Signal processing method and device combining deep learning and acoustic parametric array

Abstract

The invention provides a signal processing method and device combining deep learning and an acoustic parametric array. According to the method, after a to-be-processed audio signal is obtained, a trained preprocessing model is adopted to preprocess the to-be-processed audio signal, and a corresponding ultrasonic signal is output; and acquiring a self-demodulation audio signal after the output ultrasonic signal passes through the configured ultrasonic parametric array loudspeaker, wherein the signal is the self-demodulation audio signal approaching the actual audio signal. According to the method, a signal preprocessing algorithm of the preprocessing model can be obtained through end-to-end back-stepping of the neural network, so that the sound quality problem of the ultrasonic parametric array loudspeaker can be improved.

Description

Technical Field

[0001] The present application relates to the field of acoustic processing technology, and more specifically, to a signal processing method and device combining deep learning and acoustic parametric array. Background Art

[0002] Acoustic parametric array has received extensive attention in the field of audio directional broadcasting technology. Its working principle is based on signal preprocessing and nonlinear acoustics. First, the audio signal is loaded onto the ultrasonic carrier in a specific preprocessing and modulation method to form a composite signal form, and then it is transmitted into the air medium with the help of a transducer. In the propagation process, the nonlinear characteristics of the air medium play a key role. Due to the nonlinear effect of the interaction between air molecules, the original audio signal begins to gradually self-demodulate from the ultrasonic carrier. At the same time, the ultrasonic wave itself, as the carrier of the audio signal, has a high degree of directivity. In the propagation process, the ultrasonic wave can effectively concentrate the energy in a specific direction, so that the self-demodulated audio signal propagates along this predetermined direction, thereby achieving the directional propagation effect of the sound. This feature has irreplaceable application potential in many fields such as audio broadcasting, acoustic imaging, and sound transmission in specific environments. The core theoretical basis of the acoustic parametric array is the classic Berktay formula. This formula describes that the audio signal obtained after the modulation signal with ultrasonic wave as the carrier is self-demodulated by the nonlinear effect of air is proportional to the quadratic derivative of the square of the envelope of the original modulation signal. Therefore, if we want to significantly improve the quality of audio restoration and achieve a near-ideal audio reproduction effect, we must focus on reducing harmonic distortion and optimizing related signal processing algorithms. The difficulty of the preprocessing algorithm lies in the fact that the air self-demodulation process involves nonlinear acoustics, and the entire nonlinear process is affected by various factors such as air temperature and humidity in actual scenes. There are a lot of errors when relying solely on the Berktay formula. Summary of the invention

[0003] The purpose of the embodiments of the present application is to provide a signal processing method and device combining deep learning and acoustic parametric array, so as to solve the above-mentioned problems existing in the prior art. Through the end-to-end reverse preprocessing algorithm of the neural network, the sound quality problem of the ultrasonic parametric array speaker can be improved.

[0004] In a first aspect, a signal processing method combining deep learning and acoustic parameter array is provided, and the method may include: Obtaining the audio signal to be processed; The trained preprocessing model is used to preprocess the audio signal to be processed and output a corresponding ultrasonic signal; the trained preprocessing model is obtained by combining the trained nonlinear processing model and inputting the audio signal input sample into the preprocessing model to be trained; A self-demodulated audio signal is collected after the ultrasonic signal passes through the configured ultrasonic parametric array loudspeaker, wherein the self-demodulated audio signal is an audio signal that is close to a real audio signal in the audio signal to be processed.

[0005] In a possible implementation, the training process of the nonlinear processing model includes: Acquire historical ultrasonic signals in different frequency bands; Placing a microphone at a preset distance value in the output direction of the ultrasonic parametric array speaker, and obtaining a self-demodulated microphone signal received by the microphone after the historical ultrasonic signal passes through the ultrasonic parametric array speaker; The historical ultrasonic signal is used as an input sample, and the microphone signal is used as a sample label to train the neural network model to obtain a trained nonlinear processing model.

