Recording pen audio compression storage method, device and equipment

Abstract

The invention relates to a recording pen audio compression storage method, device and equipment. The method comprises the following steps: preprocessing original audio data to obtain an audio frequency domain signal, and carrying out operation to obtain a frequency reference signal and a frequency variation; performing K-SVD algorithm training to obtain a spectrum feature vector, performing formant detection, and constructing compressed mask data; inputting the spectrum feature vector into a variational auto-encoder for nonlinear dimension reduction to obtain feature encoding data, and performing complexity calculation to obtain a compression control parameter; performing orthogonal matching pursuit operation based on the compression control parameters and the compression mask data to obtain a standard compression matrix; and performing matrix transformation and CP-ABE encryption on the standard compression matrix and the original audio data to obtain an encrypted audio stream, and storing the encrypted audio stream into a memory in the recording pen. According to the invention, the compression processing efficiency and the audio quality are improved, so that the recording pen can realize a longer-time recording function in a limited storage space.

Description

Technical Field

[0001] The present invention relates to the technical field of audio compression, and in particular to a method, device and equipment for compressing and storing audio of a voice recorder. Background Art

[0002] As an important audio acquisition device, voice recorders are widely used in various fields. However, voice recorders will generate a large amount of audio data during long-term recording, which not only takes up a lot of storage space, but also affects the battery life and user experience of the voice recorder.

[0003] Traditional audio compression methods mainly use lossy compression or lossless compression technology. These methods often require periodic segmentation of audio data during the compression process, resulting in low compression efficiency and difficulty in ensuring audio quality. At the same time, existing audio compression algorithms lack in-depth analysis of audio signal characteristics and cannot perform adaptive compression based on the characteristics of different types of audio, resulting in unstable compression effects. Summary of the invention

[0004] The main purpose of the present invention is to provide a method, device and equipment for compressing and storing audio of a voice recorder. The present invention improves the compression processing efficiency and audio quality, so that the voice recorder can achieve a longer recording function within a limited storage space.

[0005] To achieve the above object, the present invention provides a method for compressing and storing audio of a voice recorder, comprising the following steps: Preprocessing the original audio data to obtain an audio frequency domain signal, performing a high-order synchronous compression FSSTN operation on the audio frequency domain signal to obtain a frequency reference signal and a frequency variation; Performing K-SVD algorithm training on the audio frequency domain signal to obtain a spectrum feature vector, and performing resonance peak detection on the spectrum feature vector based on the frequency reference signal and the frequency variation to construct compression mask data; Inputting the spectrum feature vector into a variational autoencoder for nonlinear dimensionality reduction to obtain feature encoding data, and performing complexity calculation based on the feature encoding data to obtain compression control parameters; Performing an orthogonal matching pursuit operation based on the compression control parameter and the compression mask data to obtain a standard compression matrix; The standard compression matrix and the original audio data are subjected to matrix transformation and CP-ABE encryption to obtain an encrypted audio stream, and the encrypted audio stream is stored in a memory in the voice recorder.

[0006] The present invention also provides a recording pen audio compression storage device, comprising: A preprocessing module is used to preprocess the original audio data to obtain an audio frequency domain signal, and perform a high-order synchronous compression FSSTN operation on the audio frequency domain signal to obtain a frequency reference signal and a frequency variation; A detection module, used for performing K-SVD algorithm training on the audio frequency domain signal to obtain a spectrum feature vector, and performing resonance peak detection on the spectrum feature vector based on the frequency reference signal and the frequency variation to construct compression mask data; An encoding module, used for inputting the spectrum feature vector into a variational autoencoder for nonlinear dimensionality reduction to obtain feature encoding data, and performing complexity calculation based on the feature encoding data to obtain compression control parameters; A calculation module, used for performing an orthogonal matching pursuit operation based on the compression control parameter and the compression mask data to obtain a standard compression matrix; The encryption module is used to perform matrix transformation and CP-ABE encryption on the standard compression matrix and the original audio data to obtain an encrypted audio stream, and store the encrypted audio stream into a memory in the voice recorder.

[0007] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0008] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0009] In summary, the technical solution provided by the present invention can realize feature extraction without periodic segmentation of audio data by introducing high-order synchronous compression FSSTN operation and K-SVD algorithm, thereby significantly improving the compression processing efficiency and audio quality; adopting variational autoencoder for nonlinear dimensionality reduction and combining with CBOW model to analyze audio features, realizes adaptive compression of audio data and makes the compression effect more stable and reliable; constructs a standard compression matrix through orthogonal matching pursuit algorithm to ensure the reversibility of the compression process and effectively avoid the distortion of audio data during the compression process; introduces CP-ABE encryption technology to encrypt and store compressed audio data, thereby achieving data security protection while ensuring compression efficiency; establishes a complete audio data compression-encryption-decryption-restoration process, so that the voice recorder can realize longer recording function within limited storage space. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 Schematic diagram of the steps of a method for compressing and storing audio in a voice recorder according to an embodiment of the present invention; Figure 2It is a structural block diagram of an audio compression storage device of a voice recorder in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0011] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0013] Reference Figure 1 This embodiment provides a method for compressing and storing audio of a voice recorder, comprising the following steps: S1, preprocessing the original audio data to obtain an audio frequency domain signal, performing a high-order synchronous compression FSSTN operation on the audio frequency domain signal to obtain a frequency reference signal and a frequency variation; Among them, the external sound signal is captured by the built-in audio acquisition unit of the recorder. The audio acquisition unit in the recorder is composed of a high-sensitivity microphone, which can effectively pick up audio signals in different environments and reduce the interference of background noise, ensure the quality of the signal, and obtain the original audio data. The original audio data is converted into digital quantity, and the analog signal is converted into a digital signal to obtain digital audio data. The digital audio data is segmented according to a fixed time length. The setting of the time length needs to be optimized according to the actual application scenario, between a few milliseconds and tens of milliseconds, to ensure that each audio segment can fully express its spectral characteristics without generating computational redundancy. Through the segmentation operation, the continuous audio signal is divided into multiple independent segments to form an audio segment sequence. The audio segment sequence is input into the discrete Fourier transform unit, and the Fourier transform is used to convert each audio data from the time domain to the frequency domain to generate the corresponding audio frequency domain signal, which contains the frequency component and amplitude information of the audio. The audio frequency domain signal is subjected to periodic signal duration extraction, and the periodic components in the signal are identified and analyzed for duration, thereby determining the jitter characteristics of the audio signal, which are usually caused by small frequency changes or amplitude fluctuations. The audio jitter data is input into the Fourier high-order synchronous compression FSSTN processing unit for feature decomposition. High-order synchronous compression can more accurately extract the instantaneous frequency components in the signal by combining time domain and frequency domain information. The FSSTN processing unit uses synchronous sparse time-frequency transform technology to identify the main frequency components in the signal, and simultaneously analyzes its phase information to generate a frequency feature sequence. The frequency feature sequence is subjected to phase feature extraction to obtain a frequency reference signal. The phase change law of the frequency signal is used to restore the reference frequency component of the signal. The frequency reference signal is a stable reference signal that represents the main frequency component in the audio signal and is used to reconstruct the main characteristics of the audio data. The frequency reference signal is subjected to time domain differential operation to obtain the frequency change, which reflects the subtle change trend of the signal in the time dimension.

