Sound transmission system based on encryption and decryption

Abstract

The invention relates to the technical field, in particular to a sound transmission system based on encryption and decryption, which comprises a sound signal acquisition module, a sensitive weight calculation module, an encryption density execution module, a fingerprint embedding transmission module and a decryption verification module. The invention specifically relates to the technical field of password communication, a voiceprint feature set is established in a voice signal and a dynamic encryption parameter is formed in combination with a semantic weight, framing adaptive encryption strength control of voice content can be realized, the encryption process is kept balanced between fidelity and security, and the encryption efficiency is improved. A density regulation and control mechanism enables a highly sensitive voice area to obtain higher disturbance intensity, so that anti-eavesdropping and tamper-proofing capabilities are improved, a traceable identifier is formed in encrypted voice in a frequency spectrum fingerprint embedding mode, uniqueness identification and traceability verification of data transmission are realized, and through dual verification of disturbance indexes and frequency spectrum features in a decryption stage, a decryption effect is improved. The accuracy of data integrity and consistency judgment is improved, and the tracing reliability and the anti-interference robustness are integrally enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of password communication, in particular to a sound transmission system based on encryption and decryption. BACKGROUND

[0002] The technical field of password communication includes methods for ensuring the security and privacy of information during transmission and storage through encryption technology, the core of which is to study and apply various encryption algorithms, key management protocols and security protocols to prevent unauthorized access, eavesdropping, tampering and forgery. This field systematically introduces the basic theories from information theory, number theory, computational complexity theory to symmetric encryption, asymmetric encryption, hash function, digital signature, identity authentication, secure multi-party computation, etc., aiming to build a secure and reliable communication infrastructure.

[0003] Among them, a sound transmission system based on encryption and decryption refers to a technology that encrypts sound data and transmits it through a secure channel, and the receiving end decrypts the received encrypted sound data to restore the original sound. Through targeted technical matters, it covers the collection of sound data, the selection and implementation of encryption algorithms, the packaging of encrypted data and transmission protocols, and the implementation of decryption algorithms at the receiving end. Specifically, after collecting sound data, the sound data block is encrypted using the AES algorithm, and the encrypted data is transmitted through the TCP / IP protocol. At the receiving end, the received encrypted data stream is decrypted using the AES algorithm according to the pre-set key, and the decrypted data is restored to playable sound.

[0004] The prior art uses a fixed encryption algorithm to perform uniform processing on sound data blocks, lacking adaptive adjustment capability under different voice content and signal feature conditions, resulting in mismatch between encryption strength and content sensitivity of voice information. When the voice signal energy changes greatly or the semantic distribution is uneven, some information may be over-encrypted, causing transmission delay, or insufficient encryption, leading to reduced security. At the same time, data is decrypted only by the key, lacking a verifiable data traceability mechanism. If disturbances or tampering occur during transmission, the system cannot effectively detect them through feature comparison, affecting the security and controllability of voice communication. SUMMARY

[0005] The purpose of the present application is to solve the shortcomings in the prior art and to provide a sound transmission system based on encryption and decryption.

[0006] In order to achieve the above object, the present application adopts the following technical scheme: a voice transmission system based on encryption and decryption, the system comprises, obtaining a voice signal, extracting amplitude sampling points, frequency band energy values, fundamental frequency variation rates and resonance peak spacings by frame, performing ratio calculation and weighted accumulation, recording amplitude sampling points as encryption reference, generating encryption voiceprint reference data; A sensitive weight calculation module: based on the encryption voiceprint reference data, extracting phoneme duration, semantic symbol frequency and energy average, performing ratio calculation and fusion, weighting the results and the voiceprint confidence, and generating encryption control sensitive parameters; An encryption density execution module: comparing the encryption control sensitive parameters with the density threshold value, when the threshold value is exceeded, calling the sampling points and the key offset sequence to perform reordering and offset, when not exceeding, calling the key sampling points to perform disturbance, and generating encryption voice frame structure information; A fingerprint embedding transmission module: based on the encryption voice frame structure information, decomposing the frequency spectrum and partitioning, performing energy density and phase offset difference value calculation, embedding a pseudo-random sequence to form a frequency spectrum fingerprint, and generating fingerprint mark transmission data; A decryption verification and checking module: based on the fingerprint mark transmission data, calling the disturbance index and the key mapping table to perform reverse ordering and amplitude compensation, restoring the decrypted voice, extracting the frequency spectrum data and comparing the frequency spectrum fingerprint, and generating a decryption consistency checking result.

[0007] As a further scheme of the present application, the encryption voiceprint reference data comprises a voiceprint numerical feature set, amplitude sampling point reference information and encryption reference parameters, the encryption control sensitive parameters specifically are semantic and voiceprint fusion sensitive coefficients, encryption intensity control factors and semantic weight correction terms, the encryption voice frame structure information comprises disturbance index records, key mapping additional tables and synchronization mark fields, the fingerprint mark transmission data specifically are frequency spectrum fingerprint structures, pseudo-random sequence embedding information and encryption voice data packages, and the decrypted voice signal comprises a reset sample point sequence, an amplitude compensation result and a decrypted voice restored waveform.

[0008] As a further scheme of the present application, an input voice signal is obtained, the signal is frame-extracted to obtain an acoustic amplitude sampling point sequence according to a time interval, the amplitude change amplitude between adjacent sampling points in each frame is detected and the sampling frame sequence length is recorded, the average amplitude and the change rate of each frame amplitude sequence are calculated to form an amplitude statistical set, and an acoustic amplitude characteristic value is generated; An energy frequency calculation submodule: based on the acoustic amplitude characteristic value, extracting the frequency band energy value and the fundamental frequency variation rate of the corresponding frame, calculating the ratio of the frequency band energy value and the fundamental frequency variation rate, calling the ratio sequence between adjacent frames to perform difference and accumulation to form a ratio change sequence, and generating an energy-frequency ratio coefficient; The voiceprint feature generation sub-module: according to the energy-frequency ratio coefficient, a frame-to-frame resonance peak spacing sequence is called, a ratio change sequence and a resonance peak spacing change amplitude are calculated, a weighted value is calculated, a frame-level weighted cumulative operation is performed to form a feature vector set, and the voice amplitude sampling point sequence is taken as an encryption reference to generate encrypted voiceprint reference data.

[0009] As a further scheme of the present application, the sensitive weight calculation module further comprises: The phoneme duration analysis sub-module: the duration of each phoneme sequence in the encrypted voiceprint reference data is obtained, the start and end time points corresponding to the phoneme boundary are detected and the time interval is calculated, the time intervals of all phonemes are summed and grouped to obtain the average duration interval value of the phoneme sequence under different speech segments, and the phoneme duration average value is generated; The semantic frequency ratio sub-module: based on the semantic symbol sequence marked in the encrypted voiceprint reference data, the number of occurrences of each semantic symbol in the speech sample is counted and the occurrence frequency is calculated, the ratio operation is performed according to the phoneme duration average value and the occurrence frequency of each semantic symbol, the relative duration proportion of the corresponding semantic symbol at the phoneme level is obtained, and the semantic ratio parameter is generated; The acoustic energy fusion calculation sub-module: the acoustic energy sequence in the encrypted voiceprint reference data is called, the energy amplitude average value is calculated for each time slice, the fusion calculation is performed according to the semantic ratio parameter and the acoustic energy average value, the weighted superposition of the two data under the same time slice is calculated, and the integrated energy weight value is generated; The linear weighting control sub-module: according to the voiceprint confidence value output in the voiceprint recognition stage, the integrated energy weight value is called, the linear weighting operation is performed on the two, the proportion contribution of the weight coefficient and the confidence is calculated, the fusion parameter for controlling the encryption intensity is obtained, and the semantic and voiceprint fusion sensitive coefficient is generated.