[0006] In a possible implementation, obtaining a self-demodulated microphone signal received by the microphone includes: Collect the initial signal after self-demodulation received by the microphone; Using a minimum value statistical method, the initial signal after self-demodulation is processed to obtain a noise power spectrum in the current environment; The self-demodulated initial signal is processed by using a preset spectral subtraction method to obtain a self-demodulated microphone signal.

[0007] In one possible implementation, the method further includes: After the nonlinear processing model is trained, the neural network model to be trained is trained to obtain a trained preprocessing model, wherein the training process of the preprocessing model includes: The acquired historical audio signal is used as a sample label, and the signal obtained by mixing the historical audio signal with different noise signals is used as different audio signal input samples; For any audio signal input sample, obtaining a corresponding ultrasonic signal sample output after the audio signal input sample is input into the preprocessing model to be trained; Collecting a self-demodulated audio signal sample after the ultrasonic signal sample passes through the configured ultrasonic parametric array speaker; Using a trained nonlinear processing model to process the ultrasonic signal sample, and output an audio signal output sample that fits and approximates the historical audio signal; A preset loss function is used to process the historical audio signal and the corresponding audio signal output samples of each iteration to obtain a loss value. If the loss value does not meet the iteration termination condition, the model parameters are optimized and iterative training is continued until the iteration termination condition is met to obtain a trained preprocessing model.

[0008] In a possible implementation, before training the neural network model to be trained to obtain a trained preprocessing model, the method further includes: Set the hyperparameters, learning rate and optimization function for model training.

[0009] In one possible implementation, the loss function is a mean square error loss function.

[0010] In a second aspect, a signal processing device combining deep learning and acoustic parametric array is provided, and the device may include: An acquisition unit, used for acquiring an audio signal to be processed; A processing unit, used to pre-process the audio signal to be processed by using a trained pre-processing model, and output a corresponding ultrasonic signal; the trained pre-processing model is obtained by combining a trained nonlinear processing model and inputting an audio signal input sample into the pre-processing model to be trained; The collecting unit is used to collect the self-demodulated audio signal after the ultrasonic signal passes through the configured ultrasonic parametric array loudspeaker, and the self-demodulated audio signal is an audio signal that is close to the real audio signal in the audio signal to be processed.

[0011] In a third aspect, an electronic device is provided, the electronic device comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; The processor is used to implement any method step described in the first aspect when executing the program stored in the memory.

[0012] In a fourth aspect, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, any method step described in the first aspect is implemented.

[0013] The signal processing method combining deep learning and acoustic parametric array provided in the embodiment of the present application uses a trained preprocessing model to preprocess the audio signal to be processed after obtaining the audio signal to be processed, and outputs the corresponding ultrasonic signal; the trained preprocessing model is obtained by combining the trained nonlinear processing model, and inputting the audio signal input sample into the preprocessing model to be trained; the self-demodulated audio signal after the ultrasonic signal passes through the configured ultrasonic parametric array speaker is collected, and the self-demodulated audio signal is an audio signal that approximates the actual audio signal in the audio signal to be processed after self-demodulation. This method can reversely derive the signal preprocessing algorithm of the preprocessing model through an end-to-end neural network, thereby improving the sound quality problem of the ultrasonic parametric array speaker. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1 A flowchart of a signal processing method combining deep learning and acoustic parameter array provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a signal processing device combining deep learning and acoustic parameter array provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. Unless otherwise defined, the technical terms or scientific terms used in this application should be the common meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in this application do not indicate any order, quantity or importance, but are only used to distinguish different components. "Including" or "including" and other similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected", "coupled" or "connected" and other similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0017] For ease of understanding, the terms involved in the embodiments of the present application are explained below: The input signal of the Minimum Statistics (MS) method is a signal containing noise. The method aims to estimate the power spectral density (PSD) of noise from a mixed signal containing speech and background noise. Specifically, the Minimum Statistics method does not require pure noise samples, but directly processes the mixed signal containing speech and noise.

