A method and system for secure encryption of user voice data

Abstract

The present invention relates to the field of data encryption processing, and in particular to a method and system for securely encrypting user voice data. The method comprises the following steps: performing a short-time Fourier transform on a user voice signal to obtain a spectrogram; calculating the degree of voiceprint representation of each data point in the spectrogram; obtaining a pre-obtained frequency interval; setting the frequency segmentation ratio of each data point according to the degree of voiceprint representation; randomly arranging the frequency segmentation ratios of all data points at a moment, and segmenting the frequency intervals according to the order of any arrangement result to obtain the segmentation result of that arrangement; analyzing the frequency offset degree of the segmentation result of each arrangement; selecting the segmentation result of the arrangement corresponding to the maximum frequency offset degree as the target segmentation result; randomly selecting an integer in each interval of the target segmentation result as the frequency encryption data of the corresponding data point; completing the amplitude encryption of each data point; thereby realizing encryption of the user voice signal. This improves the voice signal encryption effect.

Description

Technical Field

[0001] The present invention relates to the field of data encryption processing, and in particular to a method and system for securely encrypting user voice data. Background Art

[0002] With the rapid development of information technology, voice interaction technology has become ubiquitous in our lives and work, such as voice control and voice payment. Voiceprint recognition, as a biometric identification technology, has played an increasingly important role in identity authentication, access control, and other fields in recent years. Due to individual differences in physiological characteristics such as vocal cord shape, vocal tract length and shape, oral cavity shape, and tooth and tongue position, the sound produced by each person has unique voiceprint characteristics, making voiceprint recognition highly accurate and reliable. However, precisely because voiceprint information uniquely identifies an individual, once voiceprint data is leaked, it can pose serious security risks to users, such as identity theft, privacy infringement, and financial loss.

[0003] Existing voice data security technologies typically use encryption algorithms for transmission and storage of voice content to prevent theft or tampering. However, encryption protection measures for voiceprint information are less comprehensive. Many systems may only use simple encryption methods or even store voiceprint information in plain text when collecting and storing it, exposing voiceprint data to the risk of unauthorized access during transmission and storage in databases.

[0004] Traditional encryption algorithms do not consider the voiceprint characteristics of speech information and use a unified encryption method for all speech data. As a result, the encryption method cannot effectively conceal the voiceprint characteristics of the speech information. Therefore, the voiceprint information in the speech can be extracted through simple analysis and processing. Therefore, how to effectively conceal the voiceprint information in the speech through encryption has become the research focus of this invention. Summary of the Invention

[0005] In order to solve the problem of how to better conceal the voiceprint information in speech through encryption, the present invention provides a method and system for securely encrypting user voice data.

[0006] In a first aspect, the present invention provides a method for securely encrypting user voice data, which adopts the following technical solution:

[0007] A method for securely encrypting user voice data, comprising the steps of:

[0008] Obtain user voice signals;

[0009] Performing a short-time Fourier transform on the user's speech signal to obtain a spectrogram; calculating the degree of voiceprint representation of each data point in the spectrogram, wherein the degree of voiceprint representation is negatively correlated with the distance between the data point and the formant, and is also negatively correlated with the frequency difference and variation pattern of the data point and harmonics;

[0010] Obtain a pre-obtained frequency interval; set a frequency segmentation ratio for each data point based on the voiceprint characterization degree, wherein the frequency segmentation ratio of the data point is positively correlated with the voiceprint characterization degree; randomly arrange the frequency segmentation ratios of all data points at a moment, and segment the frequency interval according to the order of any arrangement result to obtain a segmentation result of that arrangement; analyze the frequency offset degree of each segmentation result, wherein the frequency offset degree reflects the difference between the center of the interval corresponding to the frequency segmentation ratio of each data point and the original frequency value of the data point; select the segmentation result of the arrangement corresponding to the maximum frequency offset degree as the target segmentation result; and randomly select an integer in each interval of the target segmentation result as the frequency encryption data of the corresponding data point;

[0011] Complete the amplitude encryption of each data point to achieve user voice signal encryption.