[0010] As a further scheme of the present application, the encryption density execution module further comprises: The threshold judgment sub-module: the encryption control sensitive parameter and the encryption density threshold value are obtained, the numerical comparison operation is performed on the two, when the encryption control sensitive parameter exceeds the encryption density threshold value, the amplitude sampling point number sequence of the corresponding frame is extracted, and the threshold offset proportion value is calculated, the offset proportion value is differentially compared with the key offset sequence length, and the density trigger coefficient is generated; The time axis rearrangement sub-module: based on the density trigger coefficient, the amplitude sampling point sequence and the key offset sequence length are called, the time axis rearrangement operation is performed on the same time period index, the interval difference value after rearrangement is recorded, and the time index moving rate interval is calculated, and the time rearrangement rate interval value is obtained; Amplitude offset calculation sub-module: according to the time rearrangement rate interval value, the corresponding value of the voiceprint feature matrix and the amplitude data are called, the multiplication and addition offset operation is executed, the product sum of each voiceprint feature matrix value and the amplitude sampling point amplitude is calculated, and the disturbance energy change rate of each frame is obtained, and the amplitude disturbance change rate value is generated; Encryption frame generation sub-module: call the amplitude disturbance change rate value and the key sampling point amplitude parameter, execute the limited amplitude disturbance for the sample whose amplitude disturbance change rate value is lower than the density threshold interval, and keep the phase fixed, combine the disturbance index value and the key mapping table number sequence to form the frame structure additional information, and generate the encrypted speech frame structure information.

[0011] As a further scheme of the application, the fingerprint embedding transmission module comprises: Spectrum decomposition sub-module: obtain the encrypted speech signal, perform spectrum decomposition on the encrypted speech frame structure information of the speech signal, divide the frequency domain signal into fixed width sections, detect the energy density value and the phase offset rate value of each section, calculate the energy balance difference value between adjacent sections according to the energy density distribution characteristics, and adjust the section distribution boundary according to the phase offset rate change range, to generate a section energy difference value sequence; Phase difference embedding sub-module: call the previous section energy difference value sequence, calculate the phase difference value range for adjacent frequency bands, obtain a pseudo-random sequence, and perform corresponding matching on the pseudo-random sequence and the phase difference value range, execute phase offset adjustment according to the energy difference value and the amplitude ratio of the pseudo-random sequence, generate a phase difference value matrix embedded with pseudo-random information, and obtain the frequency band phase embedding coefficient; Fingerprint fusion sub-module: based on the previous frequency band phase embedding coefficient, select the encrypted speech data frame of the corresponding frequency band, weight and fuse the spectrum amplitude and phase embedding coefficient of the speech frame, calculate the amplitude-phase balance degree value of the fused spectrum, group the weighted fusion result with the original encrypted speech frame data to form a complete transmission data packet, and generate the fingerprint marked transmission data.

[0012] As a further scheme of the application, the decryption verification check module further comprises: Fingerprint mark reverse reset sub-module: obtain the fingerprint mark transmission data, perform reverse sorting operation on the sample index in the data according to the disturbance index table, execute weighted compensation operation on the sample sequence after index rearrangement and the amplitude offset parameter in the key mapping table, rearrange the time sequence position according to the sample amplitude sequence after compensation, restore the time continuous structure of the signal, and generate the time sequence reset signal energy; Spectrum feature extraction sub-module: based on the time sequence reset signal energy, perform spectrum decomposition operation on the signal, obtain the amplitude set and the phase set of each frequency section, calculate the energy density value for each frequency band amplitude set, and calculate the phase offset rate value according to the phase set, call the amplitude set and the phase set for section ratio analysis, and generate the spectrum energy distribution rate; The phase consistency matching sub-module: according to the spectral energy distribution rate, the phase offset rate difference of adjacent frequency bands is extracted, the corresponding offset reference rate in the key mapping table is compared, the difference comparison coefficient is calculated, the verification sequence is formed according to the comparison coefficient of each frequency band, the segment-by-segment consistency matching is performed with the spectral fingerprint structure, and the decryption consistency check result is generated.

[0013] Compared with the prior art, the advantages and positive effects of the present application are that: In the present application, the voiceprint feature set is established in the voice signal, and the dynamic encryption parameter is formed by combining the semantic weight, so that the frame adaptive encryption intensity control of the voice content can be realized, the balance between fidelity and security is maintained in the encryption process, the density control mechanism makes the high-sensitive voice area obtain higher disturbance intensity, thereby improving the anti-eavesdropping and anti-tampering ability, the spectral fingerprint embedding mode forms a traceable identifier in the encrypted voice, realizes the unique identification and traceability verification of data transmission, and the disturbance index and spectral feature double check in the decryption stage improve the accuracy of data integrity and consistency judgment, and the encryption flexibility, traceability reliability and anti-interference robustness of voice transmission are overall enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 The system flowchart of the present application is shown in the figure. Figure 2 The module flowchart of the present application is shown in the figure. DETAILED DESCRIPTION

[0015] The technical solutions in the present application will be described below with reference to the drawings.

[0016] In the embodiments of the present application, the words such as "example", "for example" are used to represent an example, illustration or explanation. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0017] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent. "Of", "corresponding" and "corresponding" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0018] In the embodiments of the present application, sometimes the subscript such as W1 may be written in the form of non-subscript such as W1, and the meanings expressed thereby are consistent when the difference is not emphasized.

[0019] To make the technical problems, technical solutions and advantages to be solved by the present application clearer, specific embodiments will be described in detail below with reference to the drawings.

[0020] Please refer to Figure 1 and Figure 2 The present application provides a technical solution: a voice transmission system based on encryption and decryption, the system comprising: The sound signal acquisition module: acquires the input voice signal, extracts the sound wave amplitude sampling point sequence, the frequency band energy value, the fundamental frequency change rate and the resonance peak spacing, performs ratio operation on the frequency band energy value and the fundamental frequency change rate, performs weighted accumulation on the obtained ratio value and the resonance peak spacing change amplitude, forms the voiceprint numerical characteristic set corresponding to the frame, and records the sound wave amplitude sampling point sequence as the basic encryption reference to generate the encrypted voiceprint reference data; The sensitive weight calculation module: based on the encrypted voiceprint reference data, extracts the phoneme sequence duration, the semantic symbol occurrence frequency and the sound wave energy average value, performs ratio operation on the semantic symbol occurrence frequency and the phoneme sequence duration, and then performs fusion with the sound wave energy average value, performs linear weighting on the obtained sensitive weight value and the voiceprint confidence value to form the semantic and voiceprint fusion sensitive coefficient for controlling the encryption intensity and generating the encryption control sensitive parameter; The encryption density execution module: according to the comparison between the encryption control sensitive parameter and the encryption density threshold value, when the encryption control sensitive parameter exceeds the encryption density threshold value, the amplitude sampling points of the corresponding frame and the key offset sequence length are reordered in the time axis, and the corresponding values of the voiceprint feature matrix and the amplitude are multiplied and added to form high-density disturbance samples, when the encryption control sensitive parameter is lower than the encryption density threshold value, the key sampling points are executed for limited amplitude disturbance and the phase is kept unchanged, the disturbance index and the key mapping table are combined to form frame structure additional information, the disturbance result and the original voice frame are marked synchronously to generate the encrypted voice frame structure information; The fingerprint embedding transmission module: based on the encrypted voice frame structure information, the encrypted voice signal is executed for frequency spectrum decomposition, the frequency domain is divided into fixed width sections, the energy density and the phase offset rate of each section are called to perform phase difference value operation between sections, a pseudo-random sequence is embedded in the phase difference value of adjacent frequency bands to form a frequency spectrum fingerprint structure, the encrypted voice data and the frequency spectrum fingerprint structure are combined to form traceable transmission data packets, and the fingerprint marked transmission data is generated; Decryption verification module: Based on the fingerprint marker transmitted data, the receiving end calls the perturbation index table and key mapping table to perform reverse sorting and amplitude offset compensation, resets the sample point position to the original timing, restores the decrypted voice signal, and simultaneously performs spectral decomposition on the received data to extract the energy density and phase offset rate of each segment. It calls the phase offset rate of adjacent frequency bands and the key mapping table to perform difference comparison, performs consistency matching between the verification sequence and the spectral fingerprint structure, and generates decryption consistency verification result.