[0018] The characteristics of the input signal include: Input signal It is a voice signal and background noise The superposition of: In practical applications, background noise is usually non-stationary, that is, its statistical characteristics change over time. Speech signals are intermittent, that is, there is speech in some time periods and only noise in other time periods.

[0019] The basic steps of the minimum statistical method are: 1. Short-time Fourier transform (STFT) First, the input signal is divided into overlapping short-time frames (usually using a Hamming window or a Hanning window), and each frame is subjected to a fast Fourier transform (FFT) to obtain a short-time spectrum. The spectrum of each frame can be expressed as ,in represents the time frame, Indicates frequency.

[0020] 2. Calculate the power spectrum Take the modulus square of the spectrum of each frame to get the power spectrum:

[0021] This step converts the complex spectrum into a real power spectrum for easy subsequent processing.

[0022] 3. Minimum value estimation within the sliding window To estimate the noise power spectrum, the minimum statistical method searches for the minimum value of each frequency point in a sliding time window. Assume that the length of the sliding window is , then for each frequency point , in the time window Find the minimum power spectrum value within:

[0023] This minimum value is used as an estimate of the noise power spectrum at that frequency point.

[0024] 4. Smoothing Since directly using the minimum value may cause the estimated result to be too sensitive and easily affected by transient signals, the estimated noise power spectrum is usually smoothed. Commonly used smoothing methods include exponential weighted average (EWA):

[0025] in, is the smoothing coefficient, is the noise power spectrum estimate at the previous moment. By adjusting A balance can be achieved between response speed and stability.

[0026] 5. Noise Power Spectrum Estimation After smoothing, we get This is the estimated value of the noise power spectrum.

[0027] The preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. In addition, the embodiments and features in the embodiments of the present application may be combined with each other if there is no conflict.

[0028] Traditional signal preprocessing methods are all theoretically derived based on certain assumptions, and in fact have large errors. For this reason, this application provides a signal processing method that combines deep learning and acoustic parameter arrays. In this method, the air self-demodulation process is first modeled through a neural network. The modeling environment is located in an anechoic chamber at a constant temperature. The specific process is: collect the ultrasonic transmission signal and the microphone receiving signal at a position of 1m, and then preprocess the signal to form an input-output pair. Then build a neural network to fit the nonlinear relationship between input and output. After the training is completed, save the model and complete the nonlinear self-demodulation process. After the self-demodulation process is completed, the pre-processing module can be built. The pre-processing module is combined with the self-demodulation process, and the weight of the self-demodulation modeling part is fixed to obtain the pre-processing algorithm.

[0029] Compared with the traditional approximate solution, the method of the present application can improve the sound quality problem of the ultrasonic parametric array speaker in a specific environment through an end-to-end reverse preprocessing algorithm of a neural network.

[0030] Figure 1 A flowchart of a signal processing method combining deep learning and acoustic parameter array provided in an embodiment of the present application. Figure 1 As shown, the method may include: Step S110: Acquire the audio signal to be processed.

[0031] The audio signal to be processed may be collected using a fixed format and sampling rate. For example, the audio signal to be processed may be a 24-bit audio signal collected using a 48k sampling rate.

[0032] Step S120: Use the trained preprocessing model to preprocess the audio signal to be processed and output a corresponding ultrasonic signal.

[0033] The trained preprocessing model is obtained by combining the trained nonlinear processing model and inputting the audio signal input sample into the preprocessing model to be trained.

[0034] Step S130: collecting the self-demodulated audio signal after the output ultrasonic signal passes through the configured ultrasonic parametric array speaker.

[0035] The self-demodulated audio signal is a signal that is close to the actual audio signal in the audio signal to be processed after self-demodulation.