[0012] The present invention encrypts each information differently according to the correlation between each information and the voiceprint feature in the user voice signal, so that the information with a stronger correlation with the voiceprint feature is encrypted more strongly, thereby improving the encryption effect; further, when analyzing the correlation between each information and the voiceprint feature, the resonance peak, harmonic frequency and other information that can better reflect the voiceprint feature are used to accurately reflect that each information has the voiceprint feature, thereby providing a basis for subsequent precise encryption; further, when encrypting the user voice signal, the encrypted data is set by randomly selecting data in the interval, thereby increasing the randomness of the encrypted data, better concealing the regular information of the data, and improving the encryption effect; further, when encrypting the user voice signal, the segmentation ratio is set according to the degree of voiceprint representation, thereby making the selection interval of the encrypted data of the information with a relatively large correlation with the voiceprint feature larger, increasing the randomness of the encrypted data of the information with a relatively large correlation with the voiceprint feature, and improving the encryption effect of the information with a large correlation with the voiceprint feature. Furthermore, in order to make the frequency of the data point have a large difference with the encrypted data, the frequency offset degree is introduced to accurately measure the frequency difference between the data point and the corresponding interval, thereby laying the foundation for subsequent improvement of the encryption effect.

[0013] Preferably, the step of calculating the voiceprint representation degree of each data point in the spectrogram includes:

[0014] Obtain the resonance peak area, harmonic frequency and change model of each data point in the spectrogram; if the data point belongs to the resonance peak area of the spectrogram, set the peak distance of the data point to the preset first anti-zero value, otherwise, connect the data point with the geometric center of the resonance peak area, and take the distance from the data point to the intersection of the connection line as the peak distance; add the difference between the frequency of the data point and the harmonic frequency with the smallest frequency difference plus the preset second anti-zero value to obtain the harmonic difference; obtain the change law reference model of the data point, add the difference between the change model of the data point and the change law reference model plus the preset third anti-zero value to obtain the change law difference; calculate the voiceprint characterization degree according to the peak distance, harmonic difference and change law difference of the data point, and the voiceprint characterization degree is negatively correlated with the peak distance, harmonic difference and change law difference.

[0015] The present invention takes into account that the frequencies of resonance peaks and harmonics can strongly reflect voiceprint features. Therefore, the correlation between data points and voiceprint features is accurately reflected by introducing the peak distance between data points and resonance peaks, the difference between the frequencies of data points and harmonic frequencies, and the difference in change rules.

[0016] Preferably, the method for obtaining the change model includes:

[0017] A preset area is obtained with each data point as the center, a Gaussian mixture model is constructed based on all data points in the preset area, and the Gaussian mixture model of the preset area is used as the change model of each data point.

[0018] Preferably, the method for obtaining the variation law reference model includes:

[0019] The similarity of the change model of each two data points is obtained and recorded as the change similarity. The change similarities of all two data points are clustered into two categories. The data points corresponding to the change similarity in the category with the largest change similarity mean are recorded as regular reference data points. The regular reference data points are clustered according to the change similarity, and the regular reference data points in a category are recorded as regular data points of the same type. The mean value of the corresponding parameter of the change model of the regular data points of the same type is recorded as the comprehensive parameter, and the change model composed of the comprehensive parameters is used as the regular model.

[0020] The regular model with the smallest difference from the variation model of the data point is obtained as the reference model of the variation pattern of the data point.

[0021] There are multiple rules that characterize voiceprint features in the speech signal of the present invention. By introducing classification, data with different rules are separated, thereby increasing the accuracy of regular model fitting.

[0022] Preferably, the step of setting the frequency division ratio of each data point according to the voiceprint characterization degree includes:

[0023] The voiceprint representation degree of the data point is divided by the cumulative sum of the voiceprint representation degrees of all data points at a moment, and the obtained quotient is used as the frequency division ratio of the data point.