[0014] S2, training the audio frequency domain signal with a K-SVD algorithm to obtain a spectrum feature vector, and performing resonance peak detection on the spectrum feature vector based on a frequency reference signal and a frequency variation to construct compressed mask data; Specifically, the audio frequency domain signal is sub-divided according to a preset frequency interval. The selection of the preset frequency interval is based on the spectral distribution characteristics of the audio signal. These intervals will cover the entire spectrum range and divide the signal into multiple frequency band grouping data. The frequency band grouping data is sparsely represented, and the original data is approximated using as few basis vectors as possible, so that the frequency band grouping data is converted into a feature basic matrix. The feature basic matrix is ​​input into the K-SVD algorithm, and the feature basic matrix is ​​sparsely learned through the iterative optimization process of K-SVD to generate a spectrum feature vector. As an optimization algorithm for sparse signal processing, the K-SVD algorithm can improve the expression ability of the feature vector by continuously updating the dictionary and sparse representation coefficients, so that the spectrum feature vector is not only more sparse, but also can more accurately describe the spectrum information of the audio signal. The frequency reference signal and the frequency change are used to extract the features of the spectrum feature vector. The spectrum feature vector is decomposed by the singular value decomposition technology to extract the main spectrum basic features. Singular value decomposition can extract the most representative components from high-dimensional data, thereby reducing the complexity of the data while retaining its main characteristics to obtain a set of spectrum basic features. Perform a formant extraction operation on the basic features of the spectrum. Formants are frequency components with high energy and high stability in audio signals, which have an important impact on the speech characteristics and perceived quality of audio signals. By analyzing the local maximum positions of the basic features of the spectrum, the formants in the signal are identified, and the formant position data are stored as spectrum peak position data. Perform a spectrum quality assessment on the spectrum peak position data. The assessment process calculates the spectrum quality threshold by analyzing the energy distribution, stability and prominence of the formants of the signal. The spectrum quality threshold is used as a quantitative indicator to distinguish high-quality spectrum features from noise or redundant information. According to the spectrum quality threshold, the spectrum feature vector is binarized. By setting the threshold, the data in the spectrum feature vector is divided into important features and unimportant features, and a spectrum labeling matrix is ​​generated. The elements of this matrix are binary values ​​(such as 0 and 1) to identify the importance of each frequency component. The spectrum labeling matrix and the spectrum feature vector are subjected to matrix operations. By applying the labeling matrix as mask data to the spectrum feature vector, the redundant information is effectively eliminated and the key information is retained, and the compressed mask data is obtained.

[0015] S3, inputting the spectrum feature vector into the variational autoencoder for nonlinear dimensionality reduction to obtain feature encoding data, and performing complexity calculation based on the feature encoding data to obtain compression control parameters; It should be noted that the spectral feature vector is normalized to adjust the data of the feature vector to a unified numerical range, usually between 0 and 1 or -1 and 1, so as to eliminate the dimensional differences of different feature dimensions, improve the subsequent model's ability to analyze features and training efficiency, and obtain standardized feature data. The standardized feature data is input into the encoding network of the variational autoencoder for nonlinear mapping. As an important part of VAE, the encoding network compresses the high-dimensional spectral feature vector into a low-dimensional latent variable matrix through a series of nonlinear transformations. The latent variable matrix is ​​generated by the unique reparameterization technique of the variational autoencoder. During the encoding process, the variational autoencoder generates a mean and variance parameter for each latent variable, and then samples these parameters through the reparameterization technique to obtain the latent variable matrix. Reparameterization can effectively avoid the problem of gradient transfer caused by direct sampling, and ensure the continuity and distribution characteristics of the latent variable matrix. The latent variable matrix is ​​processed and converted into feature encoding data, and the feature encoding data is segmented into sequences. The segmentation operation is divided according to a preset time window or fixed length, and the continuous feature data is split into multiple small fragments to construct feature sequence data. The feature sequence data is input into the continuous bag of words model (CBOW) for vector training. The CBOW model is a natural language processing technology that predicts the center word vector through context information. In audio data processing, this technology is used to analyze the semantic association between feature sequences and build an audio semantic model. The role of the audio semantic model is to capture the implicit time and frequency relationships in the feature data and give the feature sequence a higher level of expression ability. After completing the semantic modeling, the feature density of the audio semantic model is calculated. By analyzing the compactness and sparsity of the feature distribution in the model, a complexity matrix is ​​generated to reflect the degree of change of the audio signal in different time periods and frequency intervals. The hierarchical interval division is performed based on the complexity matrix. According to the numerical range in the complexity matrix, the audio features are divided into different complexity level data, so that the compression parameters can dynamically adapt to the characteristics of audio signals of different levels. According to the complexity level data, the corresponding quantization bits, compression ratio and fidelity parameters are set. Among them, the quantization bits determine the accuracy of data representation, the compression ratio controls the degree of data compression, and the fidelity parameter is used to balance the relationship between the compression rate and the quality of the audio signal. By combining these parameters, the final compression control parameters are generated.

