Method for changing sound wave orientation and loudspeaker

Abstract

The invention discloses a method for changing sound wave orientation and a loudspeaker. The method comprises the following steps: acquiring environmental noise data acquired by a multi-channel acoustic sensor array and target audience coordinate information extracted by an infrared positioning module; constructing a noise distribution model based on the environment noise and the audience coordinate information; inputting the target audience coordinate into a pre-constructed sound field propagation model, and calculating a space directional angle adjustment parameter and a frequency band gain compensation parameter of the audio driving unit; jointly inputting the noise distribution model and the space directional angle adjustment parameter and the frequency band gain compensation parameter of the audio driving unit into a pre-trained neural network model, and outputting a rotation angle of a loudspeaker rotation mechanism and a frequency band gain value of a loudspeaker power amplification circuit; and based on the model output parameters, controlling a mechanical rotation angle of a loudspeaker rotation mechanism, and synchronously adjusting a frequency band gain value of a loudspeaker power amplification circuit. According to the method, accurate noise suppression and sound field directional optimization can be carried out on a complex acoustic environment, and the audio quality and audience experience are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of variable sound wave orientation, and particularly to a method for variable sound wave orientation and a loudspeaker. Background Art

[0002] With the rapid development of modern acoustic technologies, people have put forward higher and higher requirements for the performance and functions of audio systems. In diversified application scenarios, the limitations of traditional audio systems with fixed sound wave orientation have gradually emerged, giving rise to the variable sound wave orientation technology. In traditional audio systems, the sound wave orientation is fixed and cannot be flexibly adjusted according to the actual scenario. This results in that in stage performances, speakers with fixed orientation are difficult to cover moving actors and audiences, affecting the sound effect; car audio systems also cannot optimize the sound directivity according to the positions of passengers, reducing the experience. To solve these problems, the variable sound wave orientation technology integrates multi-field technologies such as infrared positioning, phased array, adaptive filtering, and sound field propagation models. The infrared positioning module accurately obtains the coordinates of the target audience, providing a spatial position basis for sound wave orientation adjustment; the phased array technology realizes rapid and accurate changes in the sound wave direction by controlling the phase and amplitude of the sound source; the adaptive filtering technology analyzes and processes environmental noise in real time, generates a filter coefficient matrix, and selectively attenuates the acoustic signal to ensure the clarity of the target acoustic signal; the sound field propagation model simulates the propagation characteristics of sound waves in a medium, calculates the spatial pointing angle and frequency band gain compensation parameters of the audio drive unit, and realizes accurate sound field coverage of the target audience area.

[0003] The variable sound wave orientation technology shows great application potential in multiple fields. In stage performances, it tracks the positions of actors and audiences in real time and dynamically adjusts the orientation of the speakers to ensure the best sound effect; car audio systems optimize the sound directivity according to the positions and needs of passengers, providing a personalized audio experience; public address systems project sound accurately, reduce interference, and improve the clarity and effect of the broadcast.

[0004] However, this technology also faces many challenges. In a complex acoustic environment, there are huge challenges in precisely suppressing environmental noise and simultaneously optimizing the directional propagation of the sound field. Traditional noise suppression systems often use fixed parameters, making it difficult to adapt to the dynamically changing noise characteristics and prone to excessive attenuation of useful signals. And simple sound field optimization cannot effectively isolate background noise, affecting the listener experience. How to achieve a dynamic balance between noise suppression and sound field optimization in a changing acoustic environment has become a key problem to be solved urgently. This involves the real-time fusion and processing of multi-dimensional data, including environmental noise feature extraction, target listener positioning, sound field propagation modeling, etc. At the same time, due to the complexity and uncertainty of the acoustic environment, the system also needs to have an adaptive learning ability to continuously optimize parameters according to environmental changes and listener feedback. In addition, while achieving precise noise suppression, how to ensure the integrity and clarity of useful audio signals and avoid distortion caused by overprocessing is also a technical difficulty that needs to be weighed. These factors are interrelated and mutually restrictive, constituting a complex system of technical problems. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to propose a method and a speaker with variable sound wave orientation, which can achieve precise noise suppression and sound field directional optimization for a complex acoustic environment, significantly improving the audio quality and listener experience.

[0006] According to one aspect of the present invention, a method with variable sound wave orientation is provided, which acquires environmental noise data collected by a multi-channel acoustic sensor array and target listener coordinate information extracted by an infrared positioning module;

[0007] Based on the environmental noise and listener coordinate information, a noise distribution model is constructed; the target listener coordinates are input into a pre-constructed sound field propagation model to calculate the spatial pointing angle adjustment parameter and frequency band gain compensation parameter of the audio drive unit;

[0008] The noise distribution model, the spatial pointing angle adjustment parameter and the frequency band gain compensation parameter of the audio drive unit are jointly input into a pre-trained neural network model to output the rotation angle of the speaker rotation mechanism and the frequency band gain value of the speaker power amplifier circuit;

[0009] Based on the model output parameters, the mechanical rotation angle of the speaker rotation mechanism is controlled, and the frequency band gain value of the speaker power amplifier circuit is synchronously adjusted.

[0010] In the above technical solution, the method integrates the environmental noise data collected by the multi-channel acoustic sensor array and the target listener coordinate information obtained by the infrared positioning module. The multi-channel acoustic sensor array can capture environmental noise from multiple angles and provide spatial noise distribution data; the infrared positioning module accurately obtains the listener's position, providing a basis for adjusting the sound wave orientation. The multi-modal data fusion enables the system to comprehensively perceive the acoustic environment and the listener's position, laying a foundation for optimizing the sound wave orientation.

[0011] Based on the noise distribution model constructed from the above data, it accurately depicts the intensity, distribution of environmental noise, and the relative position with the listener, covering both static and dynamic noise sources, providing a detailed "noise map" for sound wave propagation path planning and sound field adjustment, and helping to identify the noise areas that have the greatest impact on the listener's auditory experience.

[0012] By inputting the listener coordinates into the pre-constructed sound field propagation model, the adjustment parameters of the audio drive unit can be quickly calculated. This model is based on the propagation characteristics of sound waves in different media and spatial structures, predicting the sound wave propagation path, attenuation, and interaction with obstacles, enabling the system to determine in advance the adjustment of the sound wave emission direction and frequency response, ensuring that the sound wave reaches the listener's position efficiently and accurately.

[0013] By inputting the noise distribution model and the adjustment parameters into the pre-trained neural network model, deep integration of multi-source information and intelligent decision-making are achieved. The neural network model learns the complex relationships between data by learning a large amount of training data and outputs the optimal speaker control parameters. This artificial intelligence optimization strategy improves the accuracy and efficiency of sound wave orientation adjustment, adaptively generates personalized sound wave control schemes, and enhances the intelligence and user experience of the system.

[0014] The sound field propagation model is constructed based on physical principles and provides relatively accurate parameter calculation results, but the actual acoustic environment is more complex. The neural network model is trained with a large amount of actual data to learn the actual mapping relationships in complex environments. Using the results of the sound field propagation model as the input of the neural network can optimize parameter prediction on a theoretical basis and improve accuracy. The neural network can simultaneously accept multiple inputs such as the sound field propagation model and the noise distribution model, comprehensively consider the influence of various factors on the speaker control parameters, analyze the acoustic environment comprehensively, and output parameter control that meets the actual needs, realizing the intelligent and automated operation of the system.