[0024] The present invention reflects the frequency division ratio through the proportion of voiceprint representation degree, so that data points with a large voiceprint representation degree have a larger frequency division ratio, providing a basis for subsequently improving the encryption effect of information with a large voiceprint representation degree.

[0025] Preferably, the segmenting of the frequency intervals according to the arrangement order of any arrangement result to obtain the segmentation result of the arrangement includes:

[0026] According to the arrangement order of any arrangement result, intervals of corresponding proportions are intercepted from the frequency interval based on the frequency division ratio in sequence to obtain a plurality of intervals.

[0027] Preferably, analyzing the frequency deviation degree of the segmentation results of each arrangement includes:

[0028] In the segmentation results of any arrangement, the frequency value at the center point of each interval is obtained, and the absolute value of the difference between the frequency value at the center point of the data point and the interval corresponding to the frequency segmentation ratio is recorded as the frequency deviation value of the data point; the voiceprint representation degree of the data point is used as the weight, and the frequency deviation values of all data points at a moment are weighted and summed to obtain the frequency deviation degree of the segmentation result of this arrangement.

[0029] The present invention introduces the voiceprint representation degree as a weight to weight the frequency deviation value, so that the data points with a large voiceprint representation degree have a greater impact on the frequency deviation degree, providing a basis for subsequently improving the encryption effect of data points with a large voiceprint representation degree.

[0030] Preferably, completing the amplitude encryption of each data point includes:

[0031] Get the pre-obtained amplitude interval;

[0032] According to the method of obtaining the target segmentation result of the frequency interval, the target segmentation result of the amplitude interval is obtained, and an integer is randomly selected in each interval of the target segmentation result of the amplitude interval as the amplitude encrypted data of the corresponding data point.

[0033] Preferably, the method of realizing user voice signal encryption includes:

[0034] The spectrogram composed of the frequency encrypted data and the amplitude encrypted data is used as the encrypted spectrogram, and the encrypted spectrogram is inversely transformed to obtain the encrypted signal of the user voice signal.

[0035] In a second aspect, the present invention provides a user voice data security encryption system, which adopts the following technical solution:

[0036] A user voice data security encryption system includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned user voice data security encryption method is implemented.

[0037] By adopting the above technical solution, the above-mentioned user voice data security encryption method is generated into a computer program and stored in the memory to be loaded and executed by the processor, so that a terminal device is made based on the memory and the processor for easy use.

[0038] The present invention has the following technical effects:

[0039] The present invention encrypts each piece of information differently based on its correlation with the voiceprint feature in the user's voice signal, so that information with a stronger correlation with the voiceprint feature is more strongly encrypted, thereby improving the encryption effect.

[0040] Furthermore, when analyzing the correlation between each piece of information and voiceprint features, information such as resonance peaks and harmonic frequencies that can better reflect voiceprint features can be used to accurately reflect whether each piece of information has voiceprint features, providing a basis for subsequent accurate encryption;

[0041] Furthermore, when encrypting the user's voice signal, the encrypted data is set by randomly selecting data in the interval, which increases the randomness of the encrypted data, better conceals the regular information of the data, and improves the encryption effect;

[0042] Furthermore, when encrypting the user voice signal, the segmentation ratio is set according to the degree of voiceprint representation, so that the selection interval of the encrypted data of the information with a relatively large correlation with the voiceprint characteristics is larger, thereby increasing the randomness of the encrypted data of the information with a relatively large correlation with the voiceprint characteristics, and improving the encryption effect of the information with a relatively large correlation with the voiceprint characteristics.

[0043] Furthermore, in order to make the frequency of the data point have a large difference with the encrypted data, the frequency offset degree is introduced to accurately measure the frequency difference between the data point and the corresponding interval, thereby laying the foundation for subsequent improvement of the encryption effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The present invention is a flowchart of a method for securely encrypting user voice data. DETAILED DESCRIPTION

[0045] The embodiment of the present invention discloses a method for securely encrypting user voice data. Figure 1 , including steps S1 to S5:

[0046] S1: Obtain user voice signal.