[0016] S4, performing an orthogonal matching pursuit operation based on the compression control parameter and the compression mask data to obtain a standard compression matrix; Specifically, the compression control parameters are input into the compression domain mapping unit, and the parameters are transformed through the unit to obtain the initial measurement matrix. By analyzing the sparse characteristics of the audio signal, the sparsity analysis is performed on the initial measurement matrix to determine the optimal basis functions. These basis functions represent the sparse directions with the most information content of the signal and can effectively capture the main features of the audio signal. The optimal basis functions and the compressed mask data are input into the orthogonal matching pursuit (OMP) unit to generate the compressed basis matrix through iterative optimization. OMP is a greedy algorithm that can select the basis function with the highest matching degree with the signal in each iteration, and continuously adjust the current solution to approach the target signal. In this process, the optimal basis function and the compressed mask data are combined to achieve accurate sparse decomposition of the signal through multiple rounds of iterations, and a set of compressed basis matrices with high representation capabilities are obtained. The Gram-Schmidt orthogonalization operation is performed on the compressed basis matrix. Orthogonalization is an important step to improve the independence of the basis matrix. By transforming the original basis vectors into a set of mutually orthogonal basis vectors, the correlation between the columns in the matrix can be effectively reduced, and the stability of the compressed matrix can be enhanced. The orthogonalized basis data form orthogonal basis data, which has stronger numerical robustness and computational stability. The orthogonal basis data is linearly transformed to generate reversible compressed data. The design of linear transformation combines the characteristics of audio signals to ensure reversibility during data compression, so that high-quality signal restoration can be achieved in the decompression stage. The condition number is calculated for the reversible compressed data. By analyzing the condition number of the matrix, its numerical stability and sensitivity to signal disturbances can be evaluated. A matrix with a high condition number will lead to a decrease in decompression accuracy. The optimization parameters are generated by condition number calculation to guide matrix adjustment. On this basis, the optimization parameters are combined with the reversible compressed data to adjust the matrix to generate an adaptive matrix. The generation process of the adaptive matrix takes into account the complexity, sparsity and characteristics of the compression control parameters of the signal to achieve adaptive optimization for different signals. Time-frequency parameter optimization is performed on the adaptive matrix. By adjusting the time domain and frequency domain characteristics of the matrix, it can better adapt to the dynamic changes and frequency distribution of the audio signal during the compression process. The time-frequency parameter optimization combines technologies such as short-time Fourier transform (STFT) and compressed sensing (CS) to adjust the time-frequency resolution and smoothness of the matrix and ultimately generate a standard compressed matrix.

[0017] S5, matrix transformation and CP-ABE encryption are performed on the standard compression matrix and the original audio data to obtain an encrypted audio stream, and the encrypted audio stream is stored in the memory in the voice recorder.

[0018] Among them, the original audio data is subjected to matrix multiplication operation with the standard compression matrix, and the high-dimensional original audio data is projected into the low-dimensional space constructed by the standard compression matrix to obtain matrix compressed data. The matrix compressed data is rounded according to the quantization step size, and the quantization process discretizes the continuous signal value to obtain the initial compressed data. The initial compressed data is converted to a fixed-point number according to the preset number of quantization bits to obtain quantized data. The compressed data represented by the floating point number is converted into an integer format of fixed length, thereby simplifying the complexity of data storage and transmission, and controlling the data accuracy to meet different storage requirements. The CP-ABE access policy is constructed for the quantized data. CP-ABE encryption is an attribute-based access control method. By defining a set of access policies, the access rights of different users or devices to encrypted data can be flexibly controlled. By constructing a ciphertext policy associated with the audio data, ciphertext policy data is generated, and the ciphertext policy data contains information such as user permissions and access conditions. The ciphertext policy data and the quantized data are encrypted to generate an encrypted audio stream. The encryption operation combines the access control capability of CP-ABE encryption technology and the efficient storage characteristics of matrix compression technology. By embedding the ciphertext strategy into the audio data, the security protection and efficient compression of the data are integrated. The encrypted audio stream has a small data volume and contains a complete description of the original audio data. After the encrypted audio stream is generated, the compression information and key information inside it are identified and encapsulated. The encapsulation process clarifies the attributes and structure of the data by defining file identification data, such as recording compression ratio, quantization accuracy, encryption key identifier and other information. These identification data will provide necessary meta information for subsequent data decryption and recovery. After the identification encapsulation is completed, the file identification data and the encrypted audio stream are format converted. The encrypted audio stream is organized into a specific data structure that meets the storage requirements of the recorder. This process includes the insertion of file header information, the alignment of data blocks, and the adaptation of specific file formats, such as the extension of compression formats such as WAV and MP3. The data after format conversion is the storage data. The storage data is written and saved to the memory of the recorder to complete the entire compression encryption storage process.

[0019] File identification data is read from the memory, which contains compressed information and encrypted information generated during the storage process. The encryption parameters and decryption keys are separated through information extraction operations. The encryption parameters are used for subsequent data decompression and matrix reconstruction, while the decryption key is the key input for decrypting audio data. The extracted decryption key is input into the CP-ABE decryption unit, which performs a decryption operation on the encrypted audio data based on the access control policy and the decryption key to generate decrypted audio data. The decrypted audio data is subjected to a quantization and anti-normalization operation. Quantization and anti-normalization are the inverse processes of quantization and fixed-point number conversion during the compression process, which restores the discretized compressed data to a floating-point number representation close to the original value. By combining the decrypted audio data and the encryption parameters, the anti-normalization operation can restore the dynamic range and precision of the audio data. The standard compression matrix is ​​reconstructed based on the encryption parameters used during the storage process. The linear transformation applied in the compressed storage is reversely derived to restore the standard compression matrix used during compression, and the reconstructed reduction matrix can be used for the inverse transformation of the decompressed data. The decompressed data and the reconstructed reduction matrix are subjected to an inverse transformation operation. By applying the inverse matrix transformation during compression, the audio signal is mapped from the low-dimensional space back to the original high-dimensional space to generate reconstructed audio data. The accuracy of the inverse transformation operation directly affects the restoration quality of the audio data. During the implementation process, it is necessary to ensure the accuracy of the restoration matrix and the stability of the inverse transformation algorithm. The generated reconstructed audio data is numerically close to the original audio signal. In order to verify its usability, it is subjected to quality evaluation. The waveform, spectrum and signal energy distribution of the reconstructed audio data are analyzed by the quality standard of the preset fidelity range. The fidelity range is an important indicator for audio data restoration, which is used to measure the similarity between the decompressed audio data and the original audio signal. If the evaluation result meets the preset fidelity range, the audio data is determined to be qualified audio data. The qualified audio data is input into the digital-to-analog converter, which converts the audio data in digital format into an analog signal to generate a playback signal. The playback signal is input into the power amplifier unit. The power amplifier unit performs signal enhancement on the playback signal to drive the speaker of the audio playback unit. The sound data is output through the speaker to restore the original audio content stored in the recorder.

