Audio output control method and device, equipment and storage medium

Abstract

The invention relates to the technical field of audio processing, and discloses an audio output control method, device and equipment and a storage medium, and the method comprises the steps: detecting an input voltage signal and collecting an audio input signal through an MCU control unit, obtaining a voltage parameter matrix and an audio parameter matrix, and combining the voltage parameter matrix and the audio parameter matrix to form a mixed feature matrix; inputting the mixed feature matrix into a deep neural network for feature extraction to obtain an audio feature space; constructing a cross attention model, calculating an association weight value among the audio parameters, and generating an audio curve switching control sequence; switching a control sequence based on an audio curve, dynamically adjusting the voltage of the power amplifier through a feedback controller, and collecting working state data of the power amplifier; and performing nonlinear distortion compensation and noise suppression processing on the working state data of the power amplifier to generate an audio output quality parameter, thereby effectively solving the problem of unstable audio output under different power supply conditions, and realizing intelligentization and precision of audio output control.

Description

Technical Field

[0001] The present invention relates to the field of audio processing technology, and in particular to an audio output control method, device, equipment and storage medium. Background Art

[0002] Traditional audio output control methods usually use fixed audio processing parameters and cannot be dynamically adjusted according to different power supply conditions and usage scenarios, resulting in unstable audio output quality. Especially when using fast charging technology, rapid changes in voltage will have a significant impact on audio output quality.

[0003] The audio processing solutions currently on the market generally have problems such as insufficient sound quality control accuracy and weak anti-interference ability. Especially in complex usage environments, due to the lack of an effective voltage adaptation mechanism and audio feature extraction method, audio output is prone to distortion, noise interference and other problems, which seriously affects the user experience. Although there are some audio optimization solutions in the prior art, most of them are designed for a single scenario or fixed conditions, and lack comprehensive consideration of multi-dimensional factors such as voltage changes and device status. At the same time, these solutions have limited capabilities in handling nonlinear distortion and noise suppression, and cannot meet the needs of high-quality audio output. Summary of the invention

[0004] The present invention provides an audio output control method, device, equipment and storage medium, which effectively solves the problem of unstable audio output under different power supply conditions and realizes intelligent and precise audio output control.

[0005] In a first aspect, the present invention provides an audio output control method, the audio output control method comprising:

[0006] The MCU control unit detects the input voltage signal and collects the audio input signal to obtain a voltage parameter matrix and an audio parameter matrix, and combines the voltage parameter matrix with the audio parameter matrix to form a mixed feature matrix;

[0007] Inputting the mixed feature matrix into a deep neural network for feature extraction, performing time-frequency domain analysis on the audio signal, and obtaining an audio feature space;

[0008] Constructing a cross attention model according to the audio feature space, calculating the association weight values ​​between audio parameters, and generating an audio curve switching control sequence;

[0009] Based on the audio curve switching control sequence, dynamically adjusting the power amplifier voltage through a feedback controller, and collecting the power amplifier working status data;

[0010] Nonlinear distortion compensation and noise suppression processing are performed on the power amplifier working state data to generate audio output quality parameters, wherein the audio output quality parameters include signal-to-noise ratio parameters, harmonic distortion parameters, and frequency response flatness parameters.

[0011] In a second aspect, the present invention provides an audio output control device, the audio output control device comprising:

[0012] An acquisition module is used to detect an input voltage signal and acquire an audio input signal through an MCU control unit to obtain a voltage parameter matrix and an audio parameter matrix, and combine the voltage parameter matrix with the audio parameter matrix to form a mixed feature matrix;

[0013] A feature extraction module is used to input the mixed feature matrix into a deep neural network for feature extraction, perform time-frequency domain analysis on the audio signal, and obtain an audio feature space;

[0014] A calculation module, used to construct a cross attention model according to the audio feature space, calculate the association weight values ​​between audio parameters, and generate an audio curve switching control sequence;

[0015] A dynamic adjustment module, used to switch the control sequence based on the audio curve, dynamically adjust the power amplifier voltage through a feedback controller, and collect the power amplifier working status data;

[0016] The generating module is used to perform nonlinear distortion compensation and noise suppression processing on the power amplifier working state data to generate audio output quality parameters, wherein the audio output quality parameters include signal-to-noise ratio parameters, harmonic distortion parameters and frequency response flatness parameters.

[0017] The third aspect of the present invention provides an audio output control device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the audio output control device executes the above-mentioned audio output control method.

[0018] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned audio output control method.

[0019] In the technical solution provided by the present invention, the coordinated detection of input voltage and audio signal is realized through the MCU control unit, and feature extraction and analysis are performed in combination with a deep neural network, which significantly improves the accuracy and adaptability of audio output control. The cross-attention model is used to dynamically calculate the weights of audio parameters, realize the intelligent switching of audio curves, and effectively solve the problem of unstable audio output under different power supply conditions. Through the dynamic adjustment and feedback control mechanism of the power amplifier voltage, combined with nonlinear distortion compensation and noise suppression processing, the signal-to-noise ratio and frequency response flatness of the audio output are greatly improved. The deep learning algorithm is innovatively applied to audio feature extraction and parameter optimization, which realizes the intelligence and precision of audio output control and enables the system to have stronger environmental adaptability. The modular design concept is adopted, and the interfaces between the functional units are standardized, which improves the scalability and maintainability of the system and facilitates subsequent functional upgrades and optimization. Through multi-level signal processing and optimization strategies, the audio output quality is comprehensively improved, meeting the high-quality audio output requirements in different application scenarios.

[0020] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of an embodiment of an audio output control method in an embodiment of the present invention;

[0023] Figure 2 A schematic diagram of an embodiment of an audio output control device in an embodiment of the present invention;

[0024] Figure 3 FIG. 4 is a schematic diagram of an embodiment of an audio output control device in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0027] To facilitate understanding of this embodiment, firstly, an audio output control method disclosed in an embodiment of the present invention is described in detail. Figure 1 As shown, the method comprises the following steps:

[0028] 101. Detect the input voltage signal and collect the audio input signal through the MCU control unit to obtain a voltage parameter matrix and an audio parameter matrix, and combine the voltage parameter matrix and the audio parameter matrix to form a mixed feature matrix;

[0029] It is understandable that the execution subject of the present invention may be an audio output control device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0030] Specifically, the MCU control unit identifies the fast charging protocol of the input voltage signal to obtain the protocol type identifier, which includes USB PD3.0, PD2.0 and BC1.2 boost fast charging protocol identifiers. The identification of the fast charging protocol is achieved through the preliminary analysis of the voltage signal. According to the different characteristics of the protocol type, the MCU performs pattern matching and handshake signal analysis on the input signal to determine the protocol standard followed by the current voltage signal to ensure that the system can adapt to different fast charging protocols. After completing the protocol identification, the input voltage signal is sampled and detected according to the protocol type identifier to generate a voltage sampling sequence. The voltage sampling sequence is implemented by a high-precision ADC module, and its sampling frequency and resolution are dynamically adjusted according to the protocol requirements to capture the fluctuation characteristics and dynamic changes of the voltage signal. The voltage sampling sequence is passed to the feature extraction module, in which the characteristic parameters of the voltage are extracted by signal processing methods such as Fourier transform, including the average value, peak value, frequency component and other information of the voltage. These features are integrated to form a voltage parameter matrix. At the same time, multi-channel sampling is performed on the audio input signal. Through multi-channel sampling technology, the time domain data of the audio signal is separated into sampling data of different channels. The data of each channel contains detailed information of the audio signal on its own sampling path. The sampled audio data is input into the audio feature extraction module. The module processes the data through short-time Fourier transform and other frequency domain analysis methods to extract the frequency response, phase characteristics and dynamic range data of the audio. After integration, these data form an audio parameter matrix, in which the frequency response data describes the amplitude change of the audio signal at different frequencies and reflects the overall spectrum characteristics of the audio signal; the phase characteristic data records the phase distribution information of the audio signal, which helps to restore the time domain morphology of the audio; the dynamic range data reflects the minimum and maximum amplitude change range of the signal and evaluates the performance of the signal at different intensities. The voltage parameter matrix is ​​standardized to obtain a normalized voltage matrix, and the audio parameter matrix is ​​standardized to obtain a normalized audio matrix. The linear transformation method is used to map all parameter values ​​to a unified numerical range (usually 0 to 1). The normalized voltage matrix and the normalized audio matrix are combined into a hybrid feature matrix.