[0021] The encrypted voiceprint reference data includes a set of voiceprint numerical features, amplitude sampling point reference information, and encryption reference parameters. The encryption control sensitive parameters specifically include the semantic and voiceprint fusion sensitivity coefficient, encryption strength adjustment factor, and semantic weight correction term. The encrypted speech frame structure information includes a perturbation index record, a key mapping supplementary table, and a synchronization marker field. The fingerprint marker transmission data specifically includes a spectral fingerprint structure, pseudo-random sequence embedding information, and an encrypted speech data encapsulation body. The decrypted speech signal includes a reset sample sequence, amplitude compensation result, and decrypted speech reconstruction waveform.

[0022] The set of numerical features of the voiceprint, the amplitude sampling point reference information, and the encryption reference parameters in the encrypted voiceprint reference data collectively constitute the basic content of the execution process. First, for each voiceprint sample signal in the set of numerical features of the voiceprint, the sampling rate is set to 16kHz, and the speech sample is divided into frames with a length of 25ms each and a frame shift of 10ms. For each frame signal, the energy, zero-crossing rate, fundamental frequency, and 12-dimensional Mel-frequency cepstral coefficients (MFCC) feature values ​​are calculated, and the voiceprint feature matrix V=[v1,v2,…,v] is obtained in sequence. n The matrix V is calculated by normalizing the amplitude of each frame's feature vector (vᵢ). Each feature value xᵢ is then adjusted according to the formula xᵢ′=(xᵢ−μ) / σ, where μ is the sample mean and σ is the standard deviation. When σ < 0.05, resampling is performed to avoid over-compression, resulting in the normalized speaker matrix V′. Next, the amplitude sequence A=[a1,a2,…,a] of each speech sample is selected from the amplitude sampling point reference information. m The data is divided into 50 segments according to time sequence. The mean amplitude of each segment is calculated as αⱼ=Σ(a k ) / nⱼ, where nⱼ is the number of sampling points in this segment, and by comparing two adjacent sets of αⱼ and αⱼ +1 The difference determines the amplitude fluctuation threshold T. a When |αⱼ +1 −αⱼ|>T a The region is marked as a characteristic mutation region; T a Based on the population standard deviation σ a Set to 0.1σ a If σ a =0.25, then T a=0.025. For the encrypted reference parameter part, set the semantics and voiceprint fusion sensitive coefficient η, the encryption strength control factor β, and the semantic weight correction term γ, take η∈[0.4, 0.7] when executing to balance the proportion of semantics and voiceprint features, when the voiceprint feature dimension is more than 30% higher than the semantic vector dimension, η takes the lower limit value 0.4; β∈[1.0, 2.5] is set according to the voiceprint feature change amplitude classification, if the normalized feature variance σᵥ′=0.12, then β=1.5; γ corrects the weight of the semantic feature in the fusion vector, take γ=Σ(w s s ᵢ) / Σ(wᵥᵢ), where w s ᵢ is the semantic feature weight, wᵥᵢ is the voiceprint weight, when the total weight of the semantic part is 3.6 and the voiceprint part is 4.0, γ=0.9. In the execution process, the fusion feature F=η·V′+γ·S is calculated in turn, and the result is multiplied by β to obtain the encrypted input matrix E=β·F, which is used in the subsequent encrypted voiceprint generation process. In the example, if the sample feature mean μ=0.42 and the standard deviation σ=0.15, after normalization xᵢ′=(0.42−0.42) / 0.15=0, the final E matrix corresponding to the encrypted feature output of the first sample is a 0 vector form, which is used as the input of the subsequent module; The sound signal acquisition module comprises: The sound wave sampling sub-module: acquires the input voice signal, extracts the sound wave amplitude sampling point sequence according to the time interval, detects the amplitude change amplitude between adjacent sampling points in each frame and records the sampling frame sequence length, calculates the average amplitude and change rate of each frame amplitude sequence and forms an amplitude statistical set, and generates a sound wave amplitude feature value; Further, in this submodule, first, the input speech signal is obtained, usually from a microphone or other audio acquisition device, which is converted into a digital signal by sampling and then enters the processing flow. Next, according to the set time interval (such as every 10 milliseconds), the continuous speech signal is divided into multiple frames, each frame containing a certain number of sampling points (such as 100 sampling points per frame). For each frame of sampling points, the amplitude change needs to be detected, that is, the amplitude difference between adjacent sampling points in the frame is compared. If the amplitude change between adjacent sampling points exceeds a certain set threshold (such as 0.5), it is considered that the amplitude has changed significantly, and this change is recorded and stored in the amplitude change sequence. Through the processing of multiple frames of data, the length of the sampling sequence of each frame and the amplitude change value of each frame can be obtained. In the calculation process, in order to obtain the amplitude characteristics of each frame, the amplitude of each frame of sampling points needs to be averaged first to calculate the average amplitude of each frame. Next, the change rate of each frame is calculated, that is, the ratio of the amplitude difference between adjacent sampling points to the time interval. Through the accumulation and statistics of these amplitude characteristics of all frames, a feature set of sound wave amplitude is finally formed. For example, if the amplitude sequence of the first frame in a recorded signal is [0.2, 0.3, 0.4, 0.5], the average amplitude of the frame is (0.2+0.3+0.4+0.5) / 4=0.35; if the amplitude changes greatly in the next frame of sampling points, the amplitude change rate of the frame may be found to be 0.1 when calculating the change rate. The sound wave amplitude feature value generated in this process is used for subsequent analysis; Energy frequency calculation submodule: based on the sound wave amplitude feature value, the frequency band energy value and the fundamental frequency change rate of the corresponding frame are extracted, the ratio of the frequency band energy value and the fundamental frequency change rate is calculated, the ratio sequence between adjacent frames is called to difference and accumulate to form a ratio change sequence, and an energy frequency ratio coefficient is generated; In the energy frequency calculation submodule, first, based on the sound wave amplitude eigenvalue generated in the last stage, the energy value of each frame is extracted. The energy value of each frame can be converted to the frequency domain by fast Fourier transform (FFT), and the energy of the spectrum in a certain frequency band (such as low frequency, medium frequency, and high frequency) is calculated. The fundamental frequency change rate refers to the change of the fundamental frequency (the basic frequency of the tone) between consecutive frames. The fundamental frequency is usually calculated using short-time Fourier transform (STFT) or other frequency estimation methods. The change rate of the fundamental frequency can be obtained by comparing the difference between the fundamental frequency of the current frame and the fundamental frequency of the previous frame. For example, if the fundamental frequency of the current frame is 220Hz and the fundamental frequency of the previous frame is 210Hz, then the fundamental frequency change rate is 10Hz / frame. Then, by calculating the ratio of the energy value of the frequency band to the change rate of the fundamental frequency, the energy-frequency ratio of each frame can be obtained. This ratio reflects the proportion of the energy of the frequency band to the change of the fundamental frequency, and further reflects some characteristics of the speech signal. For example, if the energy of a certain frame is 0.5 and the change rate of the fundamental frequency is 5Hz, then the energy-frequency ratio is 0.5 / 5=0.1. Next, the difference between the energy-frequency ratios of consecutive frames is calculated. By difference operation, the change value of the energy-frequency ratio between each pair of adjacent frames is obtained. These difference values are accumulated to form a ratio change sequence. Finally, the energy-frequency ratio change sequence of all frames is combined to generate an energy-frequency ratio coefficient. This coefficient is used to describe the energy distribution characteristics and dynamic characteristics of the fundamental frequency change of the speech signal. The voiceprint feature generation submodule: according to the energy-frequency ratio coefficient, the inter-frame formant spacing sequence is called, the weighted value of the ratio change sequence and the formant spacing change amplitude is calculated, the frame-level weighted accumulation operation is performed to form a feature vector set, and the sound wave amplitude sampling point sequence is taken as an encryption reference to generate encrypted voiceprint reference data. The first step of the voiceprint feature generation submodule is to extract the formant spacing of each frame based on the energy-frequency ratio coefficient. The frequency difference between different formants is usually calculated by analyzing the formant position of the signal. The change of the formant spacing reflects the change of the vocal tract shape, which is closely related to the pronunciation characteristics of the speaker. Next, the weighted value of the ratio change sequence and the formant spacing change amplitude is calculated. The calculation method of the weighted value is usually to set a weight coefficient to give different weight values to different frames. For example, if the ratio change of the current frame is 0.05 and the formant spacing change amplitude is 0.02, and the weight coefficient of the frame is 0.8, then the weighted value of the frame is 0.05 0.02 0.8=0.0008, the feature value of each frame is obtained by weighted operation, then frame-level weighted cumulative operation is performed, the weighted value of each frame is accumulated to form a feature vector set, each vector of the feature vector set represents the voiceprint feature of the corresponding frame, and finally encrypted voiceprint reference data is generated. These reference data can be used for identity verification or other related applications in voiceprint recognition, for example, if the amplitude of the sound wave sampling point sequence changes greatly and the frequency band energy changes significantly in a recording, the generated voiceprint reference data will have strong recognition.