[0020] In one example, the original audio data is preprocessed to obtain an audio frequency domain signal, and a high-order synchronous compression FSSTN operation is performed on the audio frequency domain signal to obtain a frequency reference signal and a frequency variation, including: The external sound signal is collected by the audio collection unit in the recording pen to obtain original audio data, and the original audio data is converted into digital quantity to obtain digital audio data; Segmenting the digital audio data according to a fixed time length to obtain an audio segment sequence, and performing discrete Fourier transform on the audio segment sequence to obtain an audio frequency domain signal; The audio frequency domain signal is subjected to periodic signal duration extraction to obtain audio jitter data, and the audio jitter data is input into the Fourier high-order synchronous compression FSSTN processing unit for feature decomposition to obtain a frequency feature sequence; The phase feature is extracted from the frequency feature sequence to obtain a frequency reference signal, and a time domain differential operation is performed on the frequency reference signal to obtain a frequency variation.

[0021] In this example, the audio acquisition unit of the voice recorder is used to collect the sound of the external environment. The audio acquisition unit consists of a high-sensitivity microphone and a front-end amplifier, which can capture sound signals in a wide frequency band and improve the signal-to-noise ratio through noise reduction and filtering. Assume that the collected sound signal is ,in It's time. is the amplitude of the analog signal. After the signal is collected, the analog signal is converted to Convert to digital signal ,in is a discrete-time index, Represents the sampled digital audio signal. The sampling process is implemented through the sampling theorem, and its sampling frequency satisfy ,in The sampling frequency is the highest frequency component in the signal to avoid aliasing effects. For example, if a speech signal is collected, its maximum frequency is about 4kHz, then the sampling frequency can be selected kHz. The sampled signal is expressed as: ; in is the sampling period. The collected digital audio data is sampled at a fixed time length Perform segmentation processing to obtain an audio segment sequence. Assume that the number of sampling points in each segment is , the segmented signal is expressed as ,in is a segment index, Indicates This segmentation process ensures that the length of each signal segment is fixed, which is convenient for subsequent frequency domain analysis. In order to extract the frequency characteristics, each audio signal segment is Perform discrete Fourier transform to convert it from time domain signal to frequency domain signal. Frequency domain signal The calculation formula is: ; in represents the frequency index, It is Segment frequency The complex amplitude on contains amplitude and phase information. Extract the duration of periodic signals. Extract the periodic features of audio signals by detecting the duration of stable components in the frequency domain signal. Define the duration of periodic signals as , which is calculated based on how fast the amplitude of the frequency components changes: ; in Indicates the change in spectrum amplitude. If the amplitude of a frequency component remains stable in multiple consecutive segments, then The extracted audio jitter data is input into the Fourier high-order synchronous compression (FSSTN) processing unit for feature decomposition. FSSTN is a high-order time-frequency analysis method that can accurately represent the instantaneous frequency of the signal. Assume that the time-frequency representation of the signal is , its instantaneous frequency Calculated by the following formula: ; in represents the time derivative of the time-frequency distribution, Represents the imaginary part operation. Through FSSTN decomposition, the frequency feature sequence is obtained , which contains the main frequency components of each signal and its changing rules. Perform phase feature extraction to obtain a frequency reference signal. Phase feature extraction determines the reference frequency by analyzing the phase change of the frequency component. : ; in is the phase difference between adjacent moments, is the time interval. Based on the frequency reference signal, the frequency change is obtained by time domain differential operation. : ; The frequency variation reflects the dynamic change characteristics of the signal frequency over time.

[0022] In one example, a K-SVD algorithm is trained on an audio frequency domain signal to obtain a spectrum feature vector, and resonance peak detection is performed on the spectrum feature vector based on a frequency reference signal and a frequency variation to construct compressed mask data, including: Divide the audio frequency domain signal into sub-bands according to a preset frequency interval to obtain frequency band grouping data, and sparsely represent the frequency band grouping data to obtain a feature basic matrix; The characteristic basic matrix is ​​optimized through the K-SVD iterative algorithm to obtain the spectrum characteristic vector; Perform singular value decomposition on the spectrum feature vector based on the frequency reference signal and the frequency variation to obtain the basic characteristics of the spectrum; Performing resonance peak extraction operation on basic features of the spectrum to obtain spectrum peak position data, and performing spectrum quality evaluation on the spectrum peak position data to obtain a spectrum quality threshold; The spectrum feature vector is binarized according to the spectrum quality threshold to obtain a spectrum label matrix, and the spectrum label matrix is ​​matrix-operated with the spectrum feature vector to obtain compressed mask data.