[0031] 102. Input the mixed feature matrix into a deep neural network for feature extraction, perform time-frequency domain analysis on the audio signal, and obtain an audio feature space;

[0032] Specifically, the normalized voltage matrix in the mixed feature matrix is ​​input into the feature extraction module of the deep neural network, which consists of three convolutional layers. Each convolutional layer uses the ReLU activation function to enhance the nonlinear mapping ability. At the same time, a batch normalization layer is introduced after each convolutional layer to accelerate convergence and prevent the gradient from disappearing. After multi-level processing of the voltage matrix through this structure, the voltage deep feature vector describing the characteristics of the voltage signal is extracted. At the same time, the normalized audio matrix is ​​input into the spectrum analysis module in the deep neural network, which consists of five layers of convolutional neural networks, each of which is followed by a batch normalization layer and a LeakyReLU activation function. LeakyReLU has better performance in processing the nonlinear relationship of audio signals. It can retain a certain gradient in the negative value range, thereby effectively alleviating the "dead neuron" problem of the neural network. During the convolution process, the audio signal is parsed layer by layer, and its frequency distribution, timing characteristics, etc. are gradually extracted, and finally an audio deep feature vector containing rich spectrum information is obtained. The voltage deep feature vector and the audio deep feature vector are fused. The different dimensions of the two features are combined through feature concatenation or attention mechanism to generate a fused feature vector. The fused feature vector represents the joint characteristics of the voltage signal and the audio signal, and captures the potential complex relationship between the two. The fused feature vector is input into the four-layer fully connected neural network in the deep neural network. The feature mapping capability of the fully connected layer is used to project the fused features from the low-dimensional space to the high-dimensional space to generate a high-dimensional feature vector. The high-dimensional feature vector is subjected to deep residual calculation. The deep residual network effectively alleviates the gradient vanishing problem in the deep network by introducing jump connections, while retaining the correlation between the low-level information and the high-level information of the features, and obtains the time-frequency feature matrix. The time-series feature matrix is ​​extracted through time series analysis to capture the dynamic characteristics of the signal changing over time. The time-series feature extraction uses recurrent neural network (RNN) or long short-term memory network (LSTM) to ensure the model's ability to handle time correlation. The output of the time-series feature extraction is a dynamic feature vector, which describes the performance of the signal in a dynamic scene. The dynamic feature vector is input into the decoder network for feature reconstruction. The decoder network contains four deconvolution layers and two fully connected layers. The deconvolution layers gradually restore the original spatial information of the features, while the fully connected layers adjust the distribution of the output data. Through the processing of the decoder network, the audio feature space is finally generated. The audio feature space describes the frequency response characteristics, phase characteristics, and harmonic characteristics of the audio signal. The frequency response characteristics record the amplitude performance of the signal at different frequencies and reflect the overall spectrum distribution of the audio signal; the phase characteristics describe the phase relationship between the signal at different frequency components; and the harmonic characteristics reflect the harmonic components of the audio signal.

[0033] 103. Construct a cross attention model according to the audio feature space, calculate the correlation weight values ​​between the audio parameters, and generate an audio curve switching control sequence;

[0034] Specifically, feature projection is performed on the frequency response feature data in the audio feature space, and the high-dimensional frequency response features are mapped to low-dimensional frequency feature vectors through linear transformation, thereby retaining key information while reducing data redundancy. At the same time, feature projection is performed on the phase feature data and harmonic feature data in the audio feature space, and their main components are extracted and integrated into audio representation vectors. The frequency feature vector is used as the query vector, and the audio representation vector is used as the key value vector. The similarity between the query vector and the key value vector is calculated through the dot product operation in the cross attention model to generate a similarity matrix. The result of the dot product operation reflects the correlation strength between the frequency feature and the phase and harmonic features. The similarity matrix is ​​normalized, and the softmax function is used to map the values ​​in the matrix to the range of 0 to 1 to obtain a normalized similarity matrix. The normalized similarity matrix and the audio representation vector are weighted and summed to generate an attention feature vector. According to the attention feature vector, an audio feature association graph is constructed, and the nodes of the association graph represent different audio parameters, while the weights of the edges represent the correlation strength between the parameters. By extracting features from the association graph using a graph convolutional network, the model can extract high-level global information from the local connection structure, generate an audio parameter association matrix, and describe the complex relationships between audio parameters in different dimensions. The audio parameter association matrix is ​​clustered, and similar audio parameters are divided into different categories using a clustering algorithm to generate an audio curve category matrix. The audio curve category matrix assigns a representative feature to each category of parameters, which corresponds to different optimization directions in the audio system. Based on the audio curve category matrix, the curve templates in the preset audio curve library are matched, including gain curves, frequency curves, and phase curves. The matching process measures the similarity between the category matrix and the templates in the curve library to select the candidate curve sequence that best matches the current audio features. The candidate curve sequence is screened and adjusted through similarity sorting and optimization. Similarity sorting uses a correlation measurement method to prioritize the curves according to their matching degree with the audio features, while sequence optimization adjusts the order of the curves through dynamic programming or greedy algorithms to maximize the overall quality of the output audio. The optimized sequence finally generates an audio curve switching control sequence, which includes gain control parameters, frequency adjustment parameters and phase compensation parameters, which are used to adjust the gain level, frequency distribution and phase characteristics of the output audio respectively.

[0035] 104. Based on the audio curve switching control sequence, dynamically adjust the power amplifier voltage through the feedback controller and collect the power amplifier working status data;