[0023] The sensitive weight calculation module further comprises: The phoneme duration analysis submodule obtains the duration of each phoneme sequence in the encrypted voiceprint reference data, detects the start and end time points corresponding to the phoneme boundary and calculates the time interval, sums up the time intervals of all phonemes and performs grouping average to obtain the average duration interval value of the phoneme sequence under different speech segments, and generates the average phoneme duration value; In the process of obtaining the duration of each phoneme sequence in the encrypted voiceprint reference data, first, the boundary of each phoneme is extracted from the audio file to accurately locate the start and end time of the phoneme, and the boundary of the phoneme is determined by analyzing the energy change or acoustic characteristics (such as frequency, amplitude) of the signal. This process uses audio processing tools such as audio framing technology, short-time Fourier transform (STFT) or linear prediction analysis (LPC) to analyze the audio signal, locate the start and end points of each phoneme, and calculate the time interval. In a simple example, if the audio file contains the word "hello", the start time of the "H" phoneme is determined to be 0.0 seconds and the end time is 0.2 seconds by the audio processing algorithm, and the duration is calculated to be 0.2 seconds. The same method is applied to "e", "l" and "o" phonemes to calculate the duration of each phoneme. Next, the duration of each phoneme is summed up and the average duration value in different speech segments is calculated. Assuming that the phoneme duration in "hello" is 0.2 seconds, 0.15 seconds, 0.1 seconds and 0.1 seconds respectively, the average duration of the phoneme sequence in this speech segment is (0.2+0.15+0.1+0.1) / 4=0.1375 seconds, the average duration of the phoneme is obtained, the phoneme duration reference of the subsequent speech segment is generated, and finally the time interval of the required phoneme sequence and the corresponding average duration interval value are obtained by summing and grouping average.

[0024] The semantic frequency ratio submodule: based on the semantic symbol sequence labeled in the encrypted voiceprint reference data, the number of occurrences of each semantic symbol in the speech sample is counted and the occurrence frequency is calculated, the ratio operation is performed according to the average phoneme duration and the occurrence frequency of each semantic symbol, the relative duration proportion of the corresponding semantic symbol at the phoneme level is obtained, and the semantic ratio parameter is generated. In the process of counting the number of each semantic symbol appearing in the speech sample and calculating its frequency, first, all the annotated semantic symbol sequences are extracted from the speech data. Based on these annotation information, the number of each symbol appearing in the entire sample is counted. For example, assuming that the word "hello" appears in a certain speech sample, and the annotation information shows that it contains the semantic symbol "greet", the frequency of this symbol is 1 time. On this basis, the frequency value of each semantic symbol is obtained by calculating the frequency. If a semantic symbol appears different number of times in multiple speech samples, the average frequency is further calculated by adding up the number of appearances in all samples. For example, assuming that the "greet" symbol appears 5 times in total in multiple samples, and the total number of speech samples is 10, then its frequency is 5 / 10 = 0.5. After obtaining the frequency, the frequency value is compared with the average phoneme duration to calculate the relative duration ratio of the semantic symbol at the phoneme level. This step fuses the phoneme duration and semantic symbol frequency through a weighting algorithm to obtain the duration ratio of each semantic symbol and its corresponding phoneme. Finally, the generated semantic ratio parameter is obtained.

[0025] Sound energy fusion calculation submodule: call the sound wave energy sequence in the encrypted voiceprint reference data, calculate the energy amplitude mean value for each time slice, perform fusion calculation according to the semantic ratio parameter and the sound wave energy mean value, weight and superimpose the two data under the same time slice, and calculate the comprehensive value to generate the fusion energy weight value; In the process of performing sound energy fusion calculation, first, the sound wave energy sequence in the encrypted voiceprint reference data is called to calculate the energy amplitude mean value in each time slice. Assuming that the length of each time slice is 10 ms, the sound wave energy in 10 ms can be obtained by integrating the signal intensity of each time slice, for example, in a certain time slice, the signal intensity is [0.2, 0.3, 0.4, 0.5], and the energy amplitude mean value is calculated as (0.2+0.3+0.4+0.5) / 4=0.35. Then, the two data under the same time slice are superimposed through the weighted calculation between the semantic ratio parameter and the energy amplitude mean value, for example, in a certain time slice, the ratio of the semantic symbol is 0.5, and the energy amplitude mean value is 0.35, then the fusion energy weight value is calculated as 0.5 0.35=0.175, generating the fusion energy weight value of the time slice.

[0026] Linear weighting control submodule: according to the voiceprint confidence value output in the voiceprint recognition stage, call the fusion energy weight value, perform linear weighting operation on the two, calculate the proportional contribution of the weight coefficient and the confidence, obtain the fusion parameter for controlling the encryption intensity, and generate the semantic and voiceprint fusion sensitivity coefficient; In the linear weighting control calculation process, first, the output result of the voiceprint recognition stage is obtained to obtain the voiceprint confidence value, assuming that the confidence of a voiceprint recognition result is 0.8, then the fusion energy weight value obtained in the previous step is called, the weight value here is 0.175, next, the voiceprint confidence and the fusion energy weight value are executed linear weighting operation to obtain the weighted fusion parameter, for example, a simple linear weighting formula is set: fusion parameter = confidence weight value, according to the formula, the value of the fusion parameter is 0.8 0.175 = 0.14, finally, the linear weighting result is used to calculate the fusion sensitivity coefficient, control the encryption strength, and then affect the effect of the encryption process and the confidentiality of the voice sample.