[0023] In this example, the frequency range is set according to the preset frequency range. Sub-band division is performed. These frequency intervals are determined according to the audio application scenario. For example, speech signals are usually divided into low frequency band, medium frequency band and high frequency band. The result of sub-band division is expressed as ,in Indicates frequency bands, Contains frequency domain data within the corresponding frequency range. By dividing, the complex frequency domain signal is decomposed into multiple frequency band grouping data, which is convenient for subsequent sparse processing. , and the sparse representation method is used to extract its main features. The goal of sparse representation is to approximate the original signal with as few basis vectors as possible. Assume that the signal In sparse basis The sparse representation of , the relationship is: ; in is a sparse basis matrix, is a sparse coefficient vector, Is the error. By optimizing the problem: ; Find the sparsest ,in is a sparse vector norm, which indicates the number of non-zero elements, is the permissible error range. By grouping all frequency bands Perform sparse representation to obtain the feature basic matrix , each column of which corresponds to a sparse representation of a subband. The K-SVD iterative algorithm is used to iterate the feature matrix K-SVD is a dictionary learning algorithm that continuously updates the sparse basis and the sparse coefficient , to minimize the reconstruction error. The core step is to alternately optimize: fix ,optimization ;fixed ,renew The final optimized sparse coefficient matrix is Based on the optimized spectrum feature vector, combined with the frequency reference signal and frequency variation ,right Perform singular value decomposition. Singular value decomposition will This breaks down into: ; in is the left singular vector matrix, indicating the direction of the feature; is a diagonal matrix of singular values, representing the weights of the features; is a right singular vector matrix, representing the time distribution of features. Singular value matrix The first few non-zero values ​​of represent the main spectrum basic features of the signal. Perform resonance peak extraction on the spectrum basic features. The resonance peak is the local maximum value in the spectrum, reflecting the strongest harmonic component in the signal. Assume that the spectrum basic features are , the position of the resonance peak is determined by the following conditions: ; Extracted formant position data Indicates the key frequency components in the signal. Perform spectral quality assessment on the resonance peak position data to determine the ratio of useful components and noise in the signal. Define spectral quality threshold for: ; The numerator is the total energy of the resonance peak, and the denominator is the total energy of the entire spectrum. If the spectrum quality is higher than the set threshold, it is considered that the spectrum quality is high, otherwise further processing is required. , and binarize the spectral feature vector. The binarization operation is performed by setting the threshold , classifying the elements of the feature vector into high quality (1) and low quality (0): ; Generate spectrum label matrix after binarization , used to identify which frequency components are important. Perform matrix operations with the spectrum feature vector S to obtain the compressed mask data C: ; in Represents element-by-element multiplication. Compressed mask data Contains optimized and filtered spectral information, with a significantly reduced size while still retaining the main characteristics of the signal.

[0024] In one example, the spectrum feature vector is input into a variational autoencoder for nonlinear dimensionality reduction to obtain feature encoding data, and complexity calculation is performed based on the feature encoding data to obtain compression control parameters, including: Perform normalization operation on the spectral feature vector to obtain standard feature data, and input the standard feature data into the encoding network of the variational autoencoder for nonlinear mapping to obtain a latent variable matrix; Reparameterization sampling is performed on the latent variable matrix to obtain feature encoding data, and sequence segmentation is performed on the feature encoding data to obtain feature sequence data; The feature sequence data is input into the CBOW model for vector training to obtain an audio semantic model, and the feature density of the audio semantic model is calculated to obtain a complexity matrix; Based on the complexity matrix, hierarchical interval division is performed to obtain complexity level data, and the number of quantization bits, compression ratio, and fidelity parameters are set according to the complexity level data to obtain compression control parameters.

[0025] In this example, assume that the spectral feature vector is ,in Indicates In order to eliminate the magnitude differences between different characteristic components, Perform normalization operation. The normalization formula is: ; in is the mean of the eigenvectors, is the standard deviation of the eigenvector. The normalized standard eigendata The mean is 0 and the standard deviation is 1. Input the encoding network of the variational autoencoder (VAE), which compresses the high-dimensional feature vector into a low-dimensional latent variable matrix through nonlinear mapping. Assume that the encoder network of VAE is represented by the function , then the latent variable matrix It is expressed as: ; in and is the network weight matrix, and is the bias vector, is the activation function. Through nonlinear mapping, the latent variable matrix The main information in the data is extracted and the dimension is reduced. Perform reparameterized sampling, VAE generates the mean of each latent variable during encoding and variance , and the sampling process is: ; in is a random variable with a standard normal distribution. The reparameterization technique ensures the differentiability of the sampling process, thus supporting end-to-end training of the network. The feature encoding data obtained by sampling Represents the low-dimensional sparse features of the audio signal. Feature encoding data Segment the sequence, assuming that the length of each segment is , then the segmentation result is : ; in Indicates segment index. Feature sequence data is a time series, each segment contains a set of low-dimensional encoding features. The segmented feature sequence data is input into the continuous bag of words model (CBOW). The CBOW model is a context-based vector training method, whose goal is to predict the central feature vector through the context window. Assume that the context window is , then the loss function of CBOW is: ; in is the prediction probability, which is calculated by the softmax function of the network output. The audio semantic model generated after CBOW training can capture the contextual relationship of the feature sequence and generate a semantic feature representation. The feature density of the audio semantic model is calculated to obtain the complexity matrix The complexity calculation is achieved by analyzing the distribution density of the characteristic components, and the formula is: ; in It is a specific dimension in the semantic model The number of co-activations, is the total number of time steps. The complexity matrix The value of reflects the complexity of the signal distribution in the feature space. Based on the complexity matrix , and divide it into graded intervals. The division rules are based on the complexity value, which is divided into multiple levels, for example: ; in is the lower bound of the interval, is the upper bound. The complexity level data provides an accurate description of the characteristic distribution of each interval. According to the complexity level data, the number of quantization bits is set , compression ratio and fidelity parameters For example, for intervals with lower complexity, a lower quantization bit number and a higher compression ratio are selected, while for intervals with higher complexity, a higher quantization bit number and a lower compression ratio are selected. The final expression of the compression control parameter is: ; in is the total number of complexity classes.

[0026] In one example, an orthogonal matching pursuit operation is performed based on the compression control parameter and the compression mask data to obtain a standard compression matrix, including: The compression control parameters are input into the compression domain mapping unit for transformation to obtain an initial measurement matrix, and sparsity analysis is performed on the initial measurement matrix to obtain an optimal basis function; The optimal basis function and the compressed mask data are input into the orthogonal matching pursuit unit for iteration to obtain the compressed basis matrix; Performing Gram-Schmidt orthogonalization operation on the compressed basis matrix to obtain orthogonal basis data, and performing linear transformation on the orthogonal basis data to obtain reversible compressed data; Calculate the condition number of the reversible compression data to obtain the optimization parameters, and perform matrix adjustment on the optimization parameters and the reversible compression data to obtain an adaptive matrix; The time-frequency parameters of the adaptive matrix are optimized to obtain a standard compressed matrix.