[0036] In some embodiments, before executing the above step S110, the configuration of the ultrasonic parametric array speaker, the training of the preprocessing model and the nonlinear processing model are performed in the following order: first, the configuration of the ultrasonic parametric array speaker is performed, then the nonlinear processing model is trained, and finally the preprocessing model is trained, that is, the preprocessing algorithm of the preprocessing model is inverted from the neural network end of the nonlinear processing model to the neural network end of the preprocessing model. Specifically: (1) The configuration process of the ultrasonic parametric array loudspeaker is mainly the beamforming loudspeaker module of the array element. This loudspeaker module is a relatively mature module. Its purpose is to enable the preprocessing model to be able to backpropagate.

[0037] (2) The training process of the nonlinear processing model may include: Historical ultrasonic signals of different frequency bands are acquired; wherein ultrasonic signals of different frequency bands can be randomly prepared.

[0038] A microphone is placed at a preset distance value in the output direction of the ultrasonic parametric array speaker, and after the historical ultrasonic signal passes through the ultrasonic parametric array speaker, a self-demodulated microphone signal received by the microphone is obtained; The collected historical ultrasonic signals are used as input samples, and the self-demodulated microphone signals are used as sample labels to train the neural network model to be trained, thereby obtaining a trained nonlinear processing model.

[0039] The neural network model to be trained can be any network model such as a multi-layer perceptron, a convolutional neural network, a recurrent neural network, or a combination of multiple network models. The number of network layers and the number of neuron nodes of the neural network model can be flexibly adjusted according to the signal characteristics.

[0040] It should be noted that the training of the neural network model to be trained can be iterative training, and the input signal and output signal of the iteratively trained model can be processed by the configured loss function to update the model parameters until the model converges. The specific parameter updating process is not described in detail in this application.

[0041] In some embodiments, the step of obtaining a self-demodulated microphone signal received by the microphone may include: preprocessing the collected initial microphone signal to remove environmental noise, specifically: collecting the self-demodulated initial signal received by the microphone; using the minimum statistical method to process the self-demodulated initial signal to obtain the noise power spectrum in the current environment; then, using a preset spectral subtraction method to process the self-demodulated initial signal to obtain a self-demodulated microphone signal.

[0042] (3) The training process of the preprocessing model may include: The acquired historical audio signal is used as a sample label, and the signal obtained by mixing the historical audio signal with different noise signals is used as a different audio signal input sample; wherein the sampling rates of the historical audio signal and the mixed signal range from 4k to 48k.

[0043] For any audio signal input sample, obtain the corresponding ultrasonic signal sample output after the audio signal input sample is input into the preprocessing model to be trained; wherein the neural network model to be trained can be any network model such as multi-layer perceptron, convolutional neural network, recurrent neural network, or a combination of multiple network models. The number of network layers and the number of neuron nodes of the neural network model can be flexibly adjusted according to the signal characteristics.

[0044] Collect the self-demodulated audio signal samples after the output ultrasonic signal samples pass through the configured ultrasonic parametric array loudspeaker; The trained nonlinear processing model is used to process the ultrasonic signal samples, and the audio signal output samples that are close to the historical audio signal are output; The preset loss function is used to process the audio signal input samples and the corresponding audio signal output samples of each iteration to obtain the loss value. If the loss value does not meet the iteration termination condition, the model parameters are optimized and the iterative training is continued until the iteration termination condition is met to obtain the trained preprocessing model. In other words, during the training of the preprocessing model, only the model parameters in the preprocessing model are optimized and updated, and the parameters of the configured ultrasonic parametric array speaker and the trained nonlinear processing model remain unchanged.

[0045] In some embodiments, the loss function involved in the present application may be a mean square error loss function.

[0046] In some embodiments, before training the neural network model to be trained, hyperparameters, learning rate, and optimization function for model training may be set.