[0047] Specifically, user voice signals are collected.

[0048] S2: Perform short-time Fourier transform on the user's voice signal to obtain a spectrogram; calculate the voiceprint representation degree of each data point in the spectrogram, wherein the voiceprint representation degree is negatively correlated with the distance between the data point and the resonance peak, and is also negatively correlated with the difference in frequency and harmonics of the data point and the difference in variation patterns.

[0049] It should be noted that in order to better conceal the voiceprint features in the speech signal, it is necessary to focus on encrypting the data information that has strong voiceprint feature description capabilities. Therefore, it is necessary to analyze the description capabilities of each data information in the speech signal for the voiceprint features.

[0050] S20: Perform short-time Fourier transform on the user's speech signal to obtain a spectrogram.

[0051] It should be noted that the spectrogram obtained by the short-time Fourier transform (SFT) reflects both the frequency domain and timing information of the signal. To better analyze the ability of various information in the speech signal to describe the voiceprint characteristics, the speech signal needs to be SFT to obtain the spectrogram. SFT of speech signals is a well-known technique and will not be further described here.

[0052] S21: Calculate the voiceprint representation degree of each data point in the spectrogram.

[0053] It should be noted that the formant, harmonics and amplitude variation characteristics can better reflect the voiceprint characteristics. Therefore, the ability of each information in the speech signal to reflect the voiceprint characteristics can be analyzed through the formant, harmonics and amplitude variation characteristics.

[0054] Preferably, as an example, calculating the voiceprint representation degree of each data point in the spectrogram includes:

[0055] Obtain the formant area, harmonic frequency and change model of each data point in the spectrogram;

[0056] It should be noted that there are many existing methods for obtaining formant regions and harmonic frequencies in a spectrogram, and any of these methods can be selected to obtain the formant regions and harmonic frequencies. Here is an example method: obtaining the formant regions through image processing methods such as threshold segmentation, obtaining the fundamental frequency through autocorrelation, and then amplifying the fundamental frequency by integer multiples to obtain the harmonic frequencies.

[0057] The peak distance, harmonic difference and variation pattern difference are calculated based on the formant area, harmonic frequency and variation model in the spectrogram.

[0058] The reciprocal of the product of the normalized value of the peak distance of the data point, the normalized value of the harmonic difference, and the normalized value of the variation law difference is taken as the voiceprint representation degree of the data point.

[0059] The above embodiments involve peak distance, harmonic difference and variation law difference. The following describes the method for determining the peak distance, harmonic difference and variation law difference.

[0060] Among them, the method of obtaining the peak distance is first introduced.

[0061] If the data point belongs to the resonance peak area of the spectrogram, the peak distance of the data point is set to the preset first anti-zero value. Otherwise, the data point is connected to the geometric center of the resonance peak area, and the distance between the data point and the intersection of the connection line is used as the peak distance.

[0062] It is understandable that a larger peak distance indicates that the data point is farther away from the formant region, and thus the data point contributes less information to the formant description. Formant is an important information in describing voiceprint features, so the data point reflects less information about the voiceprint features.

[0063] Then the method of obtaining harmonic differences is introduced.

[0064] The difference between the frequency of the data point and the harmonic frequency with the smallest frequency difference is added to the preset second anti-zero value to obtain the harmonic difference. Exemplarily, the difference between the frequency of the data point and the harmonic frequency with the smallest frequency difference can be calculated by calculating the absolute value or standard deviation of the difference.

[0065] It is understandable that harmonic frequency is also important information reflecting voiceprint characteristics. If the frequency of a data point is significantly different from the harmonic frequency, the data point contains less information describing the voiceprint characteristics.

[0066] Finally, the method of obtaining the difference in change patterns is introduced.