[0027] In this example, the compression control parameter Input compression domain mapping unit, compression control parameters include quantization bit , compression ratio and fidelity parameters The mapping unit generates the initial measurement matrix based on these parameters , whose dimension is determined by the signal compression ratio If the original signal length is And the target compression ratio is ,but The dimension is ,in The elements of the initial measurement matrix are randomly generated from a Gaussian distribution to ensure that the signal has good sparsity after projection. Perform sparsity analysis to evaluate its ability to represent sparse signals. Assume the input signal is , and the projection result is , then the goal of analyzing sparsity is to find the optimal basis function , so that the sparse representation s after projection is the sparsest, that is, it satisfies: ; Obtained through optimization This is the optimal basis function, which can minimize the number of non-zero elements while maintaining the reconstruction accuracy. and compression mask data Input the orthogonal matching pursuit (OMP) unit for iteration. OMP is a greedy algorithm that selects the basis function that best explains the residual in each iteration, updates the sparse representation s and the residual . Let the initial residual be , No. The steps of the iteration are: Select basis function: ; in express No. Column. Update the sparse coefficient: ; in is the selected basis function set. Update the residual: ; Through multiple iterations, the compressed basis matrix is ​​obtained . For the compressed basis matrix Perform the Gram-Schmidt orthogonalization operation to generate orthogonal basis data. The goal of orthogonalization is to convert the column vectors of the matrix into a set of mutually orthogonal unit vectors. The specific operation is: ; in are the original basis vectors, is the orthogonalized basis vector. The result of orthogonalization is a set of orthogonal basis data , whose column vectors are linearly independent and normalized. Perform linear transformation to generate reversible compressed data The linear transformation uses diagonal weighting: ; in Is a diagonal matrix used to adjust the amplitude of the basis vector to ensure the reversibility and numerical stability of the matrix. Perform condition number calculation to evaluate the stability of the matrix. The condition number is defined as: ; in Represents the norm of the matrix. If the condition number is high, the matrix is ​​more sensitive to the input, resulting in numerical instability. Generate optimization parameters through condition number calculation , adjust the matrix to reduce the condition number, and obtain the adaptive matrix : ; in is the identity matrix, is the adjustment coefficient. For the adaptive matrix Perform time-frequency parameter optimization to adjust the distribution of the matrix based on the time and frequency domain characteristics of the signal. For example, analyze the time-frequency distribution of the signal through short-time Fourier transform: ; in is a window function. Optimization of distribution characteristics , generating a standard compressed matrix .

[0028] In one example, a standard compression matrix and original audio data are subjected to matrix transformation and CP-ABE encryption to obtain an encrypted audio stream, and the encrypted audio stream is stored in a memory in a voice recorder, including: Perform matrix multiplication operation on the original audio data and the standard compression matrix to obtain matrix compression data, and round the matrix compression data according to the quantization step size to obtain initial compression data; The initial compressed data is converted into fixed-point numbers according to the preset quantization bit number to obtain quantized data, and the CP-ABE access strategy is constructed for the quantized data to obtain ciphertext strategy data; Performing encryption operation on the ciphertext strategy data and the quantized data to obtain an encrypted audio stream, and marking and encapsulating the compression information and key information in the encrypted audio stream to obtain file identification data; The file identification data and the encrypted audio stream are format-converted to obtain storage data, and a write operation is performed on the storage data to obtain encrypted compressed data in the storage device of the recording pen.

[0029] In this example, the raw audio data is represented as a matrix , where each column Represents a segment of an audio signal, with the number of rows corresponding to the sampling points of the time step. Standard compression matrix is a sparse matrix whose column vectors are the compressed basis and whose dimensions are ,in represents the compressed dimension, Represents the dimensions of the original data. Through matrix multiplication operation , get matrix compression data . Each column of is a compressed data vector, which represents the low-dimensional projection of the original audio signal on the compressed basis. According to the quantization step The rounding operation is performed to reduce the precision of the data, thereby compressing the data volume. The quantization process is expressed as: ; in yes The elements in , round(·) indicates the rounding operation, is the quantized value. The rounded result is the initial compressed data , where each element is an integer multiple of the quantization step, thus significantly reducing the accuracy requirements for stored data. After that, according to the preset quantization bit Perform fixed-point conversion on it. The goal of fixed-point conversion is to map quantized data to a limited numerical range, such as using - bits represent integer range The formula for fixed-point number conversion is: ; in Indicates that Restricted to the interval Inside, is the value after fixed-point conversion. and The matrices are The result of this process is quantitative data , whose value is limited to the range of discrete integers. Combined with CP-ABE (attribute-based encryption) technology, access policy is constructed for quantitative data to generate ciphertext policy data. CP-ABE encryption allows setting access rights based on user attributes, such as setting a user attribute set and access policies , only satisfied The ciphertext policy data is encrypted by the encryption key and access policies Together, it is expressed as: ; Among them, Enc Indicates encryption operation. The ciphertext policy data and the quantized data Combined, encryption operation is performed to generate an encrypted audio stream The encryption process formula is: ; in Represents a bitwise XOR operation, used to combine quantized data and the key , and EncData(·) is the application of encryption algorithms such as AES. Encrypted audio stream Data privacy is protected, and even if it is intercepted, the original data cannot be directly restored. To facilitate storage and management, the compression information and key information in the encrypted audio stream are marked and encapsulated. The mark encapsulation contains metadata of the data structure, such as the quantization step size. , quantization bit number , matrix dimension and , and the summary hash (CipherPolicy) of the ciphertext policy data. The encapsulation result is the file identification data , whose contents are usually organized as a file header. and encrypted audio stream Perform format conversion to generate storage data The format conversion operation includes data alignment, adding checksum information, and adjusting to a file format that meets the requirements of the recording pen storage system, such as a binary stream or a file block structure. The generated storage data is written into the recording pen's memory, and the encrypted compressed data stored in the memory is .

