement data includes:

[0102] The transfer function from each loudspeaker to each microphone position is determined based on the least squares estimation according to the measurement data: H(ω)=(X^H*X)^-1*X^H*Y,

[0103] Where X is the input signal matrix, which contains the signals emitted from the speakers; Y is the output signal matrix, which contains the signals received by the microphones; H(ω) is the estimated transfer function, and X^H represents the conjugate transpose of X; Assuming there is a set of input signals and corresponding output signals, by collecting these data and applying the least squares estimation algorithm, the transfer function from each speaker to each microphone position can be obtained. These transfer functions can be used to analyze how sound propagates in space and how to improve the sound quality by adjusting the speaker settings.

[0104] Obtain point data corresponding to the measurement data of the listening area, construct a 3D sound field model covering the listening area according to the point data, and optimize the acoustic characteristics in the room by implementing a continuous sound field representation in space according to the Kriging interpolation method:

[0105] Z(s_0)=∑[i=1ton]λ_i*Z(s_i), that is, the formula:

[0106]

[0107] Among them, Z(s_0) is the model value of the acoustic characteristics of the point s_0 to be estimated, Z(s_i) is the model value of the acoustic characteristics of the known point s_i, λ_i is the weight, and n is the total number of known measurement points. The continuous sound field representation obtained by Kriging interpolation here can be used to analyze and optimize the acoustic characteristics in the room. For example, areas with uneven sound distribution can be identified, or locations that may cause echoes and resonances can be found.

[0108] S3: Based on the GCC-PHAT algorithm, the TDOA estimation results of the channel signals between different microphone positions are calculated, the listening position is obtained and combined with the TDOA estimation results to calculate the optimal delay value and perform corresponding compensation; wherein, the arrival time difference estimation is implemented as follows: perform fast Fourier transform on each channel signal recorded by each microphone to obtain the frequency spectrum of the two signals; calculate the arrival time difference TDOA between the two signals: R_GCC(τ)=IFFT{X_1(ω)X_2^*(ω) / |X_1(ω)X_2^*(ω)|},

[0109] That is, the formula

[0110] Where R_GCC(τ) is a function of the time delay τ, and X_1(ω) and X_2(ω) are the spectra of the two signals.

[0111] In actual operation, considering that there may be multiple listening positions and the importance of each position may be different, a weighted least squares optimization algorithm is introduced to determine the optimal delay parameter. The core idea of ​​this method is to minimize the sum of squares of errors while satisfying all constraints, where the error refers to the difference between the actual arrival time and the target arrival time, that is, in step S3, obtaining the listening position combined with the TDOA estimation result to calculate the optimal delay value and perform corresponding compensation, including:

[0112] By calculating the time difference from different speakers to the listening position, the weighted least squares optimization algorithm is used to calculate the optimal compensation delay, so that the difference between the actual arrival time and the target arrival time of all weighted listening positions is minimized: minimize∑[i=1toM]w_i*||T_i-(d_i+δ_i)||^2, that is, the formula Where M is the number of listening positions, w_i is the weight of the i-th listening position, T_i is the target arrival time at the i-th listening position, d_i is the original delay of the i-th speaker, and δ_i is the compensation delay of the i-th speaker; ||·||: represents the Euclidean norm of the vector, which is used to measure the difference between the actual arrival time and the target arrival time. By adjusting the value of δ_i, the difference between the actual arrival time and the target arrival time of all weighted listening positions is minimized, which helps to ensure that the sound arrives at all important listening positions as simultaneously as possible. By precisely controlling the playback time of different speakers, the distribution of sound in space achieves the desired effect, thereby providing a more immersive and natural listening experience.

[0113] In this embodiment, a fractional delay filter is also designed: a FIR filter is designed to achieve accurate fractional sample delay, and the filter coefficients are constructed using the Lagrange interpolation method. The FIR filter is configured to achieve accurate fractional sample delay, and the filter coefficients are constructed using the Lagrange interpolation method to ensure delay accuracy: h(n) = ∏[k = 0, k ≠ ntoN-1](Dk) / (nk), that is, the formula Where h(n) is the filter response at the nth sample point, D is the required fractional delay, and N is the filter order. This method obtains the response of each sample point by calculating the product of a series of weights, which reflect the relative distances of different sample points relative to the required fractional delay position. The fractional delay filter designed by Lagrange interpolation not only provides an efficient and flexible way to achieve delays with non-integer sample intervals, but also ensures high accuracy and low distortion in the signal processing process. It is one of the indispensable tools in modern digital signal processing.