[0027] The encryption density execution module further comprises: The threshold judgment submodule: obtain the encryption control sensitive parameter and the encryption density threshold, perform numerical comparison operation on the two, when the encryption control sensitive parameter exceeds the encryption density threshold, extract the amplitude sampling point number sequence of the corresponding frame, and calculate the threshold offset proportion value, difference the offset proportion value and the key offset sequence length to generate the density trigger coefficient; First, the encryption control sensitive parameter and the encryption density threshold value are obtained, which correspond to the control parameter and the set density threshold value in the signal about encryption protection, respectively. These threshold values are usually derived from previous system parameter settings or automatically calculated and adjusted by the control module. In actual applications, the encryption control sensitive parameter may represent the encryption strength, data variation range, or other characteristics of a certain specific data frame, and the density threshold value is a standard value set to determine when further encryption protection is needed. Subsequently, a numerical comparison operation is performed to compare the size relationship between the encryption control sensitive parameter and the encryption density threshold value. When the value of the encryption control sensitive parameter is greater than the encryption density threshold value, it means that the sensitivity of the current data frame has exceeded the allowed range, and further protection measures need to be taken. In actual scenarios, such as in encrypted voice signal processing, when the variation amplitude of the voice signal exceeds the set density threshold value, it may be necessary to strengthen the data protection measures. Next, the amplitude sample point number sequence corresponding to the frame is extracted, which refers to the numbering sequence of each sample point in the data frame, indicating the amplitude data of these sample points. These sample point numbers will help the system to lock the data range that needs to be operated. By analyzing the numbering sequence of the sample points, the threshold offset proportion value is calculated, which represents the degree of data offset, i.e., the difference between the current sample point amplitude and the set threshold value. In actual calculations, for example, assuming that the current sample point amplitude is 0.8 and the threshold value is 0.5, the offset proportion value is (0.8-0.5) / 0.5=0.6. Finally, the offset proportion value is compared with the key offset sequence length, which is usually a protection parameter set by the system, indicating the influence range of the key on the sample point amplitude in the encryption operation. By comparing the difference between the two, the density trigger coefficient is generated, which can reflect the encryption level of the current data and determine the strength of the subsequent encryption operation. In actual applications, assuming that the key offset sequence length is 5 and the offset proportion value is 0.6, the difference comparison result will determine the specific value of the density trigger coefficient. Finally, this density trigger coefficient will affect the encryption protection level of the data frame. Time axis rearrangement submodule: based on the density trigger coefficient, the amplitude sample point sequence and the key offset sequence length are called, and the time axis rearrangement operation is performed for the same time period index. The interval difference value after rearrangement is recorded, and the time index movement rate interval is calculated to obtain the time rearrangement rate interval value. In this submodule, first, the amplitude sampling point sequence and the key offset sequence length are called based on the density trigger coefficient. This step uses the density trigger coefficient calculated in the previous stage to drive the rearrangement operation of the time axis. The amplitude sampling point sequence contains the amplitude information of each sampling point in the data frame, and the key offset sequence length reflects the influence range of the key in the encryption process. The data of these two sequences are used to further adjust the arrangement order of the time axis. In the specific execution process, first, the density trigger coefficient is taken as the input, and the data range that needs to be rearranged is determined through certain adjustment rules. The time axis rearrangement operation is performed for the same time period index. The time period index represents the position of the data in the time sequence. By modifying the time sequence of the sampling points, the positions of these sampling points will change according to the influence of the density trigger coefficient. At this time, the interval difference value between the frames after rearrangement is recorded, that is, the time difference between the new and old sampling points is obtained by calculating the position change of the sampling points before and after rearrangement. In specific operations, assuming that the time interval between two sampling points is originally 10 milliseconds, and after time axis rearrangement, the interval becomes 12 milliseconds, then the interval difference is 2 milliseconds. Next, the moving speed interval of the time index is calculated. The moving speed of the time index represents the speed at which the data points move on the time axis. It is usually calculated by comparing the interval difference after rearrangement with the interval before rearrangement. Through the ratio of the difference value of the time index to the interval before rearrangement, an interval value of the moving speed can be obtained. For example, adjusting the original 10 millisecond interval to 12 milliseconds means that the change amplitude of the time index moving speed interval is 20%. Finally, the time rearrangement speed interval value is the result of this process, which can reflect the degree of change of the data after time axis rearrangement, and thus affect the generation effect of the encrypted voice. Amplitude offset calculation submodule: call the corresponding value of the voiceprint feature matrix and the amplitude data according to the time rearrangement speed interval value, perform multiplication and addition offset operation, multiply and sum each voiceprint feature matrix value and amplitude sampling point amplitude, and add offset difference, to obtain the disturbance energy change rate of each frame, and generate amplitude disturbance change rate value; During the execution of the sub-module, first, the corresponding values of the voiceprint feature matrix and the amplitude data are called according to the time rearrangement rate interval value. The voiceprint feature matrix stores the voiceprint features of the data frame, and the amplitude data contains the amplitude information of each sampling point. At this time, the time rearrangement rate interval value will be used to adjust the interaction of these feature matrices and amplitude data. In practical applications, relevant data can be obtained by looking up the voiceprint features corresponding to specific time index points. The multiplication and addition offset operation is performed to multiply and sum each voiceprint feature matrix value and the amplitude sampling point amplitude, and to superimpose the offset difference. This process is actually a weighting operation on the original amplitude data. By multiplying each sampling point amplitude with a specific weight (such as the intensity of the voiceprint feature), a weighted sum is obtained, and then an offset difference is added to simulate the effect of data disturbance. In actual operation, assuming that the voiceprint feature matrix value is 0.3 and the amplitude data is 0.7, the product is 0.21, and after adding the offset difference 0.05, the disturbance energy change rate is 0.26. Finally, the disturbance energy change rate of each frame will reflect the amplitude change degree of the data frame under the influence of the voiceprint feature. The amplitude disturbance change rate value generated thereby will provide the required disturbance information for the encrypted speech frame generation module for the subsequent encryption process The encrypted frame generation sub-module calls the amplitude disturbance change rate value and the key sampling point amplitude parameter. For samples with an amplitude disturbance change rate value below the density threshold interval, limited amplitude disturbance is performed and the phase is fixed. The disturbance index value and the key mapping table number sequence are combined to form frame structure additional information to generate encrypted speech frame structure information. In this stage, the amplitude disturbance change rate value and the key sampling point amplitude parameter are first called. These amplitude parameters represent the amplitude data of the sampling points that need to be processed in the encryption process. For samples with an amplitude disturbance change rate value below the density threshold interval, limited amplitude disturbance is performed and the phase is fixed. In actual operation, for sampling points with a small disturbance change rate, the system will perform a small range of disturbance on their amplitudes to maintain the phase unchanged, thereby avoiding excessive disturbance to the data itself, which helps to improve the robustness of encryption. Assuming that the disturbance change rate of certain sampling points is 0.05 and is below the set density threshold of 0.1, the system will perform limited disturbance on the amplitudes of these sampling points, such as adjusting their amplitudes to 0.06 while keeping their phases unchanged. Next, the disturbance index value and the key mapping table number sequence are combined to form frame structure additional information. The disturbance index value refers to the position of the corresponding data point in the sequence after disturbance, and the key mapping table number sequence is the key information related to the encryption operation. By combining the two, the system can form a complete encrypted speech frame structure information for subsequent transmission and decryption processes.