[0030] In one example, the audio compression storage method of the recording pen further includes: Read the file identification data from the memory to extract information, obtain encryption parameters and decryption keys, and input the decryption keys into the CP-ABE decryption unit to perform decryption operations to obtain decrypted audio data; The decrypted audio data is quantized and denormalized to obtain decompressed data, and the standard compression matrix is ​​reconstructed based on the encryption parameters to obtain a restoration matrix; Performing an inverse transformation operation on the decompressed data and the restoration matrix to obtain reconstructed audio data, and performing a quality assessment on the reconstructed audio data within a preset fidelity range to obtain qualified audio data; The qualified audio data is converted into digital-to-analog format to obtain a playback signal, and the playback signal is power amplified and sound output is performed to obtain sound data of the audio playback unit.

[0031] In this example, the file identification data is read from the memory , the file identification data contains meta-information used for decryption and reconstruction, such as encryption parameters and the decryption key , and the quantization step size , quantization bit number , compress matrix dimensions and By parsing , extract the necessary information. Input the CP-ABE decryption unit and perform decryption operations in combination with the preset access strategy. Assume that the encrypted audio stream is , then the decrypted audio data It is expressed as: ; Where DecData(·) is the corresponding decryption algorithm operation, such as the AES decryption function. The decryption process ensures that only users who meet the permission requirements can restore the audio data. is the quantized data, and the quantization denormalization operation is performed to restore the decompressed data. Assume that the decompressed data matrix is , the formula for quantized denormalization is: ; in and Is the file identification data The minimum and maximum values ​​of the audio data stored in is the number of quantization bits. The denormalization process restores the quantized discrete data to an approximate continuous value range to obtain preliminary decompressed data. . Based on encryption parameters in the file identification data , reconstruct the standard compression matrix to obtain the restored matrix The standard compression matrix is ​​encrypted or compressed during storage to reduce storage usage. Its reconstruction formula is: ; where Reconstruct(·) is an algorithmic operation used to reconstruct the compressed matrix based on the parameters, for example, regenerating from a sparse basis and a random projection matrix The matrix has the same compressed basis properties as the original standard compressed matrix. With the reduction matrix Perform an inverse transform operation to restore the original audio data. The mathematical formula for the inverse transform is: ; in yes The pseudo-inverse matrix is ​​calculated by the following formula: ; The function of the pseudo-inverse matrix is ​​to map the compressed data in the low-dimensional space back to the high-dimensional space and reconstruct the audio data. In order to ensure the quality of the reconstructed audio data, a quality assessment is performed within a preset fidelity range. The quality assessment indicators include signal-to-noise ratio (SNR), waveform similarity (such as peak signal-to-noise ratio PSNR), and spectrum error. The fidelity assessment function is defined as: ; in is the original audio data. If the fidelity evaluation value is lower than the set threshold, the data is considered qualified and marked as qualified audio data. For qualified audio data, it is converted from digital form to analog signal for playback. The digital audio data is converted to analog signal through a digital-to-analog converter. Converted into a continuous analog signal , the formula is: ; in is the sampling interval, It is a standard synchronization function used to reconstruct a continuous signal. The playback signal after digital-to-analog conversion The signal is then boosted by a power amplifier to drive the speaker in the audio playback unit. The speaker converts the amplified analog signal into air vibrations, generating audible sound.

[0032] Reference Figure 2 This embodiment provides a recording pen audio compression storage device, including: Preprocessing module 1 is used to preprocess the original audio data to obtain an audio frequency domain signal, perform high-order synchronous compression FSSTN operation on the audio frequency domain signal, and obtain a frequency reference signal and a frequency variation; Detection module 2 is used to perform K-SVD algorithm training on the audio frequency domain signal to obtain a spectrum feature vector, and perform resonance peak detection on the spectrum feature vector based on the frequency reference signal and the frequency variation to construct compressed mask data; The encoding module 3 is used to input the spectrum feature vector into the variational autoencoder for nonlinear dimensionality reduction to obtain feature encoding data, and perform complexity calculation based on the feature encoding data to obtain compression control parameters; An operation module 4 is used to perform an orthogonal matching pursuit operation based on the compression control parameter and the compression mask data to obtain a standard compression matrix; The encryption module 5 is used to perform matrix transformation and CP-ABE encryption on the standard compression matrix and the original audio data to obtain an encrypted audio stream, and store the encrypted audio stream into the memory in the voice recorder.

[0033] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0034] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0035] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0036] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0037] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0038] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0039] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for compressing and storing audio in a recording pen, characterized in that: The following steps are involved: Preprocessing the original audio data to obtain an audio frequency domain signal, performing a high-order synchronous compression FSSTN operation on the audio frequency domain signal to obtain a frequency reference signal and a frequency variation; Performing K-SVD algorithm training on the audio frequency domain signal to obtain a spectrum feature vector, and performing resonance peak detection on the spectrum feature vector based on the frequency reference signal and the frequency variation to construct compression mask data; Inputting the spectrum feature vector into a variational autoencoder for nonlinear dimensionality reduction to obtain feature encoding data, and performing complexity calculation based on the feature encoding data to obtain compression control parameters; Performing an orthogonal matching pursuit operation based on the compression control parameter and the compression mask data to obtain a standard compression matrix; The standard compression matrix and the original audio data are subjected to matrix transformation and CP-ABE encryption to obtain an encrypted audio stream, and the encrypted audio stream is stored in a memory in the voice recorder.

2. The audio compression storage method of a voice recorder according to claim 1, characterized in that: The preprocessing of the original audio data to obtain an audio frequency domain signal, and performing a high-order synchronous compression FSSTN operation on the audio frequency domain signal to obtain a frequency reference signal and a frequency variation include: The external sound signal is collected by the audio collection unit in the recording pen to obtain original audio data, and the original audio data is converted into digital quantity to obtain digital audio data; Segmenting the digital audio data according to a fixed time length to obtain an audio segment sequence, and performing discrete Fourier transform on the audio segment sequence to obtain an audio frequency domain signal; Extracting the duration of a periodic signal from the audio frequency domain signal to obtain audio jitter data, and inputting the audio jitter data into a Fourier high-order synchronous compression FSSTN processing unit for feature decomposition to obtain a frequency feature sequence; Phase feature extraction is performed on the frequency feature sequence to obtain a frequency reference signal, and a time domain differential operation is performed on the frequency reference signal to obtain a frequency variation.