[0114] During implementation, the system used the GCC-PHAT algorithm to calculate the TDOA of the five channels at 16 microphone positions with an accuracy of 0.01ms. For the three main listening positions (center and sides of the sofa), the system calculated the optimal delay parameters: the center channel delay was 1.2ms, and the left and right surround channels were delayed by 0.8ms and 1.5ms respectively. These delays were implemented using a 64-order FIR filter to ensure a delay accuracy better than 0.1 sample.

[0115] S4: Construct a target mathematical model, transform the sound field optimization problem of multiple listening positions into a multi-objective optimization problem, generate a Pareto optimal solution based on the NSGA-II algorithm, and select a solution that balances the performance of different positions. The specific implementation includes the following steps:

[0116] 1. Target response definition: Define target responses for different listening positions, including frequency response and phase response; define ideal frequency response and phase characteristics based on professional standards (such as THX, Dolby), and consider the special needs of different types of audio content (music, movies, games).

[0117] Determine multiple objective functions f_1(x), f_2(x), ..., f_k(x) according to the flatness of the balanced frequency response, the linearity of the phase, and the accuracy of the sound image localization at multiple listening positions;

[0118] For example, frequency response flatness: Where x is the optimization variable such as the position, orientation or gain of the loudspeaker, and H(f i , x) is the transfer function at the i-th frequency point, H target (f i ) is the target frequency response, N is the number of frequency points, and M is the number of listening positions.

[0119] Phase Linearity: Among them, φ(fi,x) is the phase at the i-th frequency point, and φtarget(fi) is the target phase.

[0120] Sound and image positioning accuracy: Where Δθj(x) is the angular error of the sound image localization at the jth listening position, and θtarget is the target sound image localization angle.

[0121] Minimize multiple objective functions: minimize[f_1(x),f_2(x),...,f_k(x)],

[0122] Constraints: g_i(x)≤0,i=1,2,...,m; h_j(x)=0,j=1,2,...,n; where f_1(x), f_2(x)…f_k(x) are the 1st, 2nd…kth objective functions, and g_i(x) and h_j(x) are constraints, i.e., the maximum gain limit and the maximum movement range limit of the loudspeaker. These constitute the mathematical model of the optimization problem.

[0123] In this embodiment, the Pareto optimal solution of the herbal medicine is solved by using a genetic algorithm or a particle swarm optimization algorithm to solve the multi-objective optimization problem, generate a Pareto frontier, and provide a series of optimal trade-off solutions. That is, in step S4, the Pareto optimal solution is generated based on the NSGA-II algorithm, and the solution that balances the performance of different positions includes:

[0124] Set the initial parameters and initial population P of the algorithm;

[0125] Perform non-dominated sorting on the initial population P to classify the performance of individuals in the initial population on multiple objective functions and identify individuals on the Pareto frontier;

[0126] The crowding distance is determined by calculating the difference in objective function values ​​between adjacent individuals, thereby measuring the local density of individuals in the Pareto frontier; for each objective function, the crowding distance is calculated according to the following steps: 1. Processing boundary individuals: The individuals with the best and worst performance on a certain objective in the population (i.e., boundary individuals) are assigned infinite crowding distances. This is done to ensure that these individuals are always selected in the subsequent selection process. 2. Calculating the crowding distance of internal individuals: For other individuals in the population, the crowding distance is determined by calculating the difference in objective function values ​​between them. The specific steps are as follows: For the sorted population on objective function i, from the second individual to the second to last individual (i.e., excluding boundary individuals), calculate the crowding distance of individual j: dj = dj + (fi,j + 1-fi,j-1), where fi,j is the value of individual j on the i-th objective function, and dj is the crowding distance of individual j. Repeat the above steps for all objective functions. Subsequent accumulation of crowding distances: For each individual, its crowding distance on each objective function is accumulated to obtain the final crowding distance.

[0127] Based on the individual fitness and crowding distance, individuals are selected for crossover and mutation to generate a new generation of population Q;

[0128] Merge the initial population P and the newly generated population Q, and select the new population based on the non-dominated sorting and crowding distance;

[0129] Repeat the above steps, update the Pareto frontier, and iterate to ensure the convergence of the Pareto optimal solution. In the specific implementation, you can use programming languages ​​such as MATLAB and Python, and use optimization toolboxes (such as MATLAB's OptimizationToolbox) or specialized optimization libraries (such as Python's scipy.optimize or deap, etc.) to construct and solve multi-objective optimization problems.