[0036] Specifically, the input signal of the power amplifier is pre-processed according to the gain control parameters, frequency adjustment parameters and phase compensation parameters contained in the audio curve switching control sequence. These parameters are mapped to the signal adjustment domain, and a pre-processed signal matrix is ​​generated through a specific algorithm. The matrix includes gain compensation data, frequency adjustment data and phase correction data. Each type of data optimizes a key characteristic of the input signal of the power amplifier in a targeted manner. For example, gain compensation is used to adjust the signal amplitude to match the target output requirements, frequency adjustment data ensures that the spectrum of the audio signal is consistent with the target frequency response, and phase correction data is used to reduce phase distortion to improve sound quality. The pre-processed signal matrix is ​​input into the dynamic gain controller, and the dynamic gain controller calculates the gain adjustment coefficient through a feedback loop. The feedback loop monitors the difference between the output signal of the power amplifier and the target signal in real time, adjusts the gain parameter to minimize the error, and obtains the power amplifier drive control signal. This signal directly determines the input drive strength of the power amplifier, thereby affecting its output characteristics. Power compensation operation is performed on the power amplifier drive control signal. The power compensation operation is based on the real-time voltage feedback value, and the output power is dynamically adjusted to ensure that the power amplifier always provides a stable output under different load conditions. The calculation result is presented in the form of a power compensation matrix, which contains voltage compensation values ​​and current compensation values, which are used to optimize the voltage and current supply of the power amplifier respectively. According to the power compensation matrix, an amplifier driving voltage signal is generated, which directly drives the power amplifier to work. At the same time, the output voltage data is collected by the voltage detection unit to generate a voltage feedback matrix. The voltage feedback matrix records the instantaneous value of the output voltage and the voltage change rate, which can reflect the real-time state of the power amplifier output. The voltage feedback matrix is ​​filtered by a loop filter. The filtered voltage characteristic sequence contains voltage stability data and voltage ripple data. The former describes the long-term stability of the output voltage, and the latter records the fluctuation of the voltage in a short time. Based on the voltage characteristic sequence, a power amplifier state monitoring model is established. The model uses multi-point sampling technology to monitor the operating state of the power amplifier with high precision and generates an amplifier state parameter matrix. The state parameter matrix contains temperature data, current data and efficiency data, which respectively describe the heat dissipation performance, current consumption characteristics and energy conversion efficiency of the power amplifier. Through multi-dimensional analysis of the power amplifier state parameter matrix, the threshold judgment circuit is used to evaluate the key parameters. For example, when the temperature data exceeds the safety threshold, the system will determine that the amplifier is at risk of overheating, thereby triggering a protection mechanism; when the efficiency data is lower than the specified threshold, the power compensation strategy is adjusted to optimize energy utilization. The amplifier's operating status data is generated through state evaluation. The amplifier's operating status data includes voltage response characteristic data, temperature change characteristic data, and efficiency characteristic data, and is used to adjust the audio output control parameters in real time to form an adaptively optimized closed-loop control system.

[0037] 105. Perform nonlinear distortion compensation and noise suppression processing on the power amplifier working status data to generate audio output quality parameters, where the audio output quality parameters include signal-to-noise ratio parameters, harmonic distortion parameters, and frequency response flatness parameters.

[0038] Specifically, the voltage response characteristic data in the working state data of the power amplifier is subjected to fast Fourier transform, and the spectrum analysis matrix is ​​obtained through frequency domain analysis. The spectrum analysis matrix describes the fundamental component and high-order harmonic component data of the input signal, wherein the fundamental component represents the main energy concentration of the signal, and the high-order harmonic component reveals the frequency component offset caused by the nonlinear distortion in the signal. The spectrum analysis matrix is ​​input into the nonlinear compensation model. The nonlinear compensation model iteratively optimizes the signal characteristics through the back propagation algorithm in deep learning and calculates the compensation weight. The model determines the optimal distortion compensation coefficient matrix by minimizing the objective function (such as distortion error). The distortion compensation coefficient matrix includes amplitude compensation coefficients and phase compensation coefficients, which are used to correct the amplitude distortion and phase distortion of the signal, respectively, so that the restoration of the audio signal is closer to the real effect. According to the matrix, the audio signal is nonlinearly corrected to generate a compensated signal sequence. The compensated signal sequence has lower harmonic distortion while retaining the key characteristics of the original signal. At the same time, the temperature change characteristic data in the working state data of the power amplifier is wavelet decomposed to analyze the noise characteristics. Wavelet decomposition is a time-frequency analysis tool that extracts noise characteristics by decomposing the signal into sub-signals of different frequency ranges, and generates a noise feature vector, which includes background noise data and transient interference data. Background noise data describes the low-frequency noise characteristics that persist in the signal, while transient interference data records the short-term high-amplitude noise components in the signal. The noise feature vector is input into the Wiener filter, and the signal is adaptively filtered using the minimum mean square error criterion. The Wiener filter calculates the noise reduction coefficient sequence based on the statistical characteristics of the signal and noise, and dynamically adjusts the filter parameters to optimize the noise suppression effect. Based on the noise reduction coefficient sequence, the compensated signal sequence is subjected to noise suppression processing to generate a purified audio matrix. The purified audio matrix is ​​subjected to spectral analysis to extract its frequency response characteristic curve, and the curve is smoothed to reduce spectral fluctuations to generate a frequency response characteristic curve. The flatness of the frequency response characteristic curve is an important indicator for measuring the consistency of the audio system, which describes the uniformity of the system's response to signals of different frequencies. At the same time, the signal-to-noise ratio parameter is extracted from the purified audio matrix, which quantifies the clarity and noise suppression effect of the signal; the harmonic distortion parameter is extracted from the compensated signal sequence, which reflects the degree of correction of the signal nonlinear distortion. The signal-to-noise ratio parameter of the purified audio matrix, the harmonic distortion parameter of the compensated signal sequence, and the flatness parameter of the frequency response characteristic curve are combined to generate the audio output quality parameter.

[0039] The fundamental component data and high-order harmonic component data in the spectrum analysis matrix are classified. The frequency components of different harmonic orders are classified into independent frequency groups through the harmonic order classification method to generate a classified harmonic matrix. The classified harmonic matrix describes each frequency point and its corresponding harmonic component. The classified harmonic matrix is ​​feature mapped to convert the physical properties of each harmonic component into a feature space suitable for mathematical analysis to generate a harmonic feature mapping matrix. In the feature mapping process, the main characteristics of nonlinear distortion are accurately captured by considering the amplitude, phase and frequency correlation of the harmonic components. Based on this mapping matrix, a nonlinear compensation model is established. The nonlinear compensation model analyzes the amplitude and phase characteristics of harmonic distortion through the Taylor series expansion method to obtain a nonlinear feature vector. Taylor series expansion decomposes complex nonlinear relationships into polynomial approximations, effectively captures the slight nonlinear changes of the signal, and provides an accurate description of the high-order distortion components. The gradient value of the nonlinear feature vector is calculated through the back propagation algorithm, and the gradient information is used to optimize and update the network parameters of the model. To ensure the efficiency and stability of the optimization process, the Adam optimizer is used to iteratively update the network parameters. The Adam optimizer converges to the optimal solution quickly by adaptively adjusting the learning rate and generates a weight update matrix. The weight update matrix is ​​an important tool to reflect the amplitude of parameter adjustment during network training, and its result directly determines the accuracy of the nonlinear compensation model. Based on the weight update matrix, a Volterra series model is constructed to model the dynamic characteristics of the nonlinear system. The Volterra series model is a classic method for modeling nonlinear systems. It approximates the dynamic response of the signal by superimposing convolution kernel functions of different orders. Through this modeling process, a nonlinear dynamic response function is generated. This function can accurately describe the dynamic characteristics of the signal in frequency, amplitude and phase. After obtaining the nonlinear dynamic response function, it is optimized by the minimum mean square error criterion and the optimal compensation parameters are solved by the conjugate gradient method. The conjugate gradient method accelerates the optimization process by quadratic approximation and finally generates a compensation parameter vector, which contains amplitude and phase compensation data for each frequency point. In order to improve the practicality of the compensation parameters, the compensation parameter vector is smoothed. The smoothing process reduces the high-frequency fluctuations of the parameters through a low-pass filter or an interpolation algorithm to ensure that the transition of the compensation coefficient between different frequency points is smoother, thereby improving the consistency of the compensation effect. The processed smooth compensation sequence is used to reconstruct the distortion compensation coefficient matrix. The row vectors of the distortion compensation coefficient matrix represent different frequency points, while the column vectors correspond to the amplitude compensation coefficient and the phase compensation coefficient respectively. This matrix can intuitively describe the distortion characteristics and compensation strategy of the system at each frequency.