[0067] A preset region is obtained with each data point as the center, and a Gaussian mixture model is constructed based on all data points within the preset region. The Gaussian mixture model of the preset region is used as the change model for each data point. The similarity of the change models of each two data points is obtained and recorded as the change similarity. For example, the similarity of the change models of the two data points can be reflected by the cosine similarity of the parameters between the change models of the two data points. The parameters of the change model can be the mean variance of each single Gaussian model in the Gaussian mixture model.

[0068] Cluster the change similarities of all two data points into two categories; record the data points corresponding to the change similarities in the category with the largest change similarity mean as regular reference data points; cluster the regular reference data points based on the change similarity, and record the regular reference data points in one category as regular data points of the same type; record the upward integer value of the mean of the corresponding parameters of the change model of the same type of regular data points as the comprehensive parameter, and use the change model composed of the comprehensive parameters as the regular model; obtain the regular model with the smallest difference from the change model of the data point as the change regular reference model of the data point; add the difference between the change model of the data point and the change regular reference model to a preset third anti-zero value to obtain the change regular difference. For example, the difference between the change model of the data point and the change regular reference model can be reflected by the parameter difference between the models, and the difference between the parameters between the models can be reflected by the Euclidean distance of the parameters between the models.

[0069] It is understandable that there are a variety of change patterns in speech signals that reflect voiceprint characteristics. By decomposing the regular model of each pattern and describing it, and comparing the data point with the regular model with the smallest difference, we can determine whether the data point conforms to the change pattern of the voiceprint.

[0070] S3: Obtain the pre-obtained frequency interval; set the frequency segmentation ratio of each data point according to the voiceprint characterization degree, and the frequency segmentation ratio of the data point is positively correlated with the voiceprint characterization degree; randomly arrange the frequency segmentation ratios of all data points at a moment, and according to the arrangement order of any arrangement result, divide the frequency interval to obtain the segmentation result of that arrangement; analyze the frequency deviation degree of the segmentation result of each arrangement, and the frequency deviation degree reflects the difference between the center of the interval corresponding to the frequency segmentation ratio of each data point and the original frequency value of the data point; select the segmentation result of the arrangement corresponding to the maximum frequency deviation degree as the target segmentation result; randomly select an integer in each interval of the target segmentation result as the frequency encryption data of the corresponding data point.

[0071] It should be noted that to mask the voiceprint characteristics in the speech signal, other frequency values can be used for substitution. However, fixed-value substitution can easily be decrypted through the regularity of the encrypted data, thereby exposing the voiceprint characteristics. Therefore, a replacement value can be randomly selected within a range. The range of possible values allows for a wide variety of values in the frequency-encrypted data, making it difficult to decode the data based on regularity. To better mask information with a high degree of voiceprint representation, the range of frequencies selected for information with a high degree of voiceprint representation in the speech signal can be set to a larger value.

[0072] S30: Acquire a pre-obtained frequency range.

[0073] It should be noted that people's voices generally have a certain frequency range, so the common frequency range of people's voices can be used as the frequency interval, and the selection interval of the encrypted data of each data is set by subsequently dividing the frequency interval.

[0074] Optionally, obtain a pre-obtained frequency range, including:

[0075] The frequency range of human voice is used as the pre-obtained frequency interval. For example, the frequency range of human voice is 200–5000 Hz.

[0076] S31: setting a frequency division ratio for each data point according to the voiceprint characterization degree.

[0077] It should be noted that in order to better conceal information with a high degree of voiceprint representation, the selection interval of the encrypted frequency of information with a high degree of voiceprint representation in the speech signal can be set larger, and based on this, the selection interval of the encrypted frequency of each data point can be intercepted in the frequency interval.

[0078] Preferably, as an example, setting the frequency division ratio of each data point according to the voiceprint characterization degree includes:

[0079] The voiceprint representation degree of the data point is divided by the cumulative sum of the voiceprint representation degrees of all data points at a moment, and the obtained quotient is used as the frequency division ratio of the data point.