[0130] During implementation, the system defined the target response for three main listening positions, with a frequency response flatness requirement of ±3dB (20Hz-20kHz) and a phase response linearity deviation of less than 45° in the range of 100Hz-10kHz. Using the NSGA-II algorithm, the system generated 100 Pareto optimal solutions, each corresponding to a set of equalizer and delay parameter settings. The system selected a solution that balanced the performance of each position, achieving a frequency response flatness of ±2dB at the main position and ±4dB at the secondary position.

[0131] In this embodiment, it also includes: based on the adaptive learning mechanism combined with collaborative filtering and online gradient descent algorithms, it is possible to dynamically adjust the optimization strategy according to user habits and environmental changes, further including:

[0132] The interaction data between the user and the audio content is obtained to form a user-item rating matrix R, where the interaction data includes the user's ratings of different audio contents, playback history, adjustment settings, and data related to behavioral preferences. In R, the rows represent users, the columns represent audio content or items, and the values ​​of the elements are the user's ratings or preferences for the items. A personalized preference model is constructed using a collaborative filtering algorithm by recording the user's manual adjustment operations and listening habits.

[0133] A personalized preference model is constructed based on a collaborative filtering algorithm, so that similarities between users can be discovered and user preferences can be predicted based on similarities. The collaborative filtering algorithm uses matrix decomposition to decompose the user-item rating matrix R into two target low-dimensional matrices to output potential features;

[0134] Among them, R≈P*Q^T, R is the user-item rating matrix, P is the user factor matrix and Q is the item factor matrix.

[0135] Matrix decomposition of the present embodiment is a powerful technique that is commonly used in recommendation systems and user preference modeling. In the context of adaptive learning and dynamic optimization, matrix decomposition can help the system understand and predict user preferences, thereby achieving a personalized audio experience. The following are detailed steps on how to use matrix decomposition to achieve this goal: 1. Data collection: First, the system needs to collect user interaction data with audio content. This includes user ratings, playback history, adjustment settings, etc. for different audio content. These data constitute the user-item rating matrix R. 2. Matrix decomposition model: The goal of matrix decomposition is to decompose the user-item rating matrix R into two smaller matrices, P (user factor matrix) and Q (item factor matrix), so that R≈PQT. Here, P and Q are usually low-rank matrices, which means that they have only a few non-zero eigenvalues, thereby reducing the number of parameters and simplifying the model. 3. Training model: The model is trained by minimizing the difference between the actual rating matrix R and the estimated rating matrix. The commonly used loss function is the mean square error (MSE), and optimization techniques such as stochastic gradient descent (SGD) and alternating least squares (ALS) can be used to solve this problem. 4. Prediction and recommendation: Once the model is trained, it can be used to predict the user's rating or preference for audio content that has not been heard. This can be achieved by calculating the dot product of the user factor vector and the item factor vector: 5. Dynamic optimization and adaptive learning: As more data is collected, the model can be retrained regularly to reflect the latest user preferences. In addition, the system can dynamically adjust the relevant vectors in P and Q based on the user's real-time feedback (such as adjusting the audio settings) to achieve true adaptive learning.

[0136] During implementation, the system observed that users often increased low-frequency output when watching action movies, automatically incorporating this preference into the optimization strategy for movie mode. By analyzing audio features, the system was able to distinguish between music and movie content with 95% accuracy. During continued use, the system updated the optimization parameters every hour, gradually improving the frequency response flatness of the main position to ±1.5dB, while keeping other positions within ±3.5dB. In this way, matrix decomposition not only helps the system understand complex user preferences, but also enables the system to dynamically adapt to changes in these preferences, providing a more personalized and satisfying user experience.

[0137] Furthermore, this embodiment also includes algorithms such as collaborative filtering and online gradient descent based on an adaptive learning mechanism, which can dynamically adjust the optimization strategy according to user habits and environmental changes, and further includes:

[0138] Collect audio samples from different scenes and annotate each audio sample;

[0139] Extracting target features from the audio sample and using them as inputs of the SVM model, the target features include spectrum features, rhythm features, and time structure features;

[0140] The SVM model is trained using labeled audio samples and extracted features to best distinguish samples from different scene categories;

[0141] Deploy the trained SVM model to the actual audio system. When new audio data arrives, extract the same features, input the features into the SVM model for classification, and predict the scene category to which the audio data belongs based on the input features.