[0015] This method can adjust the sound wave direction and frequency band gain in real time according to the changes in environmental noise and the movement of the listener's position. Whether the listener moves or the noise source changes, the system responds quickly, recalculates and optimizes the sound wave propagation parameters to ensure the accurate pointing of the sound wave and the transmission of high-quality sound effects. By controlling the mechanical rotation angle of the speaker rotation mechanism and the frequency band gain value of the power amplifier circuit, the coordination of mechanical movement and electronic signal processing is realized. The mechanical rotation adjusts the sound wave propagation direction, and the frequency band gain optimizes the sound wave energy distribution, which improves the flexibility and accuracy of sound wave direction adjustment, compensates for the sound wave distortion and energy loss caused by mechanical movement, and further improves the audio output quality and stability. This method can customize the sound wave propagation scheme for the listener, meet the personalized audio quality requirements, and improve the satisfaction. In the fields of home theater, conference system, public address and virtual reality, etc., this technology has broad application potential and can optimize the sound output and improve the user experience.

[0016] In some embodiments, a noise distribution model is constructed based on environmental noise and listener coordinate information, including:

[0017] Perform a fast Fourier transform on the environmental noise data to obtain the noise spectrum characteristics;

[0018] Extract the energy peaks from the noise spectrum characteristics to determine the main frequency band set and its intensity distribution parameters;

[0019] Perform a spatial mapping of the target listener coordinates and the main frequency band set to establish a noise distribution model.

[0020] In the above technical solution, a fast Fourier transform (FFT) is performed on the environmental noise data to efficiently convert the time-domain signal into a frequency-domain signal, obtaining the noise spectrum characteristics. The FFT algorithm significantly speeds up the calculation by reducing the computational complexity of the discrete Fourier transform (DFT), enabling this method to quickly extract frequency-domain information from a large amount of noise data, laying a foundation for subsequent noise analysis and processing. Extracting the energy peaks from the noise spectrum characteristics can accurately identify the frequency components with higher energy, which usually correspond to the main noise sources. Determining the main frequency band set and its intensity distribution parameters can simplify the complex noise spectrum into the key frequency range and its intensity information, reducing the data processing complexity and focusing on the dominant noise frequency components.

[0021] Perform a spatial mapping of the target listener coordinates and the main frequency band set, combining the noise frequency characteristics and spatial position information to construct a model reflecting the spatial distribution of noise. This mapping fully considers the noise frequency characteristics and the listener's position, making the noise distribution model more accurately describe the impact on the target listener.

[0022] The noise distribution model constructed by the above steps can accurately depict the distribution of noise intensities at different positions in a specific spatial environment on the main frequency band. This model helps the system understand the noise distribution situation, identify the areas and frequencies that have the greatest impact on the listener's auditory experience, provides a reliable basis for sound wave propagation path planning and sound field adjustment, and enables the system to formulate an optimal sound wave orientation strategy according to the actual environment and the listener's position.

[0023] In some embodiments, the inputting the target listener coordinates into the sound field propagation model and calculating the spatial pointing angle adjustment parameter and frequency band gain compensation parameter of the audio driving unit includes:

[0024] Obtaining spatial position data by using the target listener coordinates and calculating the spatial pointing angle adjustment parameter of the audio driving unit through the sound field propagation model;

[0025] Calculating the frequency band gain compensation parameter according to the adjusted sound wave deflection angle;

[0026] If the frequency band gain compensation parameter meets the preset condition, generating the final audio output configuration;

[0027] Adjusting the spatial pointing angle and the frequency band gain compensation parameter of the audio driving unit according to the final audio output configuration.

[0028] In the above technical solution, the method obtains spatial position data through the target listener coordinates, providing an accurate directional target for the sound field propagation model. This coordinate information can clarify the specific position of the listener in the three-dimensional space, enabling the system to calculate the optimal spatial pointing angle of the audio driving unit accordingly. Compared with the traditional fixed pointing or approximate area pointing, this method can perform sound wave orientation more accurately, ensuring that the sound wave propagates directly and efficiently to the target listener position, reducing energy loss and interference to non-target areas.

[0029] Calculating the spatial pointing angle adjustment parameter by using the sound field propagation model gives full play to the advantages of acoustic theory and the model. Based on the propagation characteristics of sound waves in different media, as well as physical phenomena such as reflection, refraction, and diffraction, this model can accurately predict the propagation path and energy distribution of sound waves from the audio driving unit to the target listener. Based on the calculation results of the model, the system can determine in advance how to adjust the pointing angle of the audio driving unit so that the sound wave reaches the target listener with the optimal path and intensity, improving the audio transmission efficiency and quality.

[0030] When the propagation direction of the sound wave changes, the attenuation degrees of sound waves with different frequencies are different. By calculating the frequency band gain compensation parameter according to the deflection angle, the system can perform corresponding gain adjustment for different frequency components to compensate for the energy loss caused by the change in the propagation path. This ensures that the audio signal received by the target listener has appropriate intensity and clarity in each frequency band, improving the audio fidelity and auditory experience.

[0031] Set preset conditions to determine whether the frequency band gain compensation parameter meets the requirements. Only when the parameter meets the conditions can the final audio output configuration be generated. This mechanism ensures system stability and output quality. The preset conditions include parameter range limits, smoothness requirements, etc., to avoid audio distortion or noise amplification caused by unreasonable gain compensation. The final audio output configuration generated after meeting the conditions comprehensively considers the optimized results of the spatial pointing angle and frequency band gain compensation, and provides precise control instructions for the audio drive unit to ensure its audio output as expected.

[0032] The whole process constitutes a closed-loop control system. From obtaining the target listener coordinates, calculating the adjustment parameters, generating the audio output configuration to adjusting the parameters, and finally continuously monitoring and correcting the system through a feedback mechanism (such as obtaining environmental noise data and target listener coordinates again, etc.). This closed-loop control method effectively ensures the stability and reliability of the system, can detect and correct deviations or errors in a timely manner, and makes the audio output always meet the design requirements and quality standards.

[0033] In some embodiments, the neural network model employs a multi-layer perceptron.

[0034] In the above technical solution, the system uses a basic feedforward neural network, the multi-layer perceptron (MLP), for parameter prediction. The MLP consists of an input layer, one or more hidden layers, and an output layer, and is suitable for classification and regression problems. In this system, the calculation results of the sound field propagation model, the noise distribution model, and the listener coordinate information, etc. are used as the input features of the MLP. These features are non-linearly transformed and combined through the multi-layer neurons of the MLP, and finally the control parameters of the speaker are obtained at the output layer.

[0035] The MLP has the advantages of simple structure, easy implementation and training. It can comprehensively process and learn various input features, and provide effective parameter prediction for the system. Compared with other neural networks, the MLP shows advantages in real-time processing. For example, although the long short-term memory network (LSTM) is suitable for processing sequence data, its complex gating mechanism and cyclic structure result in a slow inference speed; the convolutional neural network (CNN) performs excellently in processing two-dimensional data such as images or spectrograms, but in the acoustic parameter prediction task, the MLP is more efficient.

[0036] The MLP has a simple structure, high computational efficiency, and a direct feedforward mechanism, with good real-time processing capabilities. It is suitable for acoustic signal processing scenarios that require fast response, can provide the control parameters of the speaker for the system in a timely manner, and ensure the fast and accurate adjustment of the sound wave direction.