[0040] In the embodiment of the present invention, the MCU control unit realizes the coordinated detection of the input voltage and the audio signal, and the deep neural network is combined to extract and analyze the features, which significantly improves the accuracy and adaptability of the audio output control. The cross-attention model is used to dynamically calculate the weights of the audio parameters, realize the intelligent switching of the audio curve, and effectively solve the problem of unstable audio output under different power supply conditions. Through the dynamic adjustment and feedback control mechanism of the power amplifier voltage, combined with nonlinear distortion compensation and noise suppression processing, the signal-to-noise ratio and frequency response flatness of the audio output are greatly improved. The deep learning algorithm is innovatively applied to audio feature extraction and parameter optimization, which realizes the intelligence and precision of audio output control and enables the system to have stronger environmental adaptability. The modular design concept is adopted, and the interfaces between the functional units are standardized, which improves the scalability and maintainability of the system and facilitates subsequent functional upgrades and optimization. Through multi-level signal processing and optimization strategies, the audio output quality is comprehensively improved, meeting the high-quality audio output requirements in different application scenarios.

[0041] In a specific embodiment, the process of executing step 101 may specifically include the following steps:

[0042] The MCU control unit identifies the input voltage signal through the fast charging protocol to obtain the protocol type identifier, which includes USB PD3.0, PD2.0 and BC1.2 boost fast charging protocol identifiers;

[0043] The input voltage value is sampled and detected according to the protocol type identifier to obtain a voltage sampling sequence, and voltage features are extracted from the voltage sampling sequence to obtain a voltage parameter matrix;

[0044] Perform multi-channel sampling on the audio input signal to obtain audio sampling data, and perform audio feature extraction on the audio sampling data to obtain an audio parameter matrix, where the audio parameter matrix includes frequency response data, phase characteristic data, and dynamic range data;

[0045] The voltage parameter matrix is ​​standardized to obtain a normalized voltage matrix, and the audio parameter matrix is ​​standardized to obtain a normalized audio matrix. The normalized voltage matrix and the normalized audio matrix are combined to obtain a mixed feature matrix.

[0046] Specifically, the fast charging protocol recognition is realized through the MCU control unit. The fast charging protocol recognition is based on the preliminary analysis of the voltage signal. By capturing the handshake signal and mode flag, the protocol type to which the voltage signal belongs is determined. Assuming the input voltage signal is , whose characteristics include voltage waveform characteristics under different fast charging protocols. By analyzing the signal in the time domain, the protocol flag data is extracted , and determine the protocol type PD3.0,PD2.0,BC1.2 According to the protocol type identification, the input voltage signal Perform sampling detection to generate a voltage sampling sequence ,in Indicated in The voltage value of the sampling point, is the number of sampling points. The time interval of voltage sampling is determined by the sampling frequency The specific sampling value is expressed as: ;

[0047] Through this sequence, the voltage signal is feature extracted to generate a voltage parameter matrix. , which contains the average ,variance , maximum value and frequency components For example, the mean and variance are calculated by the following formulas: ;

[0048] Frequency Components Calculated by fast Fourier transform: ;

[0049] At the same time, the audio input signal Multi-channel sampling is performed. Assume that the audio signal has sampling channels, the audio sampling data generated by multi-channel sampling is represented as a matrix ,in Indicates The channel in The signal value of each sampling point. For each channel’s sampling data, the frequency response, phase characteristics, and dynamic range feature extraction are performed. For example, the frequency response Calculated by the following formula: ;

[0050] in, is the frequency component of the audio output, is the frequency component of the audio input. Phase characteristics Represents the phase shift at different frequencies: ;

[0051] Dynamic Range Indicates the ratio of the maximum amplitude to the minimum amplitude of the signal: ;

[0052] After feature extraction, the audio sample data is integrated into an audio parameter matrix , which contains frequency response data, phase characteristic data and dynamic range data. For the voltage parameter matrix and the audio parameter matrix Standardization is performed to eliminate the differences between different feature dimensions and use the normalization method:

[0053] ;

[0054] in, and They are The mean and standard deviation of each column in , and They are The mean and standard deviation of each column in the normalized voltage matrix and the normalized audio matrix Combine to generate a mixed feature matrix , which is of the form:

[0055] ;

[0056] Mixed feature matrix The features of voltage and audio signals are fused into a unified representation to provide data input for subsequent deep learning or optimization algorithms.

[0057] In a specific embodiment, the process of executing step 102 may specifically include the following steps:

[0058] The normalized voltage matrix in the mixed feature matrix is ​​input into a feature extraction network containing three convolutional layers in a deep neural network for processing. Each convolutional layer uses a ReLU activation function and a batch normalization layer to obtain a voltage deep feature vector.

[0059] The normalized audio matrix is ​​input into the five-layer convolutional neural network in the deep neural network for spectrum analysis. Each layer of the convolutional neural network is followed by a BatchNormalization layer and a LeakyReLU activation function to obtain the audio deep feature vector.

[0060] Perform feature fusion on the voltage deep feature vector and the audio deep feature vector to generate a fused feature vector, and input the fused feature vector into a four-layer fully connected neural network in the deep neural network for mapping conversion to obtain a high-dimensional feature vector;

[0061] Perform deep residual calculation on the high-dimensional feature vector to obtain the time-frequency feature matrix, and perform time series feature extraction on the time-frequency feature matrix to obtain the dynamic feature vector;

[0062] The dynamic feature vector is input into the decoder network in the deep neural network for feature reconstruction. The decoder network contains four deconvolution layers and two fully connected layers to obtain the audio feature space. The audio feature space contains frequency response feature data, phase feature data and harmonic feature data.

[0063] Specifically, the normalized voltage matrix in the mixed feature matrix Input the feature extraction network of the deep neural network. The feature extraction network consists of three convolutional layers, each of which contains a convolution operation, a ReLU activation function, and a batch normalization layer. The convolution operation is performed through a two-dimensional convolution kernel. Extract local features, and the convolution output is expressed as: ;

[0064] in is the value of the output feature map, is the bias term, and are the height and width of the convolution kernel respectively. The ReLU activation function is used to introduce nonlinearity, and the output after activation is: ;

[0065] Batch Normalization stabilizes the training process by adjusting the mean and variance of the feature map. The normalization formula is: ;

[0066] in and are the mean and variance of the batch data, and is a trainable parameter, is a small constant to prevent the denominator from being zero. After being processed by the three-layer convolutional network, the voltage depth feature vector is generated. ,in is the feature dimension. At the same time, the normalized audio matrix Enter a five-layer convolutional neural network, which is used for spectrum analysis. Each convolution layer is followed by a BatchNormalization layer and a LeakyReLU activation function. The activation formula of LeakyReLU is: ;

[0067] in is a negative slope parameter, set to 0.01. Through five layers of convolution operations, the spectral features of the audio signal are extracted, and finally the audio deep feature vector is generated. ,in is the feature dimension. The voltage depth feature vector and the audio depth feature vector Perform feature fusion: ;

[0068] Or use the attention mechanism for weighted fusion. The fused vector Input a four-layer fully connected neural network for mapping conversion. The fully connected operation of each layer is expressed as: ;

[0069] in It is The output of the layer, is the weight matrix, is the bias, is the activation function. Finally, a high-dimensional feature vector is generated . High-dimensional feature vector Generate the time-frequency feature matrix through deep residual calculation The residual calculation retains the low-level information by introducing skip connections:

[0070] ;

[0071] in is a nonlinear mapping function. The time-frequency feature matrix Input the time series model (such as LSTM or GRU) to extract time series features and generate dynamic feature vectors The dynamic feature vector is input into the decoder network for feature reconstruction. The decoder consists of four deconvolution layers and two fully connected layers. Deconvolution is used to upsample features. The formula is: ;

[0072] in is the decoded output, is the value of the input feature map. After deconvolution and full connection layer processing, the audio feature space is finally generated. , which contains the frequency response characteristic data , phase characteristic data and harmonic characteristic data .