[0142] According to the output prediction result of the SVM model, automatically switch to the corresponding optimization strategy, wherein when it is identified that the user is in a video watching mode, the low-frequency effect is enhanced to provide a better viewing experience; when it is identified that the user is in a music listening mode, the equalizer setting is adjusted to highlight the rhythm of the music;

[0143] The SVM model parameters are continuously updated through online learning to adapt to new data and changes in user preferences.

[0144] This embodiment supports the classification of the following vector machines (SVM):

[0145] f(x)=sign(∑[i=1ton]α_iy_iK(x_i,x)+b)

[0146] Among them, K(x_i,x) is the kernel function, α_i is the Lagrange multiplier, and y_i is the category label.

[0147] This embodiment uses online learning and parameter updating: real-time monitoring of system performance and user feedback, and continuous optimization of system parameters using online learning algorithms. Online gradient descent: Among them, θ_t is the parameter vector, η_t is the learning rate, and L_t is the loss function.

[0148] Through the above steps, SVM classification can be used to effectively identify audio scenes and dynamically adjust audio processing strategies based on the recognition results, thereby providing a more personalized and optimized user experience.

[0149] In this embodiment, a parallel computing architecture and intelligent caching strategy are also used for real-time processing and performance optimization, further comprising:

[0150] Parallel computing architecture design: configure multi-core DSP or FPGA for parallel signal processing, and through the pipeline structure, allocate different processing tasks to different processor cores to minimize processing delays; it can be understood as using multi-core DSP or FPGA to implement parallel signal processing, design pipeline structure, and minimize processing delays. In task scheduling, the following formula is used to calculate the total processing time T_total: T_total = max(T_i + C_i,j / P_j), where T_i is the start time of task i, C_i,j is the execution time of task i on processor j, and P_j is the performance of processor j;

[0151] Cache optimization: Preload frequently used filter coefficients into fast-access memory based on intelligent caching strategies, and use the principle of locality to optimize memory access patterns, that is, store frequently accessed data in memory with close physical locations to reduce memory access latency. Use the least recently used (LRU) cache replacement algorithm to remove the least recently used items from the cache and retain the most frequently used items.

[0152] Adaptive sampling rate conversion: According to the changes in audio content and system load, the internal processing sampling rate is dynamically adjusted according to the polyphase filter bank, and the internal processing sampling rate is dynamically adjusted according to the audio content and system load to achieve a high-quality sampling rate conversion algorithm and minimize sound quality loss. This embodiment uses a polyphase filter bank (PolyphaseFilterBank): y(n) = ∑[k = 0 to N-1] h_k(n) * x(nL-k), where h_k is a subfilter and L is the sampling rate conversion ratio.

[0153] During implementation, the system uses an 8-core DSP chip to distribute the signal processing tasks of 5.1 channels to different cores. Through parallel processing, the system controls the total delay within 5ms. The 16KB filter coefficient cache is managed using the LRU strategy to store frequently used equalizer and delay filter coefficients in fast access memory. When processing 24bit / 192kHz high-resolution audio, the system uses a sampling rate of 384kHz to provide sufficient digital signal processing margin.

[0154] This implementation is for user interface and interaction design, and the implementation steps are as follows: 1. Visual sound field presentation: Use 3D graphics technology to visualize the current sound field distribution, provide an interactive interface, and allow users to intuitively adjust the sound field. The algorithm used: Isosurface rendering, using the MarchingCubes algorithm to convert 3D sound field data into a visual grid. 2. Intelligent recommendation system: Based on the current audio content and user historical preferences, recommend the best sound field settings, and use the reinforcement learning algorithm to continuously optimize the recommendation strategy. The algorithm used: Context-aware recommendation system: R(u,i|c)=f(u,i,c) where u is the user, i is the item, c is the context information, and f is the recommendation function. 3. Voice control interface: Integrate natural language processing functions, support voice command control, implement personalized voice recognition models, and improve command recognition accuracy. The algorithm used: Sequence to Sequence Model (Seq2Seq): Use the encoder-decoder architecture to convert voice input into system commands.

[0155] When implemented, the system provides a 3D touch screen interface where users can adjust the sound field by dragging virtual sound sources. Based on the user's listening history, the system automatically recommends a sound field setting that emphasizes the sense of space when playing jazz music. Through the voice command "Optimize the sound of the front seats", the system automatically adjusts the optimization weight to favor the sound field performance of the front seats.