[0047] In some embodiments, if the number of the above-mentioned input samples obtained is insufficient, it will affect the accuracy of the model during training. Therefore, the obtained input samples can be first classified according to the neighbor information to obtain input sample clusters with different neighbor information; two input samples are selected from any input sample cluster, and linear interpolation is performed between the two input samples to generate a new input sample.

[0048] This application uses a signal preprocessing algorithm that can be inferred from the end-to-end neural network to derive the preprocessing model. In order to improve the sound quality of the ultrasonic parametric array, the self-demodulation process of the ultrasonic signal in the air is accurately modeled through a data-driven neural network. After the modeling is completed, the entire system is formed into a guideable system, so that the design of the signal preprocessing algorithm can be completed end-to-end, and finally the signal preprocessing algorithm is implemented in the actual parameter array to improve the sound quality.

[0049] Corresponding to the above method, the embodiment of the present application also provides a signal processing device combining deep learning and acoustic parameter array, such as Figure 2 As shown, the device comprises: An acquisition unit 210, configured to acquire an audio signal to be processed; The processing unit 220 is used to pre-process the audio signal to be processed by using a trained pre-processing model and output a corresponding ultrasonic signal; the trained pre-processing model is obtained by combining a trained nonlinear processing model and inputting an audio signal input sample into the pre-processing model to be trained; The acquisition unit 230 is used to acquire the self-demodulated audio signal after the ultrasonic signal passes through the configured ultrasonic parametric array speaker; the self-demodulated audio signal is a signal that is close to the actual audio signal in the audio signal to be processed after self-demodulation.

[0050] The functions of each functional unit of the signal processing device combining deep learning and acoustic parameter array provided in the above-mentioned embodiments of the present application can be realized through the above-mentioned method steps. Therefore, the specific working process and beneficial effects of each unit in the signal processing device combining deep learning and acoustic parameter array provided in the embodiments of the present application will not be repeated here.

[0051] The present application also provides an electronic device, such as Figure 3 As shown, it includes a processor 310 , a communication interface 320 , a memory 330 and a communication bus 340 , wherein the processor 310 , the communication interface 320 , and the memory 330 communicate with each other via the communication bus 340 .

[0052] Memory 330, for storing computer programs; The processor 310 is used to execute the program stored in the memory 330 to implement the following steps: Obtaining the audio signal to be processed; The trained preprocessing model is used to preprocess the audio signal to be processed and output a corresponding ultrasonic signal; the trained preprocessing model is obtained by combining the trained nonlinear processing model and inputting the audio signal input sample into the preprocessing model to be trained; A self-demodulated audio signal is collected after the ultrasonic signal passes through the configured ultrasonic parametric array loudspeaker, wherein the self-demodulated audio signal is a signal that approximates the actual audio signal in the audio signal to be processed after self-demodulation.

[0053] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0054] The communication interface is used for communication between the above electronic device and other devices.

[0055] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0056] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can 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 gate or transistor logic devices, discrete hardware components.

[0057] The implementation methods and beneficial effects of the components of the electronic device in the above embodiments to solve the problems can be seen in Figure 1 The various steps in the illustrated embodiment are implemented, therefore, the specific working process and beneficial effects of the electronic device provided by the embodiment of the present application are not repeated here.

[0058] In another embodiment provided in the present application, a computer-readable storage medium is also provided, which stores instructions. When the computer-readable storage medium is run on a computer, the computer executes the signal processing method combining deep learning and acoustic parametric array described in any of the above embodiments.

[0059] In another embodiment provided in the present application, a computer program product comprising instructions is also provided. When the computer is run on a computer, the computer executes the signal processing method combining deep learning and acoustic parametric array as described in any of the above embodiments.

[0060] Those skilled in the art will appreciate that the embodiments in the present application can be provided as methods, systems, or computer program products. Therefore, the embodiments in the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments in the present application can be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0061] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0062] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0063] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0064] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present application.