[0037] In some embodiments, based on the model output parameters, control the mechanical rotation angle of the speaker rotation mechanism, synchronously adjust the frequency band gain value of the speaker power amplifier circuit, and then further include:

[0038] Collect the actual deflection angle data of the acoustic wave deflection mechanism to obtain angle feedback information;

[0039] Extract the updated data of the listener coordinates, and integrate the angle feedback and coordinate update data into a state observation vector;

[0040] Adopt the deep Q-network algorithm, input the state observation vector, and obtain the predicted value of the target deflection angle;

[0041] If the difference between the target deflection angle and the actual deflection angle exceeds the preset threshold, adjust the parameters of the neural network model.

[0042] In the above technical solution, the system collects the actual deflection angle data of the acoustic wave deflection mechanism and the updated data of the listener coordinates in real time to master the acoustic wave propagation state and the change of the listener's position. Integrate these data into a state observation vector to provide a comprehensive and accurate information basis for subsequent decision-making. This real-time state monitoring and data integration mechanism enables the system to dynamically respond to environmental changes and listener movement, ensuring the precise control of acoustic waves.

[0043] Adopt the deep Q-network algorithm, input the state observation vector, and obtain the predicted value of the target deflection angle. This is an innovative application of reinforcement learning technology in acoustic wave control. The deep Q-network algorithm combines the powerful feature extraction ability of deep learning and the decision-making optimization idea of Q-learning, and can automatically learn the optimal action strategy from the high-dimensional state space. The system predicts the optimal target deflection angle accordingly to guide the speaker to adjust the direction.

[0044] The deep Q-network algorithm learns a large amount of training data to discover the complex relationship and potential law between the state observation vector and the target deflection angle. It can process the high-dimensional complex state space and make accurate decisions quickly in a dynamic environment. This intelligent decision-making ability enables the system to adaptively adjust the acoustic wave deflection angle according to the real-time state information, optimize the acoustic wave propagation effect, and improve the audio transmission efficiency and quality.

[0045] If the difference between the target deflection angle and the actual deflection angle exceeds the preset threshold, adjust the parameters of the neural network model. The system continuously compares the difference between the predicted value and the actual value, and adjusts the model parameters accordingly to gradually reduce the error and improve the prediction accuracy.

[0046] This method can adjust the spatial pointing angle and frequency band gain compensation parameters of the audio drive unit in real time according to the change of the target listener's position. In practical applications, whether the listener moves or the environmental conditions change, the system can maintain the optimal control of the audio output through the dynamic adjustment mechanism to ensure that the listener always enjoys the best audio effect. This real-time response and dynamic optimization ability make this method have significant advantages in complex and changeable practical scenarios and adapt to various usage scenarios and user requirements.

[0047] The whole process constitutes a closed-loop control system, from obtaining the coordinates of the target audience, calculating the adjustment parameters, generating the audio output configuration, to adjusting the parameters, and then continuously monitoring and correcting the system through a feedback mechanism. This closed-loop control method effectively ensures the stability and reliability of the system, timely discovers and corrects deviations or errors, so that the audio output always meets the design requirements and quality standards.

[0048] In some embodiments, based on the model output parameters, the mechanical rotation angle of the speaker rotation mechanism is controlled, and the frequency band gain value of the speaker power amplifier circuit is synchronously adjusted. After that, it further includes:

[0049] Collecting the actual deflection angle feedback data of the sound wave deflection mechanism and the updated listener coordinate information to construct a reinforcement learning state observation vector; using the deep Q-network algorithm to optimize the sound attenuation coefficient and the reflection surface compensation parameters in the sound field propagation model.

[0050] In the above technical solution, the method uses the Deep Q-Network algorithm to optimize the sound attenuation coefficient and the reflection surface compensation parameter in the sound field propagation model, improving the model accuracy. The sound attenuation coefficient and the reflection surface compensation parameter are key factors affecting the sound wave propagation effect. After optimization, the system can more accurately predict the propagation characteristics of sound waves in the actual environment. In different acoustic environments, the sound attenuation and reflection conditions are different. The Deep Q-Network algorithm learns the complex relationship between the state observation vector and the target parameter through a large amount of training data, and automatically adjusts the relevant parameters in the sound field propagation model, enabling the system to adapt to various complex acoustic environments. This optimization mechanism ensures that the system maintains a good audio transmission effect in different scenarios, enhancing the versatility and practicality of the system. The Deep Q-Network algorithm combines the feature extraction of deep learning and the decision-making optimization of Q-learning, can handle high-dimensional complex state spaces, and make accurate decisions quickly. The system adaptively adjusts the sound wave deflection angle accordingly, optimizes the propagation effect, and improves the audio transmission efficiency and quality. If the difference between the target deflection angle and the actual deflection angle exceeds the preset threshold, the parameters of the neural network model are adjusted. The system continuously adjusts the model parameters by comparing the predicted value with the actual value, gradually reducing the error and improving the prediction accuracy. This method can adjust the spatial pointing angle and the frequency band gain compensation parameter of the audio drive unit in real time according to the change of the target listener's position. In practical applications, whether the listener moves or the environmental conditions change, the system can maintain the optimal control of the audio output through the dynamic adjustment mechanism, ensuring that the listener always enjoys the best audio effect. This real-time response and dynamic optimization ability give this method significant advantages in complex and changing actual scenarios, adapting to various usage scenarios and user requirements. The whole process constitutes a closed-loop control system, from obtaining the target listener coordinates, calculating the adjustment parameters, generating the audio output configuration, to adjusting the parameters, and then continuously monitoring and correcting the system through the feedback mechanism. This closed-loop control method effectively ensures the stability and reliability of the system, timely discovers and corrects deviations or errors, and makes the audio output always meet the design requirements and quality standards.

[0051] In some embodiments, the Deep Q-Network algorithm is used to optimize the sound attenuation coefficient and the reflection surface compensation parameter in the sound field propagation model, including:

[0052] Obtain the initial values of the sound attenuation coefficient and the reflection surface compensation parameter in the sound field propagation model;

[0053] Take the sound attenuation coefficient and the reflection surface compensation parameter as state variables and input them into the Deep Q-Network algorithm;

[0054] Calculate the reward value of the current state variable according to the preset reward function;

[0055] If the reward value is lower than the preset threshold, adjust the action amount of the Deep Q-Network;

[0056] Update the sound attenuation coefficient and the reflection surface compensation parameter through the adjusted action amount;

[0057] If the optimization goal is not achieved, repeat the iteration.

[0058] In the above technical solution, during the entire optimization process, the Deep Q-Network algorithm learns the reward values in different states and gradually establishes a policy network that can map states to optimal actions. This learning ability enables the system to automatically adjust the acoustic attenuation coefficient and the reflector compensation parameters under changing environmental conditions to adapt to new acoustic scenarios. This method forms a closed-loop control system. By continuously obtaining state feedback, calculating reward values, adjusting the action amount, and updating parameters, the system can monitor and correct its own performance in real time. This closed-loop control mechanism can not only ensure the stability of the system during long-term operation but also effectively cope with various external interferences and internal parameter drifts, etc., ensuring that the optimization effect of sound wave propagation always meets the expected design requirements.

[0059] According to another aspect of the present invention, there is provided a loudspeaker, which includes: a multi-channel acoustic sensor array for collecting environmental noise data; an infrared positioning module for extracting target listener coordinate information; a first analysis module for constructing a noise distribution model based on the environmental noise and the listener coordinate information; a second analysis module for inputting the target listener coordinates into a pre-constructed sound field propagation model to calculate the spatial pointing angle adjustment parameter and the frequency band gain compensation parameter of the audio driving unit; a neural network module for outputting the rotation angle of the loudspeaker rotation mechanism and the frequency band gain value of the loudspeaker power amplifier circuit according to the noise distribution model and the spatial pointing angle adjustment parameter and the frequency band gain compensation parameter of the audio driving unit; a loudspeaker rotation mechanism for mechanically rotating according to the rotation angle output by the neural network model; and a loudspeaker power amplifier circuit for performing gain adjustment according to the frequency band gain value output by the neural network model.