[0156] This implementation evaluates and continuously optimizes the system, and the implementation steps are as follows:

[0157] The target parameters of the system are measured by an audio analysis instrument for performance evaluation. The target parameters include frequency response, phase response and group delay parameters. The audio analysis instrument uses a logarithmic swept frequency signal to measure the transfer function of the system. It can be understood as the first step, objective performance measurement: using a professional audio analysis instrument to measure the system's frequency response, phase response, group delay and other parameters, and perform multi-point measurements to evaluate the uniformity of the sound field in the entire listening area. The algorithm used is the swept frequency measurement method: using a logarithmic swept frequency signal to measure the transfer function of the system: H(ω) = Y(ω) / X(ω), where Y(ω) is the spectrum of the output signal and X(ω) is the spectrum of the input signal.

[0158] The sound quality of a series of audio samples is scored using the MUSHRA test method. The scoring formula is: score = (original score - low anchor score) / (reference score - low anchor score) * 100%. The low anchor score refers to the subjective score obtained by the audio sample that serves as the worst quality reference in MUSHRA. It can be understood as the second step, subjective listening evaluation: organize professional audio engineers and ordinary users to conduct double-blind tests, using standardized scoring methods such as the MUSHRA test. The third step is long-term stability test: conduct long-term continuous operation tests to evaluate the stability and reliability of the system, simulate various extreme usage scenarios, and test the robustness of the system.

[0159] During implementation, the system was swept at 9 locations in a standardized test room. The results showed that the frequency response flatness of the entire listening area was within ±2.5dB, and the linearity deviation of the phase response in the range of 100Hz-10kHz was less than 30°. In the MUSHRA test of 30 people, the system received an average subjective score of 88 points, significantly higher than the 76 points of the traditional fixed parameter system. In the two-month long-term stability test, the system ran continuously for 1440 hours without failure, and successfully coped with extreme scenarios including frequent volume changes and switching between multiple audio formats.

[0160] This embodiment also integrates and expands the ecosystem, and the implementation steps are as follows: 1. Multi-room collaborative optimization: Supports networking of audio systems in multiple rooms to achieve cross-room audio synchronization and overall sound field optimization. Use ADMM (Alternating Direction Method of Multipliers) algorithm for multi-room joint optimization:

[0161] Update x: x^k+1=argmin_x(f(x)+(ρ / 2)||xz^k+u^k||_2^2), that is

[0162] Update z:z^k+1=argmin_z(g(z)+(ρ / 2)||x^k+1-z+u^k||_2^2), that is

[0163] Update u:u^k+1=u^k+x^k+1-z^k+1, that is, u k+1 =u k +x k+1 -z k+1 ,

[0164] Here, x is the optimization variable, which usually represents a decision variable or part of the system state. In the context of multi-room sound field optimization, x may represent the acoustic parameter settings of a certain room, such as speaker volume, equalizer settings, etc. z is another optimization variable, which is usually associated with x, but may represent different system aspects or constraints. In a multi-room environment, z may represent the acoustic parameters of another room or other related control variables. u is a scaling factor, also known as a Lagrange multiplier, which is used to adjust the consistency between x and z. In each iteration, u is updated based on the latest values ​​of x and z. In this way, ADMM not only simplifies the problem solving process, but also improves computational efficiency, especially for large-scale or distributed optimization problems.

[0165] 2. Third-party device integration: Provide open APIs to allow third-party devices (such as smart speakers) to access the system to achieve cross-device audio optimization and seamless switching.

[0166] 3. Cloud services and updates: Establish a cloud database to store optimized parameters for different rooms and device configurations, provide online firmware updates, and continuously improve algorithm performance.

[0167] During implementation, the system successfully integrated audio equipment in the living room, bedroom, and study, achieving collaborative optimization of audio throughout the house. Through open APIs, the system successfully connected to Amazon Echo and Google Home smart speakers and provided optimized audio output for these devices. The system connects to the cloud server every week to download the latest algorithm updates and upload anonymized performance data for further improvement of the optimization algorithm.

[0168] The embodiments of the present invention effectively solve many challenges faced by traditional multi-channel audio systems through innovative algorithm design and system architecture. It not only significantly improves the listening experience at a single location, but also optimizes the sound field in the entire space, bringing users an unprecedented immersive audio experience. The system's adaptive learning ability and real-time optimization characteristics enable it to adapt to various acoustic environments and user needs, representing a major leap in home audio system technology. Through rigorous performance testing and user evaluation, the system has demonstrated excellent performance and stability, and is expected to be widely used in the high-end audio market and provide strong support for the development of future smart home ecosystems.