[0073] In a specific embodiment, the process of executing step 103 may specifically include the following steps:

[0074] Performing feature projection on the frequency response feature data in the audio feature space to obtain a frequency feature vector, and performing feature projection on the phase feature data and harmonic feature data in the audio feature space to obtain an audio representation vector;

[0075] The frequency feature vector is used as the query vector, the audio representation vector is used as the key vector, and the similarity matrix is ​​obtained through dot product operation;

[0076] The similarity matrix is ​​normalized and weighted summed with the audio representation vector to obtain an attention feature vector, which includes a frequency response weight and a phase harmonic weight;

[0077] An audio feature association graph is constructed based on the attention feature vector. The association features between nodes are extracted through a graph convolutional network to obtain an audio parameter association matrix. The audio parameter association matrix is ​​then clustered to obtain an audio curve category matrix.

[0078] Matching the curve templates in the audio curve library based on the audio curve category matrix to obtain a candidate curve sequence, which includes a gain curve, a frequency curve and a phase curve;

[0079] The candidate curve sequences are sorted according to similarity and sequence optimized to generate an audio curve switching control sequence, which includes a gain control parameter, a frequency adjustment parameter, and a phase compensation parameter.

[0080] Specifically, frequency response feature data is extracted from the audio feature space , which represents the response strength of the system at different frequencies. Through linear feature projection, the high-dimensional frequency response features are mapped to low-dimensional frequency feature vectors , the projection formula is: ;

[0081] in is the frequency response feature projection matrix, is the dimension of the original feature, is the dimension after dimensionality reduction. Similarly, for the phase feature data in the audio feature space and harmonic characteristic data Perform joint feature projection to obtain the audio representation vector :

[0082] ;

[0083] in and are the projection matrices of phase feature and harmonic feature respectively. As the query vector, the audio representation vector As the key value vector, calculate the dot product similarity between the two and get the similarity matrix : ;

[0084] The dot product similarity measures the correlation between frequency features and audio representations. To ensure numerical stability and interpretability, the similarity matrix Normalize and use the softmax function: ;

[0085] Normalized similarity matrix Used to perform weighted summation on the audio representation vector to generate the attention feature vector :

[0086] ;

[0087] The attention feature vector contains frequency response weights and phase harmonic weights, which is the core representation of the relationship between audio features. , construct an audio feature association graph , the nodes of the graph represent different audio features, and the edge weights are determined by the weight values ​​in the attention feature vector. The feature extraction of the association graph is performed through the graph convolutional network, and the formula is: ;

[0088] in It is The node feature matrix of the layer, is the normalized adjacency matrix, is a trainable weight matrix, It is a nonlinear activation function. After being processed by a multi-layer graph convolutional network, an audio parameter correlation matrix is ​​generated. , which describes the complex relationship between the audio parameters. Perform feature clustering and use the k-means algorithm to divide it into categories, generate audio curve category matrix , each row represents the feature center of a category. Based on the audio curve category matrix, the closest curve template is matched from the audio curve library. The matching process is performed by calculating the category matrix and the curve library template. The similarity is completed: ;

[0089] After matching, we get the candidate curve sequence , where each curve includes a gain curve, a frequency curve, and a phase curve. The candidate curve sequence is sorted by similarity, and the sequence is optimized through dynamic programming to avoid large jumps during the switching process. Finally, the audio curve switching control sequence is generated ,in is the gain control parameter, is the frequency adjustment parameter, is the phase compensation parameter.

[0090] In a specific embodiment, the process of executing step 104 may specifically include the following steps:

[0091] According to the gain control parameter, frequency adjustment parameter and phase compensation parameter in the audio curve switching control sequence, the power amplifier input signal is pre-processed to obtain a pre-processed signal matrix, which includes gain compensation data, frequency adjustment data and phase correction data;

[0092] The pre-processed signal matrix is ​​input into the dynamic gain controller, and the gain adjustment coefficient is calculated through the feedback loop to obtain the power amplifier driving control signal;

[0093] Performing power compensation calculation on the power amplifier drive control signal, dynamically adjusting the output power based on the voltage feedback value, and obtaining a power compensation matrix, which includes a voltage compensation value and a current compensation value;

[0094] An amplifier driving voltage signal is generated according to the power compensation matrix, and output voltage data is collected through a voltage detection unit to obtain a voltage feedback matrix, which includes an instantaneous voltage value and a voltage change rate;

[0095] Performing signal filtering on the voltage feedback matrix through a loop filter to obtain a filtered voltage characteristic sequence, wherein the voltage characteristic sequence includes voltage stability data and voltage ripple data;

[0096] A power amplifier state monitoring model is established based on the voltage characteristic sequence, and the power amplifier working parameters are obtained through multi-point sampling to obtain a power amplifier state parameter matrix, which includes temperature data, current data and efficiency data;

[0097] A multi-dimensional analysis is performed on the power amplifier state parameter matrix, and a state evaluation is performed based on a threshold judgment circuit to generate power amplifier working state data. The power amplifier working state data includes voltage response characteristic data, temperature change characteristic data, and efficiency characteristic data.

[0098] Specifically, the gain control parameters in the control sequence are switched based on the audio curve. , frequency adjustment parameters and phase compensation parameters , for the amplifier input signal Perform pre-processing. Assume that the input signal is ,in is the signal amplitude, is the frequency, is the initial phase. Gain compensation is achieved by adjusting The signal after gain compensation is: ;

[0099] Frequency adjustment is done by Add adjustment amount The signal after frequency adjustment is: ;

[0100] Phase compensation is achieved by Add compensation The phase-corrected signal is: ;

[0101] After gain, frequency and phase compensation, the final preprocessing signal matrix is ​​formed. , whose rows represent time sampling points, and columns represent gain compensation data, frequency adjustment data, and phase correction data, respectively. Input dynamic gain controller, dynamic gain controller adjusts the gain adjustment coefficient through feedback loop The feedback loop is based on the target output signal The actual output signal Error To make a gain adjustment: ;

[0102] Gain adjustment factor Calculated by a proportional-integral-derivative (PID) controller: ;

[0103] in , , are proportional, integral and differential gains respectively. , generate amplifier drive control signal .Will Input power compensation module, combined with voltage feedback value Dynamically adjust output power and generate power compensation matrix , which contains the voltage compensation value and current compensation value : ;

[0104] and are the target voltage and current. , generating amplifier drive voltage signal , and collect output voltage data through the voltage detection unit By calculating the instantaneous voltage value and voltage change rate , generating the voltage feedback matrix .right The input loop filter is filtered, and the voltage characteristic sequence after filtering Contains voltage stability data and voltage ripple data : ;

[0105] according to Establish a power amplifier status monitoring model and record temperature data through multi-point sampling , current data and efficiency data : ;

[0106] Integrate these parameters into the amplifier state parameter matrix .right Perform multi-dimensional analysis and evaluate key parameters through threshold judgment circuits. , it is judged to be overheated; if After comprehensive evaluation, the amplifier working status data is generated, including voltage response characteristics. , Temperature change characteristics and efficiency characteristics .