[0080] It can be understood that the frequency division ratio is reflected by the proportion of the voiceprint representation degree of the data point, so that the frequency division ratio of the data point with a large voiceprint representation degree is larger, providing a basis for subsequent stronger encryption of the data point with a large voiceprint representation.

[0081] S32: Randomly arrange the frequency segmentation ratios of all data points at a moment, and segment the frequency intervals according to the arrangement order of any arrangement result to obtain the segmentation result of that arrangement; analyze the frequency offset degree of the segmentation result of each arrangement, and the frequency offset degree reflects the difference between the interval center corresponding to the frequency segmentation ratio of each data point and the original frequency value of the data point; select the segmentation result of the arrangement corresponding to the maximum frequency offset degree as the target segmentation result.

[0082] It should be noted that in order to make the difference between the frequency of each data point after encryption and the frequency before encryption larger, the intervals need to be divided so that the intervals of the data points are farther away from the frequency before encryption. Based on this, the interval division control is performed to obtain the intervals of the data points.

[0083] Preferably, as an example, the frequency segmentation ratios of all data points at a moment are randomly arranged, and the frequency intervals are segmented according to the arrangement order of any arrangement result to obtain the segmentation result of that arrangement; the frequency offset degree of each arrangement segmentation result is analyzed, and the frequency offset degree reflects the difference between the center of the interval corresponding to the frequency segmentation ratio of each data point and the original frequency value of the data point; and the segmentation result of the arrangement corresponding to the maximum frequency offset degree is selected as the target segmentation result, including:

[0084] The frequency division ratios of all data points at a moment are randomly arranged, and according to the order of any arrangement result, intervals of corresponding ratios are intercepted on the frequency interval based on the frequency division ratio to obtain several intervals.

[0085] In the segmentation results of any arrangement, the frequency value at the center point of each interval is obtained, and the absolute value of the difference between the frequency value at the center point of the data point and the interval corresponding to the frequency segmentation ratio is recorded as the frequency deviation value of the data point; the voiceprint representation degree of the data point is used as the weight, and the frequency deviation values of all data points at a moment are weighted and summed to obtain the frequency deviation degree of the segmentation result of this arrangement combination.

[0086] The segmentation result of the arrangement corresponding to the maximum frequency shift degree is selected as the target segmentation result.

[0087] It is understandable that by selecting the segmentation result with the largest frequency shift as the target segmentation result, the encrypted data may be significantly different from the original data, thereby increasing the difficulty of decryption.

[0088] S33: Randomly select an integer in each interval of the target segmentation result as the frequency encryption data of the corresponding data point.

[0089] It is understandable that encrypting by randomly selecting data rather than using mapping relationship rules can prevent the mapping relationship rules from exposing voiceprint information and improve encryption accuracy.

[0090] S4: Complete the amplitude encryption of each data point to achieve user voice signal encryption.

[0091] It should be noted that in order to prevent the amplitude information of the data point from exposing the voiceprint information, the amplitude of each data point needs to be encrypted.

[0092] Preferably, as an example, completing the amplitude encryption of each data point to implement user voice signal encryption includes:

[0093] It should be noted that people's voices generally have a certain amplitude range, so the common amplitude range of people's voices can be used as the amplitude interval.

[0094] Optionally, obtain a pre-obtained amplitude range, including:

[0095] The amplitude range of human voices is used as the pre-obtained amplitude range. For example, the amplitude range of human voices is 20–85 Pa.

[0096] The voiceprint representation level of a data point is divided by the cumulative sum of the voiceprint representation levels of all data points at a given moment, and the resulting quotient is used as the amplitude segmentation ratio for the data point. Based on the amplitude segmentation ratio and amplitude interval, the target segmentation result for the amplitude interval is obtained in the same manner as the target segmentation result for the frequency interval. An integer is randomly selected from each interval of the target segmentation result as the amplitude encryption data for the corresponding data point. The spectrogram composed of the frequency-encrypted data and the amplitude-encrypted data is used as the encrypted spectrogram, and the encrypted spectrogram is inversely transformed to obtain the encrypted signal of the user's voice signal.