[0169] Embodiment 2

[0170] Based on the same inventive concept, this embodiment provides a sound field optimization device for a multi-channel audio system, comprising:

[0171] An initialization module, used to initialize the multi-channel audio system and obtain acoustic feature data of different environments;

[0172] An acquisition module, used to simulate the propagation path of sound waves in the room based on a sound ray tracing algorithm and identify the measurement data related to the reflection of the sound ray in the room, and to construct a 3D sound field model covering the listening area of ​​the system according to the measurement data;

[0173] A calculation module, used to calculate the TDOA estimation result of the channel signal between different microphone positions based on the GCC-PHAT algorithm, obtain the listening position and calculate the optimal delay value in combination with the TDOA estimation result, and perform corresponding compensation;

[0174] The optimization module is used to construct a target mathematical model, transform the sound field optimization problem of multiple listening positions into a multi-objective optimization problem, generate a Pareto optimal solution based on the NSGA-II algorithm, and select a solution that balances the performance of different positions. The principle is shown in Example 1 and will not be repeated here.

[0175] Based on the same inventive concept, this embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that when the computer program is executed by the processor, the steps of the sound field optimization method of the multi-channel audio system described in any one of the above items are implemented.

[0176] Based on the same inventive concept, this embodiment provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods for optimizing the sound field of a multi-channel audio system are implemented.

[0177] The computer device in this embodiment is described in detail from the perspective of hardware processing.

[0178] See also Figure 3 As shown, the computer device includes a processor 100 and a memory 101. The memory 101 stores machine executable instructions that can be executed by the processor 100. The processor 100 executes the machine executable instructions to implement the sound field optimization method of the multi-channel audio system.

[0179] further, Figure 3 The computer device shown further includes a bus 102 and a communication interface 103 , and the processor 100 , the communication interface 103 and the memory 101 are connected via the bus 102 .

[0180] Among them, the memory 101 may include a high-speed random access memory (Random Access Memory, RAM), and may also include a non-volatile memory (non-volatile memory), for example, at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 103 (which can be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used. The bus 102 can be an ISA bus, a PCI bus or an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0181] The processor 100 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor 100 or an instruction in the form of software. The above processor 100 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present disclosure may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present disclosure may be directly embodied as being executed by a hardware decoding processor, or may be executed by a combination of hardware and software modules in a decoding processor. The software module may be located in a storage medium mature in the art, such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 101, and the processor 100 reads the information in the memory 101 and completes the method steps of the above-mentioned embodiment in combination with its hardware.

[0182] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the sound field optimization method for a multi-channel audio system.

[0183] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0184] 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 invention, in essence, 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 for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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 invention.

Claims

1. A method for optimizing the sound field of a multi-channel audio system, characterized in that: include: Initialize the multi-channel audio system and obtain acoustic characteristic data of different environments; Simulating the propagation path of sound waves in the room based on a ray tracing algorithm and identifying measurement data related to the reflection of the sound rays in the room, and constructing a 3D sound field model covering the listening area of ​​the system based on the measurement data; Calculate the TDOA estimation result of the channel signal between different microphone positions based on the GCC-PHAT algorithm, obtain the listening position and calculate the optimal delay value in combination with the TDOA estimation result, and perform corresponding compensation; A target mathematical model is constructed to transform the sound field optimization problem of multiple listening positions into a multi-objective optimization problem. The Pareto optimal solution is generated based on the NSGA-II algorithm, and the solution that balances the performance of different positions is selected.

2. The sound field optimization method of a multi-channel audio system according to claim 1, characterized in that: Initializing the multi-channel audio system and acquiring acoustic feature data of different environments includes: Deploy multiple microphones within the listening area of ​​the system to form a microphone array; Play a logarithmic sweep signal covering the full frequency band to each channel in turn; Synchronously recording the room acoustic response of each channel through the microphone array to capture the propagation and reflection of sound in the room; The number and positions of the microphones are determined according to environmental characteristics, including the size and shape of the room.

3. The sound field optimization method of a multi-channel audio system according to claim 1, characterized in that: The method of simulating the propagation path of sound waves in the room based on the sound ray tracing algorithm and identifying the measurement data related to the sound ray and the reflection in the room includes: Get the incident vector and the surface normal vector, calculate the reflection vector to determine the reflection path of the sound wave, and the implementation formula is as follows: P_reflected = P_incident-2(N·P_incident)N, Among them, P_incident is the incident vector, N is the surface normal vector, and P_reflected is the reflected vector.