3. The audio compression storage method of a voice recorder according to claim 2, characterized in that: The step of performing K-SVD algorithm training on the audio frequency domain signal to obtain a spectrum feature vector, and performing resonance peak detection on the spectrum feature vector based on the frequency reference signal and the frequency variation to construct compressed mask data includes: Dividing the audio frequency domain signal into sub-bands according to a preset frequency interval to obtain frequency band grouping data, and sparsely representing the frequency band grouping data to obtain a feature basic matrix; The characteristic basic matrix is ​​optimized by a K-SVD iterative algorithm to obtain a frequency spectrum characteristic vector; Performing singular value decomposition on the frequency spectrum feature vector based on the frequency reference signal and the frequency variation to obtain basic frequency spectrum features; Performing a resonance peak extraction operation on the basic features of the spectrum to obtain spectrum peak position data, and performing spectrum quality evaluation on the spectrum peak position data to obtain a spectrum quality threshold; The spectrum feature vector is binarized according to the spectrum quality threshold to obtain a spectrum label matrix, and a matrix operation is performed on the spectrum label matrix and the spectrum feature vector to obtain compressed mask data.

4. The audio compression storage method of a voice recorder according to claim 3, characterized in that: The step of inputting the spectral feature vector into a variational autoencoder for nonlinear dimensionality reduction to obtain feature coding data, and performing complexity calculation based on the feature coding data to obtain compression control parameters includes: Performing a normalization operation on the spectral feature vector to obtain standard feature data, and inputting the standard feature data into the encoding network of the variational autoencoder for nonlinear mapping to obtain a latent variable matrix; Reparameterizing and sampling the latent variable matrix to obtain feature coding data, and performing sequence segmentation on the feature coding data to obtain feature sequence data; Inputting the feature sequence data into the CBOW model for vector training to obtain an audio semantic model, and performing feature density calculation on the audio semantic model to obtain a complexity matrix; Based on the complexity matrix, hierarchical interval division is performed to obtain complexity level data, and the quantization bit number, compression ratio, and fidelity parameters are set according to the complexity level data to obtain compression control parameters.

5. The audio compression storage method of a voice recorder according to claim 4, characterized in that: The performing an orthogonal matching pursuit operation based on the compression control parameter and the compression mask data to obtain a standard compression matrix includes: Inputting the compression control parameters into a compression domain mapping unit for transformation to obtain an initial measurement matrix, and performing sparsity analysis on the initial measurement matrix to obtain an optimal basis function; Inputting the optimal basis function and the compressed mask data into an orthogonal matching pursuit unit for iteration to obtain a compressed basis matrix; Performing a Gram-Schmidt orthogonalization operation on the compressed basis matrix to obtain orthogonal basis data, and performing a linear transformation on the orthogonal basis data to obtain reversible compressed data; Performing condition number calculation on the reversible compression data to obtain optimization parameters, and performing matrix adjustment on the optimization parameters and the reversible compression data to obtain an adaptive matrix; Time-frequency parameter optimization is performed on the adaptive matrix to obtain a standard compression matrix.

6. The audio compression storage method of a voice recorder according to claim 5, characterized in that: The method of performing matrix transformation and CP-ABE encryption on the standard compression matrix and the original audio data to obtain an encrypted audio stream, and storing the encrypted audio stream in a memory in the voice recorder, comprises: Performing a matrix multiplication operation on the original audio data and the standard compression matrix to obtain matrix compression data, and rounding the matrix compression data according to a quantization step size to obtain initial compression data; The initial compressed data is converted into fixed-point numbers according to a preset number of quantization bits to obtain quantized data, and a CP-ABE access strategy is constructed for the quantized data to obtain ciphertext strategy data; Performing encryption operation on the ciphertext strategy data and the quantized data to obtain an encrypted audio stream, and marking and encapsulating the compression information and key information in the encrypted audio stream to obtain file identification data; The file identification data and the encrypted audio stream are format-converted to obtain storage data, and a write operation is performed on the storage data to obtain encrypted compressed data in the memory of the voice recorder.

7. The audio compression storage method of a voice recorder according to claim 6, characterized in that: The recording pen audio compression storage method also includes: Read the file identification data from the memory to extract information, obtain encryption parameters and decryption keys, and input the decryption keys into the CP-ABE decryption unit to perform decryption operations to obtain decrypted audio data; quantizing and denormalizing the decrypted audio data to obtain decompressed data, and reconstructing the standard compression matrix based on the encryption parameters to obtain a restoration matrix; Performing an inverse transformation operation on the decompressed data and the restoration matrix to obtain reconstructed audio data, and performing a quality assessment on the reconstructed audio data within a preset fidelity range to obtain qualified audio data; The qualified audio data is converted into digital-to-analog format to obtain a playback signal, and the playback signal is power amplified and sound output is performed on the playback signal to obtain sound data of the audio playback unit.

8. An audio compression storage device for a recording pen, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the device comprises: A preprocessing module is used to preprocess the original audio data to obtain an audio frequency domain signal, and perform a high-order synchronous compression FSSTN operation on the audio frequency domain signal to obtain a frequency reference signal and a frequency variation; A detection module, used for performing K-SVD algorithm training on the audio frequency domain signal to obtain a spectrum feature vector, and performing resonance peak detection on the spectrum feature vector based on the frequency reference signal and the frequency variation to construct compression mask data; An encoding module, used for inputting the spectrum feature vector into a variational autoencoder for nonlinear dimensionality reduction to obtain feature encoding data, and performing complexity calculation based on the feature encoding data to obtain compression control parameters; A calculation module, used for performing an orthogonal matching pursuit operation based on the compression control parameter and the compression mask data to obtain a standard compression matrix; The encryption module is used to perform matrix transformation and CP-ABE encryption on the standard compression matrix and the original audio data to obtain an encrypted audio stream, and store the encrypted audio stream into a memory in the voice recorder.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

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

Images