[0107] In a specific embodiment, the process of executing step 105 may specifically include the following steps:

[0108] Performing a fast Fourier transform on the voltage response characteristic data in the power amplifier working state data to obtain a spectrum analysis matrix, wherein the spectrum analysis matrix includes fundamental wave component data and high-order harmonic component data;

[0109] The spectrum analysis matrix is ​​input into the nonlinear compensation model, and the compensation weight is calculated by the back propagation algorithm to obtain the distortion compensation coefficient matrix, which includes the amplitude compensation coefficient and the phase compensation coefficient;

[0110] The audio signal is nonlinearly corrected according to the distortion compensation coefficient matrix to obtain a compensated signal sequence, and the temperature change characteristic data is subjected to noise feature analysis by wavelet decomposition to obtain a noise feature vector, which includes background noise data and transient interference data;

[0111] The noise feature vector is input into the Wiener filter, and adaptive filtering is performed based on the minimum mean square error criterion to obtain a noise reduction coefficient sequence, and noise suppression processing is performed on the compensated signal sequence according to the noise reduction coefficient sequence to obtain a purified audio matrix;

[0112] The purified audio matrix is ​​subjected to spectrum analysis, and the frequency response characteristic curve is obtained through smoothing processing. The signal-to-noise ratio parameters of the purified audio matrix, the harmonic distortion parameters of the compensated signal sequence, and the flatness parameters of the frequency response characteristic curve are combined to generate audio output quality parameters.

[0113] Specifically, the voltage response characteristic data is extracted from the power amplifier working state data ,in It's time. In order to analyze the characteristics of the signal in the frequency domain, Perform fast Fourier transform to get the spectrum , and its calculation formula is:

[0114] ;

[0115] in is the frequency, is the signal length, Is an imaginary unit. is a complex number, containing the magnitude and Phase . Spectral analysis matrix From the fundamental component and higher-order harmonic components Composition, of which is the fundamental frequency, is the harmonic order. Input nonlinear compensation model, the compensation model uses deep learning framework and back propagation algorithm to calculate compensation weight. The goal is to minimize fundamental amplitude distortion. and phase distortion , define the loss function as:

[0116]

[0117] in and are the ideal amplitude and phase, and is the weight coefficient. The distortion compensation coefficient matrix is ​​obtained by gradient descent optimization , which contains the amplitude compensation coefficient and phase compensation coefficient : ;

[0118] according to Perform nonlinear correction on the audio signal to obtain the compensated signal sequence : ;

[0119] At the same time, the temperature change characteristic data in the power amplifier working status data Perform wavelet decomposition, the decomposition formula is: ;

[0120] in is the wavelet basis function, is the wavelet coefficient. By decomposing the background noise data and transient disturbance data , which is integrated into the noise feature vector The noise feature vector is input into the Wiener filter, and the noise reduction coefficient sequence is calculated based on the minimum mean square error (MMSE) criterion. : ;

[0121] in and They are signal power spectrum and noise power spectrum respectively. Perform noise suppression on the compensated signal sequence to obtain a purified audio matrix . Perform spectrum analysis on the purified audio matrix and extract the frequency response characteristic curve , and smoothed by moving average method, the formula is: ;

[0122] in is the smoothing window size. By calculating the signal-to-noise ratio (SNR) of the purified audio matrix, the formula is: ;

[0123] The calculation formula of the harmonic distortion parameter (THD) of the compensated signal sequence is: ;

[0124] in is the total harmonic number. Flatness parameter of frequency response characteristic curve It is used to measure the uniformity of response at different frequencies and is defined as: ;

[0125] The signal-to-noise ratio parameter of the purified audio matrix, the harmonic distortion parameter of the compensated signal sequence, and the flatness parameter of the frequency response characteristic curve are combined to generate the audio output quality parameter: .

[0126] Among them, before the compensated signal sequence is subjected to noise suppression processing, it also includes: decomposing the compensated signal sequence into multiple frequency band signals to obtain a multi-scale signal matrix, the multi-scale signal matrix includes low-frequency signal data, medium-frequency signal data and high-frequency signal data; performing oscillation characteristic analysis on each frequency band signal in the multi-scale signal matrix, extracting the envelope signal through Hilbert transform, and obtaining an oscillation feature vector, the oscillation feature vector includes amplitude data, frequency data and phase data; inputting the oscillation feature vector into a hierarchical prediction controller, establishing a prediction model using recursive least squares method, and obtaining a state prediction matrix, the state prediction matrix includes the predicted states of N sampling points in the future; constructing a distributed control network based on the state prediction matrix, and allocating control tasks to multiple sub-controllers, each sub-controller The controller is responsible for the stable control of a specific frequency band, and a distributed control instruction sequence is obtained; the distributed control instruction sequence is collaboratively optimized, and the state synchronization between sub-controllers is achieved through a consistency algorithm to obtain a collaborative control matrix, which contains the control parameters of each frequency band; the collaborative control matrix is ​​input into a Lyapunov stability analyzer, the system stability index is calculated, and a stability evaluation vector is obtained, which contains an energy function value and a convergence rate value; the control parameters are dynamically adjusted according to the stability evaluation vector, and a sliding mode control method is used to suppress system oscillations, and an oscillation suppression sequence is obtained, which contains compensation coefficients and stabilization parameters; the oscillation suppression sequence is applied to signals of each frequency band, and the signal is reconstructed through an adaptive filter to obtain a stabilized signal matrix.

[0127] In a specific embodiment, the execution step inputs the spectrum analysis matrix into the nonlinear compensation model, calculates the compensation weights through the back propagation algorithm, and obtains the distortion compensation coefficient matrix, wherein the distortion compensation coefficient matrix includes the amplitude compensation coefficient and the phase compensation coefficient. The process can specifically include the following steps:

[0128] Performing harmonic order classification on fundamental wave component data and high-order harmonic component data in the spectrum analysis matrix to obtain a classified harmonic matrix, and performing feature mapping on the classified harmonic matrix to obtain a harmonic feature mapping matrix;

[0129] A nonlinear compensation model is established based on the harmonic characteristic mapping matrix, and the amplitude and phase characteristics of the harmonic distortion are calculated through Taylor series expansion to obtain the nonlinear characteristic vector.

[0130] The gradient value of the nonlinear feature vector is calculated by the back propagation algorithm, and the Adam optimizer is used to iteratively update the network parameters to obtain the weight update matrix;

[0131] A Volterra series model is constructed based on the weight update matrix to model the dynamic characteristics of the nonlinear system and obtain the nonlinear dynamic response function. The nonlinear dynamic response function is then optimized using the minimum mean square error method and the optimal compensation parameters are solved using the conjugate gradient method to obtain the compensation parameter vector.

[0132] The compensation parameter vector is parameter-smoothed to obtain a smooth compensation sequence, and the smooth compensation sequence is reconstructed into a distortion compensation coefficient matrix, the row vectors of which represent different frequency points, and the column vectors represent corresponding amplitude compensation coefficients and phase compensation coefficients.