[0060] In the above technical solution, the advantages of the loudspeaker rely on the above method and will not be elaborated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0062] Figure 1 It is a schematic flowchart of an embodiment of a method for variable sound wave orientation of the present invention;

[0063] Figure 2It is a schematic spatial layout diagram of an embodiment of a method for variable sound wave orientation of the present invention;

[0064] Figure 3 It is a schematic structural diagram of an embodiment of a loudspeaker with variable sound wave orientation of the present invention. Detailed implementation manners

[0065] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be specifically pointed out that the following embodiments are only used to illustrate the present invention, but do not limit the scope of the present invention. Similarly, the following embodiments are only partial embodiments of the present invention rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0066] The present invention provides a method and a loudspeaker with variable sound wave orientation, which can achieve precise noise suppression and sound field directional optimization for complex acoustic environments, significantly improving the audio quality and the listener experience.

[0067] Embodiment 1

[0068] Please refer to Figure 1 , a method for variable sound wave orientation, the method includes:

[0069] S1. Obtain the environmental noise data collected by the multi-channel acoustic sensor array and the target listener coordinate information extracted by the infrared positioning module;

[0070] In this embodiment, take a circular home theater supporting a loudspeaker as an example. Please refer to Figure 2 . In the figure, 1-6 are multiple acoustic sensors forming a multi-channel array, and the loudspeaker rotation mechanism is used to control the rotation angle of the loudspeaker. It is also possible to adopt the method of rotating the horn. For example, a two-way loudspeaker has a tweeter and a mid-bass speaker, and the above effect can also be achieved by changing the horn orientation, which will not be specifically described here. However, it should be noted that the loudspeaker rotation mechanism adopted in this case must have the ability to rotate in the horizontal and vertical directions. This is because the sound wave propagation has three-dimensional spatial characteristics, rather than being limited to a two-dimensional plane. In order to optimize the sound coverage range and effect and make the sound wave evenly spread to every corner of the target area, the above-mentioned loudspeaker rotation mechanism with a two-axis rotation function is adopted.

[0071] The infrared positioning module is relatively small in size and can be built into the speaker. It is easy to match coordinates and can also be composed of multiple units. This is prior art and will not be elaborated here. Those skilled in the art can set it according to actual needs. In addition, the situation of multiple speakers can be set according to actual needs, and the principle of this case still applies. The reason is that in this case, a neural network is used subsequently to optimize the actual result. Therefore, in the case of multiple units, the optimal result can still be obtained by training the neural network to fit.

[0072] S2. Construct a noise distribution model based on environmental noise and listener coordinate information; input the target listener coordinates into the pre-constructed sound field propagation model, and calculate the spatial pointing angle adjustment parameter and frequency band gain compensation parameter of the audio drive unit;

[0073] Exemplarily, a multi-channel acoustic sensor array is used to collect environmental noise data. For example, the noise intensities of 65dB, 70dB, 68dB, 72dB, 69dB, and 70dB are recorded in six directions respectively, and the frequency characteristic range from 100Hz to 10kHz is obtained at the same time. According to the infrared positioning module, the three-dimensional position data of the target listener in space is obtained. For example, the listener coordinates are (x = 2.5m, y = 3.0m, z = 1.2m), and the target listener coordinate information is determined. Integrate the environmental noise data and the target listener coordinate information to generate a relationship diagram of the spatial noise distribution and listener position of the acoustic scene. For example, map the noise intensity and frequency characteristics to a three-dimensional coordinate system to form a noise distribution heat map.

[0074] In this embodiment, constructing a noise distribution model based on environmental noise and listener coordinate information includes:

[0075] S21. Perform a fast Fourier transform on the environmental noise data to obtain the noise spectrum characteristics;

[0076] S22. Extract the energy peaks from the noise spectrum characteristics to determine the main frequency band set and its intensity distribution parameters;

[0077] S23. Perform a spatial mapping of the target listener coordinates and the main frequency band set to establish a noise distribution model.

[0078] In this embodiment, fast Fourier transform (FFT) is performed on the environmental noise data to efficiently convert the time-domain signal into a frequency-domain signal, obtaining the noise spectrum characteristics. The FFT algorithm significantly speeds up the calculation by reducing the computational complexity of the discrete Fourier transform (DFT), enabling this method to quickly extract frequency-domain information from a large amount of noise data, laying a foundation for subsequent noise analysis and processing. By extracting the energy peaks from the noise spectrum characteristics, the frequency components with higher energy can be accurately identified, which usually correspond to the main noise sources. Determining the main frequency band set and its intensity distribution parameters can simplify the complex noise spectrum into the key frequency range and its intensity information, reducing the data processing complexity and focusing on the dominant noise frequency components.

[0079] Perform spatial mapping of the target listener coordinates with the main frequency band set, and combine the noise frequency characteristics and spatial position information to construct a model reflecting the spatial distribution of the noise. This mapping fully considers the noise frequency characteristics and the listener's position, enabling the noise distribution model to more accurately describe the impact on the target listener.

[0080] The noise distribution model constructed in the above steps can accurately depict the distribution of the noise intensity at different positions in a specific spatial environment on the main frequency bands. This model helps the system understand the noise distribution situation, identify the areas and frequencies that have the greatest impact on the listener's auditory experience, providing a reliable basis for sound wave propagation path planning and sound field adjustment, enabling the system to formulate the optimal sound wave orientation strategy according to the actual environment and the listener's position.

[0081] Exemplarily, obtain the environmental noise data, process the noise data using the fast Fourier transform algorithm, convert the time-domain signal into a frequency-domain signal, obtain the noise spectrum characteristics, for example, analyze the energy distribution in the frequency spectrum ranging from 20 Hz to 20 kHz. Extract the energy peaks from the noise spectrum characteristics, and determine the main frequency band set and its intensity distribution parameters through a threshold screening method, for example, identify the frequency components with energy higher than 60 dB as the main frequency bands. Obtain the target listener coordinate information, perform spatial mapping of the target listener coordinates with the main frequency band set, and use a three-dimensional coordinate system to associate the noise frequency characteristics with the listener's position. Combine the noise frequency characteristics and the target listener coordinate information to construct a model reflecting the spatial distribution of the noise, for example, mark the intensity distribution of the main frequency bands at different positions in the model.

[0082] In this embodiment, inputting the target listener coordinates into the sound field propagation model to calculate the spatial pointing angle adjustment parameters and frequency band gain compensation parameters of the audio drive unit includes:

[0083] S24. Obtain spatial position data using the target listener coordinates, and calculate the spatial pointing angle adjustment parameters of the audio drive unit through the sound field propagation model;

[0084] S25. Calculate the frequency band gain compensation parameters according to the adjusted sound wave deflection angle;

[0085] S26. If the frequency band gain compensation parameter meets the preset condition, generate the final audio output configuration;

[0086] S27. Adjust the spatial pointing angle and the frequency band gain compensation parameter of the audio drive unit according to the final audio output configuration.