4. The sound field optimization method of a multi-channel audio system according to claim 3, characterized in that: The constructing of a 3D sound field model covering the listening area of ​​the system according to the measurement data comprises: The transfer function from each loudspeaker to each microphone position is determined based on the least squares estimation according to the measurement data: H(ω)=(X^H*X)^-1*X^H*Y, Where X is the input signal matrix, Y is the output signal matrix, and H(ω) is the estimated transfer function; Obtain point data corresponding to the measurement data of the listening area, construct a 3D sound field model covering the listening area according to the point data, and optimize the acoustic characteristics in the room by implementing a continuous sound field representation in space according to the Kriging interpolation method: Z(s_0)=∑[i=1ton]λ_i*Z(s_i), Among them, Z(s_0) is the model value of the acoustic characteristics of the point s_0 to be estimated, Z(s_i) is the model value of the acoustic characteristics of the known point s_i, λ_i is the weight, and n is the total number of known measurement points.

5. The sound field optimization method of a multi-channel audio system according to claim 1, characterized in that: The TDOA estimation result of the channel signal between different microphone positions calculated based on the GCC-PHAT algorithm includes: Perform fast Fourier transform on each channel signal recorded by each microphone to obtain the spectrum of the two signals; Calculate the time difference of arrival TDOA between two signals: R_GCC(τ)=IFFT{X_1(ω)X_2^*(ω) / |X_1(ω)X_2^*(ω)|}, Where R_GCC(τ) is a function of the time delay τ, and X_1(ω) and X_2(ω) are the spectra of the two signals.

6. The sound field optimization method of a multi-channel audio system according to claim 5, characterized in that: The obtaining of the listening position and combining the TDOA estimation result to calculate the optimal delay value and perform corresponding compensation comprises: By calculating the time difference from different speakers to the listening position, a weighted least squares optimization algorithm is used to calculate the optimal compensation delay, so that the difference between the actual arrival time and the target arrival time of all weighted listening positions is minimized: minimize∑[i=1toM]w_i*||T_i-(d_i+δ_i)||^2, Where M is the number of listening positions, w_i is the weight of the i-th listening position, T_i is the target arrival time at the i-th listening position, d_i is the original delay of the i-th speaker, and δ_i is the compensated delay of the i-th speaker; Configure the FIR filter to achieve accurate fractional sample delay, and use Lagrange interpolation to construct the filter coefficients to ensure delay accuracy: h(n)=∏[k=0,k≠ntoN-1](Dk) / (nk), Where h(n) is the filter response at the nth sample point, D is the desired fractional delay, and N is the filter order.

7. The sound field optimization method of a multi-channel audio system according to claim 1, characterized in that: The objective mathematical model is constructed to transform the sound field optimization problem of multiple listening positions into a multi-objective optimization problem, including: defining target responses at different listening positions, the target responses including frequency response and phase response; Determine multiple objective functions f_1(x), f_2(x), ..., f_k(x) according to the flatness of the balanced frequency response, the linearity of the phase, and the accuracy of the sound image localization at multiple listening positions; Minimize multiple objective functions: minimize[f_1(x),f_2(x),...,f_k(x)], Constraints: g_i(x)≤0,i=1,2,...,m; h_j(x)=0,j=1,2,...,n; Among them, f_1(x), f_2(x)…f_k(x) are the 1st, 2nd…kth objective functions, and g_i(x) and h_j(x) are constraints, namely the maximum gain limit and the maximum movement range limit of the speaker.

8. The sound field optimization method of a multi-channel audio system according to claim 1, characterized in that: The Pareto optimal solution is generated based on the NSGA-II algorithm, and the solution for balancing the performance of different positions is selected, including: Set the initial parameters and initial population P of the algorithm; Perform non-dominated sorting on the initial population P to classify the performance of individuals in the initial population on multiple objective functions and identify individuals on the Pareto frontier; The crowding distance is determined by calculating the difference in the objective function values ​​between adjacent individuals, thereby measuring the local density of individuals in the Pareto frontier. Based on the individual fitness and crowding distance, individuals are selected for crossover and mutation to generate a new generation of population Q; Merge the initial population P and the newly generated population Q, and select the new population based on the non-dominated sorting and crowding distance; Repeat the above steps, update the Pareto frontier, and iterate to ensure the convergence of the Pareto optimal solution.