[0133] Specifically, from the spectrum analysis matrix Extract the fundamental component and higher-order harmonic components ,in is the fundamental frequency, is the harmonic order. The fundamental component and the higher-order harmonic component describe the main frequency component of the signal and the frequency component introduced by nonlinear distortion respectively. The components are aggregated into the corresponding classification matrix Assuming the frequency range is , then the classification matrix Row indicates the Amplitude of subharmonics and Phase :

[0134] ;

[0135] Classification Harmonic Matrix Perform feature mapping by mapping the matrix Convert to harmonic eigenmap matrix : ;

[0136] in is a trainable weight matrix used to capture the complex relationship between harmonic features. The mapped matrix Contains advanced representation of harmonic characteristics to provide input to nonlinear compensation models. , establish a nonlinear compensation model. Calculate the amplitude of harmonic distortion through Taylor series expansion and Phase , Taylor expansion formula is: ;

[0137] in is an ideal signal. is the frequency offset, is the order of expansion. By analyzing the nonlinear characteristics of amplitude and phase, the nonlinear eigenvector is obtained. , which contains the characteristics of higher-order harmonics. The nonlinear eigenvector It is input into the deep neural network and the gradient value is calculated through the back propagation algorithm to optimize the network parameters. The loss function Defined as the error between nonlinear distortion and ideal characteristics: ;

[0138] The network parameters are iteratively updated through the Adam optimizer, and the update formula is: ;

[0139] in is the learning rate, and They are the first-order and second-order momentum estimates of the gradient deviation correction. The optimized weight update matrix is ​​used to construct the Volterra series model. The Volterra model describes the dynamic characteristics through nonlinear convolution, and its expression is:

[0140] ;

[0141] in yes The Volterra kernel function of order 1 describes the dynamic characteristics of the system. The optimal compensation parameter vector is solved by the conjugate gradient method through the minimum mean square error (MSE) optimization. : ;

[0142] right Parameter smoothing is performed and a low-pass filter is used to reduce high-frequency fluctuations. The formula is: ;

[0143] Reconstruct the smoothed compensation sequence into a distortion compensation coefficient matrix , whose row vectors represent different frequency points , the column vector contains the amplitude compensation coefficients and phase compensation coefficient : .

[0144] The above describes the audio output control method in the embodiment of the present invention. The following describes the audio output control device in the embodiment of the present invention. Figure 2 , an embodiment of the audio output control device in the embodiment of the present invention includes:

[0145] The acquisition module 201 is used to detect the input voltage signal and acquire the audio input signal through the MCU control unit to obtain a voltage parameter matrix and an audio parameter matrix, and combine the voltage parameter matrix and the audio parameter matrix to form a mixed feature matrix;

[0146] A feature extraction module 202 is used to input the mixed feature matrix into a deep neural network for feature extraction, perform time-frequency domain analysis on the audio signal, and obtain an audio feature space;

[0147] A calculation module 203 is used to construct a cross attention model according to the audio feature space, calculate the association weight values ​​between the audio parameters, and generate an audio curve switching control sequence;

[0148] A dynamic adjustment module 204 is used to switch the control sequence based on the audio curve, dynamically adjust the power amplifier voltage through the feedback controller, and collect the power amplifier working status data;

[0149] The generating module 205 is used to perform nonlinear distortion compensation and noise suppression processing on the power amplifier working state data to generate audio output quality parameters, which include signal-to-noise ratio parameters, harmonic distortion parameters and frequency response flatness parameters.

[0150] Through the collaboration of the above components, the MCU control unit realizes the coordinated detection of input voltage and audio signal, and combines deep neural network for feature extraction and analysis, which significantly improves the accuracy and adaptability of audio output control. The cross-attention model is used to dynamically calculate the weight of audio parameters, realize the intelligent switching of audio curves, and effectively solve the problem of unstable audio output under different power supply conditions. Through the dynamic adjustment and feedback control mechanism of power amplifier voltage, combined with nonlinear distortion compensation and noise suppression processing, the signal-to-noise ratio and frequency response flatness of audio output are greatly improved. The innovative application of deep learning algorithms to audio feature extraction and parameter optimization realizes the intelligent and precise control of audio output, and makes the system have stronger environmental adaptability. The modular design concept is adopted, and the interfaces between functional units are standardized, which improves the scalability and maintainability of the system and facilitates subsequent function upgrades and optimization. Through multi-level signal processing and optimization strategies, the audio output quality is comprehensively improved, meeting the high-quality audio output requirements in different application scenarios.

[0151] above Figure 2 The audio output control device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The audio output control device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0152] Figure 33 is a structural diagram of an audio output control device provided by an embodiment of the present invention. The audio output control device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be temporary storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the audio output control device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the audio output control device 300 to implement the steps of the above-mentioned audio output control method.

[0153] The audio output control device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The illustrated structure of the audio output control device does not constitute a limitation on the audio output control device provided by the present invention, and may include more or fewer components than illustrated, or a combination of certain components, or a different arrangement of components.

[0154] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the audio output control method.

[0155] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0156] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0157] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An audio output control method, characterized in that: The method comprises: The MCU control unit detects the input voltage signal and collects the audio input signal to obtain a voltage parameter matrix and an audio parameter matrix, and combines the voltage parameter matrix with the audio parameter matrix to form a mixed feature matrix; Inputting the mixed feature matrix into a deep neural network for feature extraction, performing time-frequency domain analysis on the audio signal, and obtaining an audio feature space; Constructing a cross attention model according to the audio feature space, calculating the association weight values ​​between audio parameters, and generating an audio curve switching control sequence; Based on the audio curve switching control sequence, dynamically adjusting the power amplifier voltage through a feedback controller, and collecting the power amplifier working status data; Nonlinear distortion compensation and noise suppression processing are performed on the power amplifier working state data to generate audio output quality parameters, wherein the audio output quality parameters include signal-to-noise ratio parameters, harmonic distortion parameters, and frequency response flatness parameters.

2. The audio output control method according to claim 1, characterized in that: The MCU control unit detects the input voltage signal and collects the audio input signal to obtain a voltage parameter matrix and an audio parameter matrix, and combines the voltage parameter matrix with the audio parameter matrix to form a mixed feature matrix, including: The MCU control unit performs fast charging protocol identification on the input voltage signal to obtain a protocol type identifier, wherein the protocol type identifier includes USB PD3.0, PD2.0 and BC1.2 boost fast charging protocol identifiers; Sampling and detecting the input voltage value according to the protocol type identifier to obtain a voltage sampling sequence, and extracting voltage features from the voltage sampling sequence to obtain a voltage parameter matrix; Perform multi-channel sampling on an audio input signal to obtain audio sampling data, and perform audio feature extraction on the audio sampling data to obtain an audio parameter matrix, wherein the audio parameter matrix includes frequency response data, phase characteristic data, and dynamic range data; The voltage parameter matrix is ​​standardized to obtain a normalized voltage matrix, and the audio parameter matrix is ​​standardized to obtain a normalized audio matrix, and the normalized voltage matrix and the normalized audio matrix are combined to obtain a mixed feature matrix.

3. The audio output control method according to claim 2, characterized in that: The mixed feature matrix is ​​input into a deep neural network for feature extraction, and the audio signal is analyzed in the time-frequency domain to obtain an audio feature space, including: Inputting the normalized voltage matrix in the mixed feature matrix into a feature extraction network including three convolutional layers in a deep neural network for processing, each convolutional layer adopts a ReLU activation function and a batch normalization layer to obtain a voltage deep feature vector; Inputting the normalized audio matrix into the five-layer convolutional neural network in the deep neural network for spectrum analysis, each layer of the convolutional neural network is followed by a BatchNormalization layer and a LeakyReLU activation function to obtain an audio deep feature vector; Performing feature fusion on the voltage deep feature vector and the audio deep feature vector to generate a fused feature vector, and inputting the fused feature vector into a four-layer fully connected neural network in the deep neural network for mapping conversion to obtain a high-dimensional feature vector; Performing deep residual calculation on the high-dimensional feature vector to obtain a time-frequency feature matrix, and performing time series feature extraction on the time-frequency feature matrix to obtain a dynamic feature vector; The dynamic feature vector is input into a decoder network in the deep neural network for feature reconstruction, wherein the decoder network comprises four deconvolution layers and two fully connected layers, and an audio feature space is obtained, wherein the audio feature space comprises frequency response feature data, phase feature data and harmonic feature data.