[0087] Exemplarily, adopt the sound field propagation model, calculate the spatial pointing angle adjustment parameter of the audio drive unit based on the target listener coordinates. For example, determine that the optimal pointing angle of the speaker is 45 degrees by predicting the sound wave propagation path. According to the spatial pointing angle adjustment parameter, calculate the frequency band gain compensation parameter. For example, for a sound wave with a frequency of 1 kHz, calculate its propagation loss and compensate the gain by 3 dB. Set the preset condition to judge whether the frequency band gain compensation parameter meets the requirement. For example, check whether the gain parameter is within the reasonable range of 0 dB to 10 dB. If the frequency band gain compensation parameter meets the preset condition, generate the final audio output configuration. For example, determine that the pointing angle of the speaker is 45 degrees and compensate the gain of the 1 kHz frequency by 3 dB. Adjust the spatial pointing angle and the frequency band gain compensation parameter of the audio drive unit to complete the application of the noise distribution model. For example, adjust the speaker pointing and gain parameters in real time through the control system.

[0088] Exemplarily, the coordinate information of the target listener is obtained, and the spatial position data is extracted, such as extracting the exact position of the listener (x = 2.5 m, y = 3.0 m, z = 1.2 m) through a three-dimensional spatial coordinate system. According to the coordinate information of the target listener, a pre-constructed sound field propagation model is adopted. This model is based on the finite element analysis method and simulates the propagation behavior of sound waves in different media. Through the sound field propagation model, the propagation attenuation characteristics of sound waves in different media are calculated. For example, the sound wave attenuation coefficient in air is 0.1 dB / m, and in the wall it is 0.5 dB / m. Based on the sound wave propagation attenuation characteristics, a mapping relationship between the sound source position and the sound pressure level at the receiving point is established. For example, when the sound source is located at (0 m, 0 m, 1.5 m), the sound pressure level at the receiving point is 75 dB. According to the mapping relationship between the sound source position and the sound pressure level at the receiving point, the spatial pointing angle adjustment parameters of the audio driving unit are calculated. For example, the pointing angle is adjusted to θ = 30°, φ = 45°. The spatial pointing angle adjustment parameters are used to determine the pointing angle of the audio driving unit. For example, the audio driving unit is rotated to the target angle through a stepper motor. According to the sound field propagation model, the frequency band gain compensation parameters at the position of the target listener are calculated. For example, the gain compensation in the 500 Hz frequency band is +3 dB. The frequency band gain compensation parameters are used to adjust the frequency band gain of the audio driving unit. For example, the frequency band gain value is adjusted through a digital signal processor (DSP). Based on the spatial pointing angle adjustment parameters and the frequency band gain compensation parameters, the final control parameters of the audio driving unit are generated. For example, the output control signal is a pointing angle of θ = 30°, φ = 45° (the two angles respectively represent the pointing angles of the speaker in the horizontal and vertical directions), and the frequency band gain at 500 Hz = +3 dB.

[0089] In this embodiment, the method obtains spatial position data through the target listener coordinates, providing an accurate directivity target for the sound field propagation model. This coordinate information can clarify the specific position of the listener in the three-dimensional space, enabling the system to calculate the optimal spatial pointing angle of the audio driving unit accordingly. Compared with traditional fixed pointing or approximate area pointing, this method can perform sound wave orientation more accurately, ensuring that sound waves are directly and efficiently propagated to the target listener position, reducing energy loss and interference to non-target areas.

[0090] Calculating the spatial pointing angle adjustment parameters using the sound field propagation model fully utilizes the advantages of acoustic theory and models. This model is based on the propagation characteristics of sound waves in different media, as well as physical phenomena such as reflection, refraction, and diffraction, and can accurately predict the propagation path and energy distribution of sound waves from the audio driving unit to the target listener. Based on the calculation results of the model, the system can determine in advance how to adjust the pointing angle of the audio driving unit so that sound waves reach the target listener with the optimal path and intensity, improving the audio transmission efficiency and quality.

[0091] When the propagation direction of sound waves changes, the attenuation degrees of sound waves with different frequencies vary. By calculating the frequency band gain compensation parameters according to the deflection angle, the system can perform corresponding gain adjustments for different frequency components to compensate for the energy loss caused by the change in the propagation path. This ensures that the audio signals received by the target listeners have appropriate intensities and are clear in each frequency band, improving the audio fidelity and the listening experience.

[0092] Preset conditions are set to determine whether the frequency band gain compensation parameters meet the requirements. Only when the parameters meet the conditions can the final audio output configuration be generated. This mechanism ensures the system stability and output quality. The preset conditions include parameter range limitations, smoothness requirements, etc., to avoid audio distortion or noise amplification caused by unreasonable gain compensation. The final audio output configuration generated after meeting the conditions comprehensively considers the optimization results of the spatial pointing angle and the frequency band gain compensation, providing precise control instructions for the audio drive unit to ensure its audio output as expected.

[0093] The whole process constitutes a closed-loop control system. From obtaining the coordinates of the target listeners, calculating the adjustment parameters, generating the audio output configuration to adjusting the parameters, and finally continuously monitoring and correcting the system through a feedback mechanism (such as obtaining the ambient noise data and the coordinates of the target listeners again, etc.). This closed-loop control method effectively ensures the system stability and reliability, can timely detect and correct deviations or errors, and makes the audio output always meet the design requirements and quality standards.

[0094] S3. Input the noise distribution model, the spatial pointing angle adjustment parameters and the frequency band gain compensation parameters of the audio drive unit into a pre-trained neural network model, and output the rotation angle of the speaker rotation mechanism and the frequency band gain value of the speaker power amplifier circuit;

[0095] Exemplarily, when collecting environmental noise data, the sound pressure level distribution is obtained through a microphone array, and the time-domain signal is converted into a frequency-domain signal by using the fast Fourier transform to construct a noise distribution model. For example, within the frequency range of 20 Hz to 20 kHz, the noise intensity reaches a peak of 65 dB at 500 Hz. When obtaining the spatial pointing angle adjustment parameters of the audio drive unit, the current spatial pointing angle value is extracted. For example, the horizontal angle is 45 degrees and the vertical angle is 30 degrees. When obtaining the band gain compensation parameters, the gain compensation values for different bands are extracted. For example, the low-band gain compensation is +3 dB, the middle-band is -2 dB, and the high-band is +1 dB. The noise distribution model, the spatial pointing angle adjustment parameters, and the band gain compensation parameters are fused to generate an input feature vector. For example, the noise intensity, the angle value, and the gain value are normalized and combined into [0.65, 0.45, 0.30, 0.03, -0.02, 0.01]. The input feature vector is input into a pre-trained neural network model and processed using the forward propagation algorithm. For example, the output of the hidden layer is calculated through three fully connected layers and the ReLU activation function. Through the output layer of the neural network model, the rotation angle control parameters of the speaker rotation mechanism are obtained. For example, the output value is [0.6, 0.4], indicating that the target horizontal angle is 60 degrees and the vertical angle is 40 degrees. Through the output layer of the neural network model, the different band gain control parameters of the speaker power amplifier circuit are obtained. For example, the output value is [0.05, -0.03, 0.02], indicating that the low-band gain is adjusted to +5 dB, the middle-band is -3 dB, and the high-band is +2 dB. According to the rotation angle control parameters, the target rotation angle of the speaker rotation mechanism is calculated. For example, the output value is converted into the number of pulses of the stepper motor, and the mechanism is driven to rotate to the target position. According to the band gain control parameters, the gain value of the speaker power amplifier circuit is adjusted to match the output requirements of different bands. For example, the gain value is adjusted in real time through a digital signal processor to ensure that the audio output is consistent with the target parameters.

[0096] In this embodiment, the neural network model uses a multi-layer perceptron.