9. The sound field optimization method of a multi-channel audio system according to claim 1, characterized in that: Also includes: Based on the adaptive learning mechanism combined with collaborative filtering and online gradient descent algorithms, it can dynamically adjust the optimization strategy according to user habits and environmental changes, further including: Obtain the interaction data between the user and the audio content to form a user-item rating matrix R, wherein the interaction data includes the user's ratings of different audio contents, playback history, adjustment settings, and data related to behavioral preferences. In R, the rows represent users, the columns represent audio contents or items, and the values ​​of the elements are the user's ratings or preferences for the items. A personalized preference model is constructed based on a collaborative filtering algorithm, so that similarities between users can be discovered and user preferences can be predicted based on similarities. The collaborative filtering algorithm uses matrix decomposition to decompose the user-item rating matrix R into two target low-dimensional matrices to output potential features; Among them, R≈P*Q^T, R is the user-item rating matrix, P is the user factor matrix and Q is the item factor matrix.

10. The sound field optimization method of a multi-channel audio system according to claim 1, characterized in that: Also includes: Based on the adaptive learning mechanism combined with collaborative filtering and online gradient descent algorithms, it can dynamically adjust the optimization strategy according to user habits and environmental changes, further including: Collect audio samples from different scenes and annotate each audio sample; Extracting target features from the audio sample and using them as inputs of the SVM model, the target features include spectrum features, rhythm features, and time structure features; The SVM model is trained using labeled audio samples and extracted features to best distinguish samples from different scene categories; Deploy the trained SVM model to the actual audio system. When new audio data arrives, extract the same features, input the features into the SVM model for classification, and predict the scene category to which the audio data belongs based on the input features. According to the output prediction result of the SVM model, automatically switch to the corresponding optimization strategy, wherein when it is identified that the user is in a video watching mode, the low-frequency effect is enhanced to provide a better viewing experience; when it is identified that the user is in a music listening mode, the equalizer setting is adjusted to highlight the rhythm of the music; The SVM model parameters are continuously updated through online learning to adapt to new data and changes in user preferences.

11. The sound field optimization method of a multi-channel audio system according to claim 1, characterized in that: Also includes: Use parallel computing architecture and intelligent caching strategies for real-time processing and performance optimization, further including: Configure multi-core DSP or FPGA for parallel signal processing, and use pipeline structure to distribute different processing tasks to different processor cores to minimize processing delay; In task scheduling, the following formula is used to calculate the total processing time T_total: T_total=max(T_i+C_i,j / P_j), Where T_i is the start time of task i, C_i,j is the execution time of task i on processor j, and P_j is the performance of processor j; and / or, Preloading frequently used filter coefficients into fast-access memory based on intelligent caching strategies, using the principle of locality to optimize memory access patterns, i.e. storing frequently accessed data in physically close memory to reduce memory access latency; and / or, Dynamically adjusts internal processing sampling rate based on polyphase filter banks as audio content and system load change.

12. The sound field optimization method of a multi-channel audio system according to claim 1, characterized in that: Also includes: Evaluate multi-channel audio systems, including: The target parameters of the system are measured by an audio analysis instrument to perform performance evaluation, wherein the target parameters include frequency response, phase response and group delay parameters, wherein the audio analysis instrument uses a logarithmic swept frequency signal to measure the transfer function of the system; The sound quality of a series of audio samples was scored using the MUSHRA test method. The scoring formula is: score = (original score - low anchor score) / (reference score - low anchor score) * 100%, where the low anchor score refers to the subjective score obtained for the audio sample with the worst quality reference in MUSHRA.

13. A sound field optimization device for a multi-channel audio system, characterized in that: include: An initialization module, used to initialize the multi-channel audio system and obtain acoustic feature data of different environments; An acquisition module, used to simulate the propagation path of sound waves in the room based on a sound ray tracing algorithm and identify the measurement data related to the reflection of the sound ray in the room, and to construct a 3D sound field model covering the listening area of ​​the system according to the measurement data; A calculation module, used to calculate the TDOA estimation result of the channel signal between different microphone positions based on the GCC-PHAT algorithm, obtain the listening position and calculate the optimal delay value in combination with the TDOA estimation result, and perform corresponding compensation; The optimization module is used to construct the target mathematical model, transform the sound field optimization problem of multiple listening positions into a multi-objective optimization problem, generate the Pareto optimal solution based on the NSGA-II algorithm, and select the solution that balances the performance of different positions.

14. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is executed by a processor, the method steps for optimizing the sound field of a multi-channel audio system according to any one of claims 1 to 12 are implemented.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method steps for optimizing the sound field of a multi-channel audio system according to any one of claims 1 to 12 are implemented.

Images