4. The audio output control method according to claim 3, characterized in that: The method of constructing a cross attention model according to the audio feature space, calculating the association weight values ​​between audio parameters, and generating an audio curve switching control sequence includes: Performing feature projection on the frequency response feature data in the audio feature space to obtain a frequency feature vector, and performing feature projection on the phase feature data and harmonic feature data in the audio feature space to obtain an audio representation vector; The frequency feature vector is used as a query vector, the audio representation vector is used as a key value vector, and a similarity matrix is ​​obtained by performing a dot product operation; Normalizing the similarity matrix and performing weighted summation on the similarity matrix and the audio representation vector to obtain an attention feature vector, wherein the attention feature vector includes a frequency response weight and a phase harmonic weight; An audio feature association graph is constructed according to the attention feature vector, association features between nodes are extracted through a graph convolutional network to obtain an audio parameter association matrix, and feature clustering is performed on the audio parameter association matrix to obtain an audio curve category matrix; Matching curve templates in an audio curve library based on the audio curve category matrix to obtain a candidate curve sequence, wherein the candidate curve sequence includes a gain curve, a frequency curve, and a phase curve; The candidate curve sequences are sorted according to similarity and sequence optimized to generate an audio curve switching control sequence, wherein the audio curve switching control sequence includes a gain control parameter, a frequency adjustment parameter, and a phase compensation parameter.

5. The audio output control method according to claim 4, characterized in that: The switching control sequence based on the audio curve dynamically adjusts the power amplifier voltage through a feedback controller and collects the power amplifier working status data, including: Pre-processing the power amplifier input signal according to the gain control parameter, frequency adjustment parameter and phase compensation parameter in the audio curve switching control sequence to obtain a pre-processing signal matrix, wherein the pre-processing signal matrix includes gain compensation data, frequency adjustment data and phase correction data; The pre-processed signal matrix is ​​input into a dynamic gain controller, and a gain adjustment coefficient is calculated through a feedback loop to obtain a power amplifier drive control signal; Performing power compensation operation on the power amplifier drive control signal, dynamically adjusting the output power based on the voltage feedback value, and obtaining a power compensation matrix, wherein the power compensation matrix includes a voltage compensation value and a current compensation value; Generate an amplifier drive voltage signal according to the power compensation matrix, and collect output voltage data through a voltage detection unit to obtain a voltage feedback matrix, wherein the voltage feedback matrix includes an instantaneous voltage value and a voltage change rate; Performing signal filtering on the voltage feedback matrix through a loop filter to obtain a filtered voltage characteristic sequence, wherein the voltage characteristic sequence includes voltage stability data and voltage ripple data; A power amplifier state monitoring model is established according to the voltage characteristic sequence, and the power amplifier operating parameters are acquired through multi-point sampling to obtain a power amplifier state parameter matrix, wherein the power amplifier state parameter matrix includes temperature data, current data and efficiency data; A multi-dimensional analysis is performed on the power amplifier state parameter matrix, and a state evaluation is performed based on a threshold judgment circuit to generate power amplifier working state data, wherein the power amplifier working state data includes voltage response characteristic data, temperature change characteristic data, and efficiency characteristic data.

6. The audio output control method according to claim 5, characterized in that: The nonlinear distortion compensation and noise suppression processing are performed on the power amplifier working state data to generate audio output quality parameters, wherein the audio output quality parameters include a signal-to-noise ratio parameter, a harmonic distortion parameter, and a frequency response flatness parameter, including: Performing a fast Fourier transform on the voltage response characteristic data in the power amplifier working state data to obtain a spectrum analysis matrix, wherein the spectrum analysis matrix includes fundamental wave component data and high-order harmonic component data; Inputting the spectrum analysis matrix into a nonlinear compensation model, calculating compensation weights through a back propagation algorithm, and obtaining a distortion compensation coefficient matrix, wherein the distortion compensation coefficient matrix includes amplitude compensation coefficients and phase compensation coefficients; Performing nonlinear correction on the audio signal according to the distortion compensation coefficient matrix to obtain a compensated signal sequence, and performing noise feature analysis on the temperature change characteristic data by wavelet decomposition to obtain a noise feature vector, wherein the noise feature vector includes background noise data and transient interference data; Input the noise feature vector into a Wiener filter, perform adaptive filtering based on a minimum mean square error criterion to obtain a noise reduction coefficient sequence, and perform noise suppression processing on the compensated signal sequence according to the noise reduction coefficient sequence to obtain a purified audio matrix; The purified audio matrix is ​​subjected to spectrum analysis, a frequency response characteristic curve is obtained through smoothing processing, and a signal-to-noise ratio parameter of the purified audio matrix, a harmonic distortion parameter of the compensated signal sequence, and a flatness parameter of the frequency response characteristic curve are combined to generate an audio output quality parameter.

7. The audio output control method according to claim 6, characterized in that: The spectrum analysis matrix is ​​input into the nonlinear compensation model, and the compensation weight is calculated by the back propagation algorithm to obtain a distortion compensation coefficient matrix, wherein the distortion compensation coefficient matrix includes amplitude compensation coefficients and phase compensation coefficients, including: Classifying the fundamental wave component data and the high-order harmonic component data in the spectrum analysis matrix according to harmonic orders to obtain a classified harmonic matrix, and performing feature mapping on the classified harmonic matrix to obtain a harmonic feature mapping matrix; A nonlinear compensation model is established according to the harmonic characteristic mapping matrix, and the amplitude and phase characteristics of the harmonic distortion are calculated by Taylor series expansion to obtain a nonlinear characteristic vector; The nonlinear feature vector is calculated by back propagation algorithm, and the network parameters are iteratively updated by Adam optimizer to obtain a weight update matrix. A Volterra series model is constructed based on the weight update matrix to model the dynamic characteristics of the nonlinear system to obtain a nonlinear dynamic response function, and the nonlinear dynamic response function is optimized by minimum mean square error, and the optimal compensation parameter is solved by the conjugate gradient method to obtain a compensation parameter vector; The compensation parameter vector is parameter-smoothed to obtain a smooth compensation sequence, and the smooth compensation sequence is reconstructed into a distortion compensation coefficient matrix, wherein the row vectors of the distortion compensation coefficient matrix represent different frequency points, and the column vectors represent corresponding amplitude compensation coefficients and phase compensation coefficients.

8. An audio output control device, characterized in that: For executing the audio output control method according to any one of claims 1 to 7, the device comprises: An acquisition module is used to detect an input voltage signal and acquire an audio input signal through an MCU control unit to obtain a voltage parameter matrix and an audio parameter matrix, and combine the voltage parameter matrix with the audio parameter matrix to form a mixed feature matrix; A feature extraction module is used to input the mixed feature matrix into a deep neural network for feature extraction, perform time-frequency domain analysis on the audio signal, and obtain an audio feature space; A calculation module, used to construct a cross attention model according to the audio feature space, calculate the association weight values ​​between audio parameters, and generate an audio curve switching control sequence; A dynamic adjustment module, used to switch the control sequence based on the audio curve, dynamically adjust the power amplifier voltage through a feedback controller, and collect the power amplifier working status data; The generating module is used to perform nonlinear distortion compensation and noise suppression processing on the power amplifier working state data to generate audio output quality parameters, wherein the audio output quality parameters include signal-to-noise ratio parameters, harmonic distortion parameters and frequency response flatness parameters.

9. An audio output control device, characterized in that: The audio output control device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the audio output control device to execute the audio output control method according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by a processor, the audio output control method according to any one of claims 1 to 7 is implemented.

Images