[0097] Exemplarily, a multi-layer perceptron neural network model is used to process the audio output configuration data. The model inputs are audio spectrum features and listener position information, and the output rotation angle control parameter is 45 degrees and the frequency band gain control parameter is low-frequency band gain +3 dB and mid-frequency band gain +1 dB. According to the rotation angle control parameter of 45 degrees, the mechanical rotation angle of the speaker rotation mechanism is controlled, and the stepper motor is driven to rotate at a speed of 5 degrees per second to adjust the spatial directivity of the speaker sound wave radiation so that it faces the listener area. According to the frequency band gain control parameter, the low-frequency band voltage gain of the speaker power amplifier circuit is adjusted to 12 V and the mid-frequency band voltage gain is adjusted to 8 V to optimize the frequency distribution characteristics of the sound wave energy. The actual deflection angle data of the sound wave deflection mechanism is collected, and the current angle is obtained as 43 degrees through an optoelectronic encoder and compared with the target angle of 45 degrees. If the difference between the target deflection angle and the actual deflection angle exceeds the preset threshold of 2 degrees, the weight parameters of the multi-layer perceptron neural network model are adjusted, and the gradient descent algorithm is used for optimization with a learning rate of 0.01. Based on the updated neural network model parameters, the rotation angle control parameter is recalculated as 46 degrees and the frequency band gain control parameter is low-frequency band gain +4 dB and mid-frequency band gain +2 dB. According to the recalculated rotation angle control parameter of 46 degrees, the mechanical rotation angle of the speaker rotation mechanism is controlled again to rotate it to 46 degrees at a speed of 5 degrees per second to further adjust the spatial directivity of the sound wave radiation. According to the recalculated frequency band gain control parameter, the low-frequency band voltage gain of the speaker power amplifier circuit is adjusted again to 13 V and the mid-frequency band voltage gain is adjusted to 9 V to further optimize the frequency distribution characteristics of the sound wave energy.

[0098] In this embodiment, the system uses a basic feedforward neural network, the multi-layer perceptron (MLP), for parameter prediction. The MLP consists of an input layer, one or more hidden layers, and an output layer, and is suitable for classification and regression problems. In this system, the calculation results of the sound field propagation model, the noise distribution model, and the listener coordinate information, etc. are used as the input features of the MLP. These features are non-linearly transformed and combined through multiple layers of neurons in the MLP, and finally the control parameters of the speaker are obtained at the output layer. The MLP has the advantages of simple structure, easy implementation and training, can comprehensively process and learn various input features, and provides effective parameter prediction for the system. Compared with other neural networks, the MLP shows advantages in real-time processing. For example, although the long short-term memory network (LSTM) is suitable for processing sequence data, its complex gating mechanism and cyclic structure result in a slow inference speed; the convolutional neural network (CNN) performs well in processing two-dimensional data such as images or spectrograms, but in the acoustic parameter prediction task, the MLP is more efficient. The MLP has a simple structure, high computational efficiency, and a direct feedforward mechanism, has good real-time processing capabilities, is suitable for acoustic signal processing scenarios that require fast response, can provide the control parameters of the speaker for the system in a timely manner, and ensure the fast and accurate adjustment of the sound wave direction.

[0099] S4. Based on the model output parameters, control the mechanical rotation angle of the speaker rotation mechanism and synchronously adjust the frequency band gain value of the speaker power amplifier circuit.

[0100] In this embodiment, the method integrates the environmental noise data collected by the multi-channel acoustic sensor array and the target listener coordinate information obtained by the infrared positioning module. The multi-channel acoustic sensor array can capture environmental noise from multiple angles and provide spatial noise distribution data; the infrared positioning module accurately obtains the listener's position, providing a basis for adjusting the sound wave orientation. The multi-modal data fusion enables the system to comprehensively perceive the acoustic environment and the listener's position, laying a foundation for optimizing the sound wave orientation.

[0101] Based on the noise distribution model constructed from the above data, it accurately depicts the intensity, distribution of environmental noise, and its relative position to the listener, covering static and dynamic noise sources, providing a detailed "noise map" for sound wave propagation path planning and sound field adjustment, and helping to identify the noise areas that have the greatest impact on the listener's auditory experience.

[0102] Input the listener coordinates into the pre-constructed sound field propagation model, and the adjustment parameters of the audio drive unit can be quickly calculated. This model is based on the propagation characteristics of sound waves in different media and spatial structures, predicting the sound wave propagation path, attenuation, and interaction with obstacles, enabling the system to determine in advance the adjustment of the sound wave emission direction and frequency response, ensuring that the sound wave reaches the listener's position efficiently and accurately.

[0103] Input the noise distribution model and adjustment parameters into the pre-trained neural network model to achieve deep integration and intelligent decision-making of multi-source information. The neural network model learns the complex relationships between data by learning a large amount of training data and outputs the optimal speaker control parameters. This artificial intelligence optimization strategy improves the accuracy and efficiency of sound wave orientation adjustment, adaptively generates personalized sound wave control solutions, and enhances the system's intelligence and user experience.

[0104] The sound field propagation model is constructed based on physical principles and provides relatively accurate parameter calculation results, but the actual acoustic environment is more complex. The neural network model is trained with a large amount of actual data to learn the actual mapping relationship in a complex environment. Using the results of the sound field propagation model as the input of the neural network can optimize parameter prediction on a theoretical basis and improve accuracy. The neural network can simultaneously accept multiple inputs such as the sound field propagation model and the noise distribution model, comprehensively consider the influence of various factors on the speaker control parameters, analyze the acoustic environment comprehensively, output parameter control that meets the actual requirements, and realize the intelligent and automated operation of the system.

[0105] This method can adjust the acoustic wave orientation and frequency band gain in real time according to the changes in environmental noise and the movement of the listener's position. Whether the listener moves or the noise source changes, the system responds quickly, recalculates and optimizes the acoustic wave propagation parameters to ensure the accurate pointing of the acoustic wave and the transmission of high-quality sound effects. By controlling the mechanical rotation angle of the speaker rotation mechanism and the frequency band gain value of the power amplifier circuit, the coordination of mechanical movement and electronic signal processing is achieved. The mechanical rotation adjusts the acoustic wave propagation direction, and the frequency band gain optimizes the acoustic wave energy distribution. Combining to improve the flexibility and accuracy of acoustic wave orientation adjustment, compensating for the acoustic wave distortion and energy loss caused by mechanical movement, further improving the audio output quality and stability. This method can customize the acoustic wave propagation scheme for the listener, meet the personalized audio quality requirements, and improve the satisfaction. In the fields of home theaters, conference systems, public address systems, and virtual reality, etc., this technology has broad application potential and can optimize the sound output and enhance the user experience.

[0106] In this embodiment, S4. Based on the model output parameters, control the mechanical rotation angle of the speaker rotation mechanism, and synchronously adjust the frequency band gain value of the speaker power amplifier circuit. After that, it further includes:

[0107] S5. Collect the actual deflection angle data of the acoustic wave deflection mechanism to obtain angle feedback information;

[0108] S6. Extract the updated data of the listener coordinates, and integrate the angle feedback and the coordinate update data into a state observation vector;

[0109] S7. Adopt the deep Q-network algorithm, input the state observation vector, and obtain the predicted value of the target deflection angle;

[0110] S8. If the difference between the target deflection angle and the actual deflection angle exceeds the preset threshold, then adjust the parameters of the neural network model.