[0028] The fingerprint embedding transmission module includes: The spectrum decomposition submodule: obtain the encrypted voice signal, perform spectrum decomposition on the encrypted voice frame structure information of the voice signal, divide the frequency domain signal into fixed width sections, detect the energy density value and phase offset rate value of each section, calculate the energy balance difference value between adjacent sections according to the energy density distribution characteristics, and adjust the section distribution boundary according to the phase offset rate change range, to generate a section energy difference value sequence; In the execution process of this module, first, the encrypted voice signal needs to be obtained, and the encrypted voice frame structure information thereof is subjected to spectrum decomposition. The spectrum decomposition is performed by Fourier transform or similar algorithm on the audio signal to convert the time domain signal into the frequency domain signal. This process can be realized by short-time Fourier transform (STFT). Assuming that the audio signal is processed at a sampling rate of 16 kHz, the audio signal is first divided into small frames, each frame has a length of 20 milliseconds, and the frame shift is 10 milliseconds, thereby obtaining the frame structure of a signal. Each frame is converted into the frequency domain by FFT (fast Fourier transform) to obtain the frequency spectrum data of the frame. Then, the frequency domain signal is divided into fixed width sections, and the width of each section is set according to the spectral characteristics of the signal. For example, the frequency domain can be divided into sections with a width of 200 Hz. In these frequency bands, the energy density value of each frequency band is calculated. The energy density refers to the total energy of the signal in the frequency band, which is usually obtained by calculating the sum of the squares of the frequency spectrum amplitudes in the frequency band. For example, assuming that the frequency spectrum values in a 200 Hz wide frequency band are [0.1, 0.3, 0.5, 0.2], the energy density of the section is , about 0.5477. Then, the energy balance difference value between adjacent sections is obtained by calculating the energy density difference value of adjacent sections. For example, if the energy density of a section is 0.5477 and the energy density of the adjacent section is 0.6, the energy balance difference value between them is 0.6-0.5477=0.0523. According to the energy difference value and the change range of the phase offset rate, the distribution boundary of the frequency domain section is adjusted to make the energy difference between adjacent sections more balanced. Finally, a section energy difference value sequence is obtained, which records the energy difference between each pair of adjacent sections. The generation process of this sequence reflects the change characteristics of the encrypted voice signal in the frequency domain.

[0029] The phase difference embedding submodule: calls the previous section energy difference value sequence, calculates the phase difference value range for adjacent frequency bands, obtains a pseudo-random sequence, and performs corresponding matching on the pseudo-random sequence and the phase difference value range. The phase offset adjustment is performed according to the energy difference value and the amplitude ratio of the pseudo-random sequence, a phase difference value matrix after embedding the pseudo-random information is generated, and the frequency band phase embedding coefficient is obtained. In the execution of the phase difference embedding submodule, first, the energy difference data between adjacent frequency bands are extracted from the energy difference sequence obtained in the previous stage, which reflects the energy gap between different frequency bands for subsequent phase adjustment. Next, for each pair of adjacent frequency bands, the phase difference value range is calculated, which refers to the difference between the phases of adjacent frequency bands, usually obtained by calculating the phase difference of the center frequency of the frequency band. For example, if the phase of the first frequency band is 10° and the phase of the second frequency band is 20°, the phase difference is |20°-10°|=10°. Next, a pseudo-random sequence is obtained, which is generated by the encryption system according to the key. Its role is to ensure the randomness and unpredictability of encrypted data. By corresponding matching of the pseudo-random sequence and the phase difference value range, each bit of the pseudo-random sequence will be paired with the corresponding phase difference value range to generate a matching scheme for subsequent encryption. For example, if a bit of the pseudo-random sequence is 1 and the phase difference range is 10°, the amplitude ratio at that position (such as 1:10) is calculated to perform phase offset adjustment. Through this offset adjustment, the phase difference value matrix is updated, so that the original phase difference value changes after adjustment. Finally, the frequency band phase embedding coefficient is obtained, which represents the degree of phase offset of each frequency band after phase adjustment, providing necessary information for the subsequent encryption and decryption process.

[0030] Fingerprint fusion submodule: based on the previous frequency band phase embedding coefficient, select the corresponding frequency band encrypted voice data frame, weight the spectrum amplitude and phase embedding coefficient of the voice frame, calculate the amplitude-phase balance value of the fused spectrum, and form a complete transmission data packet by combining the weighted fusion result with the original encrypted voice frame data to generate a fingerprint marked transmission data; In the fingerprint fusion submodule, first, according to the frequency band phase embedding coefficient generated in the previous stage, select the corresponding frequency band encrypted voice data frame, which represents the encrypted voice signal and contains the amplitude and phase information of each frequency band. Next, weight the spectrum amplitude and phase embedding coefficient of the voice frame. In the weighting fusion process, the combination of amplitude and phase follows certain weighting rules, such as determining the weight distribution of amplitude and phase according to a certain encryption control parameter. For example, if the weight of amplitude is 0.7 and the weight of phase is 0.3, then according to these weights, weight the amplitude value and phase embedding coefficient in proportion. For example, if the amplitude of a frame is 0.5 and the phase embedding coefficient is 10°, then the fused value is , the balance degree value of the fused spectrum is calculated, that is, by measuring the balance degree between the amplitude and the phase after fusion, it is judged whether the weighted data meets the expected encryption standard, the higher the balance degree value, the closer the fusion of the amplitude and the phase, in this process, a threshold value can be set to judge whether the fused result meets the required encryption strength, finally, the weighted fusion result is matched with the original encrypted speech frame data to form a complete transmission data packet, this data packet contains the encrypted speech signal, the weighted spectrum information and the fingerprint mark for decryption, the generated fingerprint mark transmission data not only guarantees the encryption of the data, but also allows the subsequent verification of the integrity and source of the data.

[0031] The decryption verification check module further comprises: The fingerprint mark reverse reset submodule: acquires the fingerprint mark transmission data, performs reverse sorting operation on the sample index in the data according to the disturbance index table, performs weighting compensation operation on the sample sequence after index rearrangement and the amplitude offset parameter in the key mapping table, rearranges the time sequence position according to the sample amplitude sequence after compensation, restores the time continuous structure of the signal, and generates a time sequence reset signal energy; The fingerprint mark transmission data is specifically a digital audio stream data containing 1024 samples, the sample amplitude of the data stream has been encrypted and disturbed, and the original time sequence index has been disturbed, for example, the sample with the original index sequence {1, 2, 3, …, 1024} is disturbed to {512, 1, 89, …, 233} in the transmission data, the sample index in the data is inversely sorted according to the disturbance index table, first, the disturbance index table is extracted from the key mapping data packet received together with the transmission data, which records the mapping relationship between the original index and the disturbed index, for example, the table records {1->512, 2->1, 3->89, …}, the inverse sorting operation is to read the sample with the index of 512 in the transmission data and place it in the first position of the new sequence, read the sample with the index of 1 and place it in the second position of the new sequence, and so on, until all 1024 samples are placed in their original time sequence index positions to form a sample sequence after index rearrangement, then the sample sequence after index rearrangement is subjected to weighted compensation operation with the amplitude offset parameter in the key mapping table, the amplitude offset parameter table corresponding to each sample index is extracted from the key mapping data packet, which gives the amplitude offset added to each sample in the encryption process, for example, the amplitude offset parameter of the sample with the index of 1 is +15.3, and the offset parameter of the sample with the index of 2 is -8.1, the process of weighted compensation operation is to subtract the offset corresponding to each sample in the amplitude offset parameter table from the sample in the sequence after index rearrangement, specifically, if the amplitude of the first sample in the sequence after index rearrangement is 125.3, the compensated amplitude is 125.3-15.3=110.0, and if the amplitude of the second sample is 81.9, the compensated amplitude is 81.9-(-8.1)=90.0, the weight in this operation is set to 1.0, that is, the subtraction operation is directly performed without scaling the offset parameter, after the amplitude compensation of all 1024 samples is completed, the compensated sample amplitude sequence is obtained, and then the time sequence position is rearranged according to the compensated sample amplitude sequence to restore the time continuous structure of the signal, this step is to arrange the amplitude sequence obtained in the foregoing according to its index sequence {1, 2, 3, …, 1024} to form a complete time sequence data frame, which is the time reset signal energy.