[0065] Obviously, those skilled in the art can make various changes and modifications to the embodiments in the present application without departing from the spirit and scope of the embodiments in the present application. Thus, if these modifications and variations of the embodiments in the present application fall within the scope of the claims and their equivalents in the embodiments of the present application, the embodiments of the present application are also intended to include these modifications and variations.

Claims

1. A signal processing method combining deep learning and acoustic parameter array, characterized in that: The method comprises: Obtaining the audio signal to be processed; The trained preprocessing model is used to preprocess the audio signal to be processed and output a corresponding ultrasonic signal; the trained preprocessing model is obtained by combining the trained nonlinear processing model and inputting the audio signal input sample into the preprocessing model to be trained; A self-demodulated audio signal is collected after the ultrasonic signal passes through the configured ultrasonic parametric array loudspeaker, wherein the self-demodulated audio signal is a signal that approximates the actual audio signal in the audio signal to be processed after self-demodulation.

2. The method according to claim 1, characterized in that The training process of the nonlinear processing model includes: Acquire historical ultrasonic signals in different frequency bands; Placing a microphone at a preset distance value in the output direction of the ultrasonic parametric array speaker, and obtaining a self-demodulated microphone signal received by the microphone after the historical ultrasonic signal passes through the ultrasonic parametric array speaker; The historical ultrasonic signal is used as an input sample, and the microphone signal is used as a sample label to train the neural network model to obtain a trained nonlinear processing model.

3. The method according to claim 2, characterized in that Acquiring a self-demodulated microphone signal received by the microphone, comprising: Collect the initial signal after self-demodulation received by the microphone; Using a minimum value statistical method, the initial signal after self-demodulation is processed to obtain a noise power spectrum in the current environment; The self-demodulated initial signal is processed by using a preset spectral subtraction method to obtain a self-demodulated microphone signal.

4. The method according to claim 2, characterized in that The method further comprises: After the nonlinear processing model is trained, the neural network model to be trained is trained to obtain a trained preprocessing model, wherein the training process of the preprocessing model includes: The acquired historical audio signal is used as a sample label, and the signal obtained by mixing the historical audio signal with different noise signals is used as different audio signal input samples; For any audio signal input sample, obtaining a corresponding ultrasonic signal sample output after the audio signal input sample is input into the preprocessing model to be trained; Collecting a self-demodulated audio signal sample after the ultrasonic signal sample passes through the configured ultrasonic parametric array speaker; Using a trained nonlinear processing model to process the ultrasonic signal sample, and output an audio signal output sample that fits and approximates the historical audio signal; A preset loss function is used to process the historical audio signal and the corresponding audio signal output samples of each iteration to obtain a loss value. If the loss value does not meet the iteration termination condition, the model parameters are optimized and iterative training is continued until the iteration termination condition is met to obtain a trained preprocessing model.

5. The method according to claim 4, characterized in that Before training the neural network model to be trained to obtain a trained preprocessing model, the method further includes: Set the hyperparameters, learning rate and optimization function for model training.

6. The method according to claim 4, characterized in that The loss function is a mean square error loss function.

7. A signal processing device combining deep learning and acoustic parameter array, characterized in that: The device comprises: An acquisition unit, used for acquiring an audio signal to be processed; A processing unit, used to pre-process the audio signal to be processed by using a trained pre-processing model, and output a corresponding ultrasonic signal; the trained pre-processing model is obtained by combining a trained nonlinear processing model and inputting an audio signal input sample into the pre-processing model to be trained; The acquisition unit is used to acquire the self-demodulated audio signal after the ultrasonic signal passes through the configured ultrasonic parametric array speaker, and the self-demodulated audio signal is a signal that is close to the actual audio signal in the audio signal to be processed after self-demodulation.

8. An electronic device, characterized in that: The electronic device comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; A processor, used to implement any method described in claims 1-6 when executing a program stored in a memory.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images