[0111] Exemplarily, collect the actual deflection angle data of the acoustic wave deflection mechanism. For example, obtain the current deflection angle of 45 degrees through an angle sensor, and at the same time obtain the updated data of the listener coordinates through an infrared positioning module, with the coordinates being (2.5 meters, 3.0 meters). Integrate the actual deflection angle data and the listener coordinate data to construct a state observation vector. For example, after normalizing the angle data and the coordinate data, combine them into a vector [0.5, 0.6, 0.45]. Input the state observation vector into the deep Q-network algorithm to predict the target deflection angle. For example, the predicted value output by the neural network model is 50 degrees. Compare the difference between the target deflection angle and the actual deflection angle to determine whether it exceeds the preset threshold. For example, the preset threshold is 5 degrees, and the difference is 5 degrees, which does not exceed the threshold. If the difference exceeds the preset threshold, adjust the parameters of the neural network model. For example, update the network weights through the gradient descent method to optimize the prediction result.

[0112] In this embodiment, the system collects the actual deflection angle data of the acoustic wave deflection mechanism and the updated listener coordinate data in real time to grasp the acoustic wave propagation state and the change of the listener's position. These data are integrated into a state observation vector to provide a comprehensive and accurate information basis for subsequent decision-making. This real-time state monitoring and data integration mechanism enables the system to dynamically respond to environmental changes and listener movement, ensuring precise control of acoustic waves. By using the Deep Q-Network algorithm and inputting the state observation vector, the predicted value of the target deflection angle is obtained, which is an innovative application of reinforcement learning technology in acoustic wave control. The Deep Q-Network algorithm combines the powerful feature extraction ability of deep learning and the decision-making optimization idea of Q-learning, and can automatically learn the optimal action strategy from the high-dimensional state space. Based on this, the system predicts the optimal target deflection angle to guide the speaker to adjust its direction. The Deep Q-Network algorithm learns a large amount of training data to discover the complex relationship and potential law between the state observation vector and the target deflection angle. It can handle high-dimensional complex state spaces and make accurate decisions quickly in a dynamic environment. This intelligent decision-making ability enables the system to adaptively adjust the acoustic wave deflection angle according to real-time state information, optimize the acoustic wave propagation effect, and improve the audio transmission efficiency and quality. If the difference between the target deflection angle and the actual deflection angle exceeds the preset threshold, the parameters of the neural network model are adjusted. By continuously comparing the predicted value with the actual value and adjusting the model parameters accordingly, the system gradually reduces the error and improves the prediction accuracy. This method can adjust the spatial pointing angle and frequency band gain compensation parameters of the audio drive unit in real time according to the change of the target listener's position. In practical applications, whether the listener moves or the environmental conditions change, the system can maintain the optimal control of the audio output through the dynamic adjustment mechanism to ensure that the listener always enjoys the best audio effect. This real-time response and dynamic optimization ability give this method significant advantages in complex and changeable practical scenarios and adapt to various usage scenarios and user requirements. The whole process constitutes a closed-loop control system, from obtaining the target listener coordinates, calculating the adjustment parameters, generating the audio output configuration, to adjusting the parameters, and then continuously monitoring and correcting the system through the feedback mechanism. This closed-loop control method effectively ensures the stability and reliability of the system, discovers and corrects deviations or errors in a timely manner, and makes the audio output always meet the design requirements and quality standards.

[0113] In this embodiment, based on the model output parameters, the mechanical rotation angle of the speaker rotation mechanism is controlled, and the frequency band gain value of the speaker power amplifier circuit is synchronously adjusted. After that, it further includes:

[0114] S9. Collect the actual deflection angle feedback data of the acoustic wave deflection mechanism and the updated listener coordinate information, and construct a reinforcement learning state observation vector; use the Deep Q-Network algorithm to optimize the sound attenuation coefficient and reflection surface compensation parameters in the sound field propagation model.

[0115] In this embodiment, the method uses the Deep Q-Network algorithm to optimize the sound attenuation coefficient and the reflection surface compensation parameter in the sound field propagation model, improving the model accuracy. The sound attenuation coefficient and the reflection surface compensation parameter are the key factors affecting the sound wave propagation effect. After optimization, the system can more accurately predict the propagation characteristics of sound waves in the actual environment. In different acoustic environments, the sound attenuation and reflection conditions are different. The Deep Q-Network algorithm learns the complex relationship between the state observation vector and the target parameters through a large amount of training data, and automatically adjusts the relevant parameters in the sound field propagation model to make the system adapt to various complex acoustic environments. This optimization mechanism ensures that the system maintains a good audio transmission effect in different scenarios, enhancing the versatility and practicality of the system. The Deep Q-Network algorithm combines the feature extraction of deep learning and the decision-making optimization of Q-learning, can handle high-dimensional complex state spaces, and make accurate decisions quickly. The system adaptively adjusts the sound wave deflection angle accordingly, optimizes the propagation effect, and improves the audio transmission efficiency and quality. If the difference between the target deflection angle and the actual deflection angle exceeds the preset threshold, the parameters of the neural network model are adjusted. The system continuously adjusts the model parameters by comparing the predicted value with the actual value, gradually reducing the error and improving the prediction accuracy. This method can adjust the spatial pointing angle and the frequency band gain compensation parameter of the audio drive unit in real time according to the change of the target listener's position. In practical applications, whether the listener moves or the environmental conditions change, the system can maintain the optimal control of the audio output through the dynamic adjustment mechanism, ensuring that the listener always enjoys the best audio effect. This real-time response and dynamic optimization ability make this method have significant advantages in complex and changeable actual scenarios, adapting to various usage scenarios and user requirements. The whole process constitutes a closed-loop control system, from obtaining the target listener coordinates, calculating the adjustment parameters, generating the audio output configuration, to adjusting the parameters, and then continuously monitoring and correcting the system through the feedback mechanism. This closed-loop control method effectively guarantees the stability and reliability of the system, discovers and corrects deviations or errors in time, and makes the audio output always meet the design requirements and quality standards.

[0116] In some embodiments, S9. Optimize the sound attenuation coefficient and the reflection surface compensation parameter in the sound field propagation model by using the Deep Q-Network algorithm, including:

[0117] S91. Obtain the initial values of the sound attenuation coefficient and the reflection surface compensation parameter in the sound field propagation model;

[0118] S92. Input the sound attenuation coefficient and the reflection surface compensation parameter as state variables into the Deep Q-Network algorithm;

[0119] S93. Calculate the reward value of the current state variable according to the preset reward function;

[0120] S94. If the reward value is lower than the preset threshold, adjust the action amount of the Deep Q-Network;

[0121] S95. Update the sound attenuation coefficient and the reflection surface compensation parameter according to the adjusted action amount.

[0122] S96. If the optimization goal is not achieved, repeat the iteration.

[0123] Exemplarily, obtain the initial values of the sound attenuation coefficient and the reflection surface compensation parameter in the sound field propagation model. For example, the sound attenuation coefficient is 0.8 and the reflection surface compensation parameter is 1.2. Input the state observation vector and the initial parameters into the deep Q-network algorithm to optimize the sound attenuation coefficient and the reflection surface compensation parameter. For example, adjust the parameters to 0.85 and 1.15 through the Q-learning algorithm. Update the sound field propagation model according to the optimized sound attenuation coefficient and the reflection surface compensation parameter. For example, recalculate the sound wave propagation path and the attenuation characteristics. Adopt the updated sound field propagation model to recalculate the sound wave propagation path and the attenuation characteristics, and generate new sound wave control parameters. For example, adjust the rotation angle of the speaker to 48 degrees and the frequency band gain to 1.1.