[0032] The frequency spectrum feature extraction submodule: based on the time reset signal energy, the signal is subjected to frequency spectrum decomposition operation to obtain amplitude set and phase set of each frequency section, the energy density value is calculated for each frequency section amplitude set, and the phase offset rate value is calculated according to the phase set, the amplitude set and the phase set are called to perform section ratio analysis to generate the frequency spectrum energy distribution rate; Based on the time sequence reset signal energy generated in the last stage, which is composed of 1024 time sequence sample values, the signal is subjected to a spectral decomposition operation, which takes the first 512 sample values of the signal as an analysis window, performs a discrete Fourier transform on the data in the window to convert it from the time domain to the frequency domain, thereby obtaining the amplitude set and phase set of each frequency segment. The division standard for the frequency segment is to divide the entire frequency band from 0 Hz to the Nyquist frequency (for example, 22050 Hz) into 10 consecutive and non-overlapping segments, the first segment is 0-2205 Hz, the second segment is 2206-4410 Hz, and so on. The real and imaginary parts of each frequency point after transformation are used to calculate the amplitude and phase of the frequency point. The amplitudes of all frequency points falling within the first segment frequency range form the amplitude set of the first segment, and their phases form the phase set of the first segment. For example, the amplitude set of the first segment may be {25.1, 30.3, 28.5, …}, and the phase set is {0.1π, 0.12π, 0.09π, …}. The energy density value is calculated for each frequency segment amplitude set. The calculation process of the energy density value is to sum the squares of all amplitudes in the frequency segment amplitude set, and then divide by the number of amplitudes in the set. The specific calculation formula is wherein represents the energy density value, is the number of frequency points contained in the first segment, is the amplitude of the first frequency point in the segment. Assuming that the first segment contains 50 frequency points and the sum of their squares is 125000, then its energy density value is 125000 / 50 = 2500. The phase shift rate value is calculated based on the phase set. The calculation method of this value is to obtain the standard phase reference value of the segment from the key mapping table, for example, the reference value of the first segment is 0.11π. Then calculate the arithmetic mean of all phase values in the phase set of the segment, for example, the average value of the phases of the first segment is 0.10π. The phase shift rate value is the difference between the average value and the reference value divided by the reference value, i.e. (0.10π-0.11π) / 0.11π≈-0.091. Then call the amplitude set and the phase set for inter-segment ratio analysis. This analysis calculates the ratio of the energy density values of two adjacent frequency segments, for example, the energy density ratio of the second segment to the first segment. If the calculated energy density value of the second segment is 3750, then the ratio is 3750 / 2500 = 1.5. This operation is performed for all 9 pairs of adjacent frequency segments. Finally, a sequence consisting of 9 ratios is generated, which is the spectral energy distribution rate. The phase consistency matching sub-module: according to the spectral energy distribution rate, the phase offset rate difference value of adjacent frequency bands is extracted, the corresponding offset reference rate in the key mapping table is compared, the difference value comparison coefficient is calculated, the verification sequence is formed according to the comparison coefficient of each frequency band, the consistency matching is performed in each section with the spectral fingerprint structure, and the decryption consistency check result is generated; According to the spectrum energy distribution rate composed of 9 energy density ratio values generated in the last stage, the difference value of the phase shift rate of adjacent frequency bands is extracted, which is to call the phase shift rate values of each frequency band calculated in the last stage, for example, the phase shift rate values of the first three frequency bands are -0.091, -0.121, and -0.101, respectively. The difference value between the 2nd frequency band and the 1st frequency band is calculated as -0.121-(-0.091)=-0.030, and the difference value between the 3rd frequency band and the 2nd frequency band is calculated as -0.101-(-0.121)=0.020. This difference value calculation is performed on all 8 pairs of adjacent phase shift rates to obtain a sequence containing 8 difference values, which is then compared with the corresponding offset reference rate in the key mapping table to calculate the difference value comparison coefficient. The key mapping table stores a standard offset reference rate sequence, which is a preset fingerprint core structure for verification, for example, its first two values are {-0.025, 0.022}. The calculation process of the difference value comparison coefficient is to compare the actually calculated difference value with the reference rate in the key, for example, for the first difference value -0.030, the corresponding reference rate is -0.025, and the comparison coefficient is 1-|-0.030-(-0.025)) / (-0.025)|=1-|-0.005 / -0.025|=1-0.2=0.8, for the second difference value 0.020, the corresponding reference rate is 0.022, and the comparison coefficient is 1-|(0.020-0.022) / 0.022|=1-|-0.002 / 0.022|≈0.909. According to the comparison coefficients of each frequency band, a verification sequence is formed, the 8 difference value comparison coefficients calculated are arranged in order to form a verification sequence composed of 8 elements, for example, {0.8, 0.909,…}, and then a section-by-section consistency matching is performed with the spectrum fingerprint structure. The spectrum fingerprint structure here specifically refers to a preset consistency determination threshold, which is set by referring to a large amount of experimental data analysis, when the comparison coefficient is higher than 0.75, the confidence of successful matching is higher, and when it is lower than 0.6, the confidence of failed matching is higher, therefore the consistency determination threshold is set to 0.7. The execution action of section-by-section consistency matching is to traverse each comparison coefficient in the verification sequence to judge whether it is greater than 0.7, if the verification sequence is {0.8, 0.909, 0.65,…}, the first element 0.8 is greater than 0.7, which is determined as matching, the second element 0.909 is greater than 0.7, which is determined as matching, and the third element 0.65 is less than 0.7, which is determined as not matching. Only when all 8 comparison coefficients in the verification sequence are greater than 0.7, the final matching result is "consistent", otherwise, as long as one or more coefficients are not greater than 0.7, the final result is "inconsistent". This final binary determination result is the decryption consistency verification result.

[0033] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A sound transmission system based on encryption and decryption, characterized in that, The system includes: The audio signal acquisition module acquires the audio signal, extracts the amplitude sampling points, frequency band energy value, fundamental frequency change rate and formant spacing in frames, performs ratio calculation and weighted accumulation, and records the amplitude sampling points as an encrypted reference to generate encrypted voiceprint reference data. Sensitive weight calculation module: Based on encrypted voiceprint baseline data, it extracts phoneme duration, semantic symbols and energy mean, performs ratio calculation and fusion, and generates encrypted control sensitive parameters by weighting the results with voiceprint confidence. Encryption density execution module: Based on the comparison between the encryption control sensitive parameters and the density threshold, if the threshold is exceeded, the sampling points and key offset sequence are called to perform reordering and offsetting; if the threshold is not exceeded, the key sampling points are called to perform perturbation to generate encrypted voice frame structure information. Fingerprint embedding transmission module: Based on the encrypted voice frame structure information, the spectrum is decomposed and partitioned, the energy density and phase shift difference are calculated, and a pseudo-random sequence is embedded to form a spectral fingerprint to generate fingerprint markers for transmission data; Decryption verification module: Based on the fingerprint marker transmitted data, it calls the perturbation index and key mapping table to perform reverse sorting and amplitude compensation, recovers the decrypted voice, extracts the spectrum data and compares the spectrum fingerprint to generate a decryption consistency verification result.

2. The sound transmission system based on encryption and decryption according to claim 1, characterized in that: The encrypted voiceprint reference data includes a set of voiceprint numerical features, amplitude sampling point reference information, and encryption reference parameters. The encryption control sensitive parameters specifically include a semantic and voiceprint fusion sensitivity coefficient, an encryption strength adjustment factor, and a semantic weight correction term. The encrypted voice frame structure information includes a perturbation index record, a key mapping supplementary table, and a synchronization marker field. The fingerprint marker transmission data specifically includes a spectral fingerprint structure, pseudo-random sequence embedding information, and an encrypted voice data encapsulation body. The decrypted voice signal includes a reset sample sequence, amplitude compensation results, and a decrypted voice reconstruction waveform.

3. The sound transmission system based on encryption and decryption according to claim 1, characterized in that: The sound signal acquisition module includes: Sound wave sampling submodule: Acquires the input speech signal, extracts the sound wave amplitude sampling point sequence by dividing the signal into frames according to the time interval, detects the amplitude change between adjacent sampling points in each frame and records the length of the sampling frame sequence, calculates the average amplitude and change rate of each frame amplitude sequence and forms an amplitude statistics set, and generates sound wave amplitude feature values. Energy-frequency calculation submodule: Based on the acoustic wave amplitude feature value, extract the frequency band energy value and fundamental frequency change rate of the corresponding frame, calculate the ratio of frequency band energy value to fundamental frequency change rate, call the ratio sequence between adjacent frames to perform difference and accumulate to form a ratio change sequence, and generate energy-frequency ratio coefficient; Voiceprint feature generation submodule: Based on the energy-frequency ratio coefficient, it calls the inter-frame formant spacing sequence, calculates the weighted value of the ratio change sequence and the formant spacing change amplitude, performs frame-level weighted accumulation operation to form a feature vector set, and uses the sound wave amplitude sampling point sequence as an encryption reference to generate encrypted voiceprint reference data.

4. The sound transmission system based on encryption and decryption according to claim 1, characterized in that: The sensitivity weight calculation module also includes: Phoneme Duration Analysis Submodule: Obtains the duration of each phoneme sequence in the encrypted voiceprint baseline data, detects the start and end time points corresponding to the phoneme boundaries and calculates their time intervals, sums and averages the time intervals of all phonemes, obtains the average duration interval value of phoneme sequences under different speech segments, and generates the average phoneme duration value. Semantic frequency ratio submodule: Based on the semantic symbol sequence annotated in the encrypted voiceprint benchmark data, it counts the number of times each semantic symbol appears in the speech sample and calculates the frequency of occurrence. It performs a ratio operation based on the average phoneme duration and the frequency of occurrence of each semantic symbol to obtain the relative duration ratio of the corresponding semantic symbol at the phoneme level and generate semantic ratio parameters. Sound Energy Fusion Calculation Submodule: Calls the sound wave energy sequence in the encrypted soundprint baseline data, calculates the average energy amplitude for each time slice, performs fusion calculation based on the semantic ratio parameter and the average sound wave energy, weights and superimposes the two data in the same time slice and calculates the comprehensive value to generate the fusion energy weight value; Linear weighted control submodule: Based on the voiceprint confidence value output from the voiceprint recognition stage, it calls the fusion energy weight value, performs a linear weighting operation on the two, calculates the proportional contribution of the weight coefficient to the confidence value, obtains the fusion parameters used to control the encryption strength, and generates the semantic and voiceprint fusion sensitivity coefficient.

5. A sound transmission system based on encryption and decryption according to claim 1, characterized in that: The encryption density execution module also includes: Threshold Judgment Submodule: Obtains the encryption control sensitive parameter and the encryption density threshold, performs a numerical comparison operation on the two, and when the encryption control sensitive parameter exceeds the encryption density threshold, extracts the amplitude sampling point number sequence of the corresponding frame, calculates its over-threshold offset ratio, performs a differential comparison with the key offset sequence length, and generates the density trigger coefficient. Time axis rearrangement submodule: Based on the density trigger coefficient, the amplitude sampling point sequence and key offset sequence length are called. Time axis rearrangement operation is performed for the same time period index, the inter-frame interval difference after rearrangement is recorded, and the time index movement rate range is calculated to obtain the time rearrangement rate range value. Amplitude offset calculation submodule: Based on the time rearrangement rate interval value, call the corresponding value of the voiceprint feature matrix and amplitude data, perform multiply-add offset operation, multiply and sum the values ​​of each voiceprint feature matrix and amplitude sampling point, and add the offset difference to obtain the perturbation energy change rate of each frame and generate the amplitude perturbation change rate value. Encrypted frame generation submodule: Calls the amplitude perturbation change rate value and the amplitude parameter of key sampling points, performs limited amplitude perturbation on samples whose amplitude perturbation change rate value is lower than the density threshold range and keeps the phase fixed, combines the perturbation index value with the key mapping table number sequence to form additional frame structure information, and generates encrypted voice frame structure information.

6. The sound transmission system based on encryption and decryption according to claim 1, characterized in that: The fingerprint embedding transmission module includes: Spectrum decomposition submodule: Acquires encrypted speech signal, performs spectrum decomposition on the encrypted speech frame structure information of the speech signal, divides the frequency domain signal into fixed-width segments, detects the energy density value and phase offset rate value of each segment, calculates the energy balance difference between adjacent segments based on the energy density distribution characteristics, and adjusts the segment distribution boundary according to the phase offset rate variation range to generate a segment energy difference sequence. Phase difference embedding submodule: Calls the preceding segment energy difference sequence, calculates the phase difference range for adjacent frequency bands, obtains a pseudo-random sequence, matches the pseudo-random sequence with the phase difference range, performs phase offset adjustment according to the amplitude ratio of the energy difference and the pseudo-random sequence, generates a phase difference matrix after embedding pseudo-random information, and obtains the frequency band phase embedding coefficient. Fingerprint fusion submodule: Based on the preceding frequency band phase embedding coefficients, select the encrypted voice data frames of the corresponding frequency band, perform weighted fusion of the spectral amplitude and phase embedding coefficients of the voice frames, calculate the amplitude-phase balance value of the fused spectrum, combine the weighted fusion result with the original encrypted voice frame data to form a complete transmission data packet, and generate fingerprint-marked transmission data.

7. A sound transmission system based on encryption and decryption according to claim 1, characterized in that: The decryption verification module also includes: Fingerprint tag reverse reset submodule: acquires fingerprint tag transmission data, performs reverse sorting operation on the sample point index in the data according to the perturbation index table, performs weighted compensation operation on the sample point sequence after index rearrangement and the amplitude offset parameter in the key mapping table, rearranges the timing position according to the compensated sample point amplitude sequence, restores the time continuous structure of the signal, and generates timing reset signal energy; Spectrum Feature Extraction Submodule: Based on the energy of the time-series reset signal, the signal is decomposed into a spectrum to obtain the amplitude set and phase set of each frequency band. The energy density value is calculated for the amplitude set of each frequency band, and the phase offset rate value is calculated based on the phase set. The amplitude set and phase set are called to perform inter-segment ratio analysis to generate the spectrum energy distribution rate. Phase consistency matching submodule: Based on the spectral energy distribution rate, extract the phase offset rate difference between adjacent frequency bands, compare it with the corresponding offset benchmark rate in the key mapping table, calculate the difference comparison coefficient, form a verification sequence based on the comparison coefficient of each frequency band, perform segment-by-segment consistency matching with the spectral fingerprint structure, and generate decryption consistency verification results.

Images