[0124] In this embodiment, during the entire optimization process, the deep Q-network algorithm learns by the reward values in different states, and gradually establishes a policy network that can map the state to the optimal action. This learning ability enables the system to automatically adjust the sound attenuation coefficient and the reflection surface compensation parameter under changing environmental conditions to adapt to new acoustic scenarios. This method forms a closed-loop control system. By continuously obtaining state feedback, calculating the reward value, adjusting the action amount, and updating the parameters, the system can monitor and correct its own performance in real time. This closed-loop control mechanism can not only ensure the stability of the system during long-term operation, but also effectively cope with various external interferences and internal parameter drifts, etc., to ensure that the optimization effect of sound wave propagation always meets the expected design requirements.

[0125] Embodiment Two

[0126] Please refer to Figure 3 , a speaker, the speaker includes: a multi-channel acoustic sensor array for collecting environmental noise data; an infrared positioning module for extracting target listener coordinate information; a first analysis module for constructing a noise distribution model based on the environmental noise and the listener coordinate information; a second analysis module for inputting the target listener coordinates into a pre-constructed sound field propagation model to calculate the spatial pointing angle adjustment parameter and the frequency band gain compensation parameter of the audio drive unit; a neural network module for outputting the rotation angle of the speaker rotation mechanism and the frequency band gain value of the speaker power amplifier circuit according to the noise distribution model and the spatial pointing angle adjustment parameter and the frequency band gain compensation parameter of the audio drive unit; a speaker rotation mechanism for mechanically rotating according to the rotation angle output by the neural network model; a speaker power amplifier circuit for performing gain adjustment according to the frequency band gain value output by the neural network model.

[0127] In the above technical solution, the advantages of the loudspeaker rely on the above method, which will not be elaborated here.

[0128] Furthermore, the loudspeaker further includes a first feedback module, which is used to collect the actual deflection angle data of the acoustic wave deflection mechanism, obtain angle feedback information; extract the listener coordinate update data, and integrate the angle feedback and coordinate update data into a state observation vector; adopt the deep Q-network algorithm, input the state observation vector, and obtain the predicted value of the target deflection angle; if the difference between the target deflection angle and the actual deflection angle exceeds a preset threshold, then adjust the neural network model parameters.

[0129] Furthermore, the loudspeaker further includes a second feedback module, which is used to collect the actual deflection angle feedback data of the acoustic wave deflection mechanism and the updated listener coordinate information, and construct a reinforcement learning state observation vector; adopt the deep Q-network algorithm to optimize the sound attenuation coefficient and reflection surface compensation parameters in the sound field propagation model.

[0130] The above are only some embodiments of the present invention, and thus do not limit the protection scope of the present invention. Any equivalent device or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A method for changing the direction of sound waves, characterized in that: The method comprises, Obtaining environmental noise data collected by a multi-channel acoustic sensor array and target audience coordinate information extracted by an infrared positioning module; A noise distribution model is constructed based on the ambient noise and the listener coordinate information; the target listener coordinates are input into the pre-constructed sound field propagation model to calculate the spatial directivity angle adjustment parameters and frequency band gain compensation parameters of the audio drive unit; The noise distribution model, the spatial directivity angle adjustment parameters of the audio drive unit, and the frequency band gain compensation parameters are input into a pre-trained neural network model to output the rotation angle of the speaker rotation mechanism and the frequency band gain value of the speaker power amplifier circuit; Based on the model output parameters, the mechanical rotation angle of the speaker rotating mechanism is controlled, and the frequency band gain value of the speaker power amplifier circuit is adjusted synchronously.

2. A method for changing the direction of sound waves as claimed in claim 1, characterized in that: A noise distribution model is constructed based on the ambient noise and listener coordinate information, including: Perform fast Fourier transform on environmental noise data to obtain noise spectrum characteristics; Extract energy peaks from noise spectrum characteristics and determine the main frequency band set and its intensity distribution parameters; The target audience coordinates are spatially mapped with the main frequency band set to establish a noise distribution model.

3. A method for changing the direction of sound waves as claimed in claim 1, characterized in that: The step of inputting the target audience coordinates into the sound field propagation model and calculating the spatial directivity angle adjustment parameters and the frequency band gain compensation parameters of the audio driving unit includes: The spatial position data is obtained by using the target audience coordinates, and the spatial directional angle adjustment parameters of the audio driver unit are calculated through the sound field propagation model; Calculate the frequency band gain compensation parameter according to the adjusted sound wave deflection angle; If the frequency band gain compensation parameters meet the preset conditions, the final audio output configuration is generated; Adjust the spatial directivity angle and frequency band gain compensation parameters of the audio driver unit according to the final audio output configuration.

4. A method for changing the direction of sound waves as claimed in claim 1, characterized in that: The neural network model adopts a multi-layer perceptron.

5. A method for changing the direction of sound waves as claimed in claim 1, characterized in that: Based on the model output parameters, the mechanical rotation angle of the speaker rotating mechanism is controlled, and the frequency band gain value of the speaker power amplifier circuit is adjusted synchronously, and then it also includes: Collect actual deflection angle data of the acoustic wave deflection mechanism to obtain angle feedback information; Extract the audience coordinate update data, and integrate the angle feedback and coordinate update data into a state observation vector; Using the deep Q network algorithm, the state observation vector is input to obtain the target deflection angle prediction value; If the difference between the target deflection angle and the actual deflection angle exceeds a preset threshold, the neural network model parameters are adjusted.

6. A method for changing the direction of sound waves as claimed in claim 1, characterized in that: Based on the model output parameters, the mechanical rotation angle of the speaker rotating mechanism is controlled, and the frequency band gain value of the speaker power amplifier circuit is adjusted synchronously, and then it also includes: The actual deflection angle feedback data of the sound wave deflection mechanism and the updated audience coordinate information are collected to construct a reinforcement learning state observation vector. The deep Q-network algorithm is used to optimize the acoustic attenuation coefficient and reflection surface compensation parameters in the sound field propagation model.

7. A method for changing the direction of sound waves as claimed in claim 6, characterized in that: The deep Q network algorithm is used to optimize the acoustic attenuation coefficient and reflection surface compensation parameters in the sound field propagation model, including: Obtaining initial values ​​of acoustic attenuation coefficient and reflection surface compensation parameters in the acoustic field propagation model; The acoustic attenuation coefficient and the reflection surface compensation parameter are input into the deep Q network algorithm as state quantities; According to the preset reward function, calculate the reward value of the current state; If the reward value is lower than the preset threshold, adjust the action amount of the deep Q network; The acoustic attenuation coefficient and the reflection surface compensation parameter are updated through the adjusted action amount; If the optimization goal is not achieved, the iteration is repeated.

8. A speaker, characterized in that: The speaker comprises: A multi-channel acoustic sensor array for collecting ambient noise data; Infrared positioning module, used to extract target audience coordinate information; A first analysis module, for constructing a noise distribution model based on environmental noise and listener coordinate information; The second analysis module is used to input the target audience coordinates into the pre-built sound field propagation model to calculate the spatial directivity angle adjustment parameters and frequency band gain compensation parameters of the audio drive unit; A neural network module, configured to output a rotation angle of a speaker rotating mechanism and a frequency band gain value of a speaker power amplifier circuit according to the noise distribution model and the spatial directivity angle adjustment parameters and frequency band gain compensation parameters of the audio drive unit; a speaker rotation mechanism for mechanically rotating according to a rotation angle output by the neural network model; The loudspeaker power amplifier circuit is used to adjust the gain according to the frequency band gain value output by the neural network model.

Images