[0097] It should be noted that the method of obtaining a signal by inverse transforming the encrypted spectrogram is an existing technology and will not be described in detail here.

[0098] It can be understood that the above encryption sets different interval ranges according to whether the data points contain voiceprint features, so that the encrypted data of data points containing more voiceprint feature information has more data selection types, improves the randomness of the encrypted data, and strengthens the encryption effect.

[0099] It should be supplemented that the key acquisition method includes: taking the difference between the frequency of the data point and the frequency at the center point of the interval corresponding to the frequency division ratio as the frequency offset value.

[0100] The amplitude deviation value of the data point is obtained in the same way.

[0101] Sort the voiceprint representation degrees of all data points at a moment in time according to the segmentation order of the target frequency segmentation result to obtain a voiceprint representation degree sequence. Similarly, sort the frequency offset values of all data points at a moment in time to obtain a frequency offset value sequence. The frequency offset value sequence is concatenated after the voiceprint representation degree sequence to obtain the frequency key at a moment.

[0102] The amplitude key at a moment is obtained in the same way.

[0103] S5: Perform decryption processing.

[0104] Preferably, as an example, the decryption process includes:

[0105] Performing short-time Fourier transform on the encrypted signal of the user voice signal to obtain an encrypted spectrum graph;

[0106] It should be noted that the time window length of the short-time Fourier transform is a length agreed upon in advance by the encryptor and decryptor, and the time window lengths of all encryption and decryption short-time Fourier transforms are the same.

[0107] Obtain the frequency key at each moment, and decompose the voiceprint characterization degree sequence and frequency offset value sequence in the frequency key. Calculate the frequency division ratio according to each voiceprint characterization degree in the voiceprint characterization degree sequence, and divide the frequency interval according to the frequency division ratio and the corresponding sorting method of the voiceprint characterization degree; obtain the frequency interval of each data point in the encrypted spectrum in the segmentation result and record it as the analysis interval of each data point. According to the order of the voiceprint characterization degree corresponding to the analysis interval in the voiceprint characterization degree sequence, obtain the frequency offset value corresponding to the analysis interval in the frequency offset value sequence, add the frequency offset value to the center point of the analysis interval, and use the obtained sum as the original frequency value of the data point corresponding to the analysis interval. In the same way, decrypt the original amplitude of the data point according to the amplitude key.

[0108] The spectrogram composed of the original amplitude and original frequency values of all data points is inversely transformed to obtain the original speech signal.

[0109] An embodiment of the present invention further discloses a user voice data security encryption system, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a user voice data security encryption method according to the present invention is implemented.

[0110] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0111] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as resistive random access memory, dynamic random access memory, static random access memory, enhanced dynamic random access memory, high bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store the required information and can be accessed by an application, module, or both. Any such computer storage medium may be part of, accessible to, or connectable to the device.

Claims

1. A method for securely encrypting user voice data, characterized in that: Including steps: Obtain user voice signals; Perform short-time Fourier transform on the user's speech signal to obtain a spectrogram; Calculating the degree of voiceprint representation of each data point in the spectrogram, including: obtaining the resonance peak area, harmonic frequency and change model of each data point in the spectrogram; if the data point belongs to the resonance peak area of the spectrogram, setting the peak distance of the data point to a preset first anti-zero value; otherwise, connecting the data point with the geometric center of the resonance peak area, and taking the distance between the data point and the intersection of the connection line as the peak distance; adding the difference between the frequency of the data point and the harmonic frequency with the smallest frequency difference plus a preset second anti-zero value to obtain the harmonic difference; obtaining the change law reference model of the data point, adding the difference between the change model of the data point and the change law reference model plus a preset third anti-zero value to obtain the change law difference; calculating the degree of voiceprint representation based on the peak distance, harmonic difference and change law difference of the data point, wherein the voiceprint representation degree is negatively correlated with the peak distance, harmonic difference and change law difference; Obtain a pre-obtained frequency interval; set a frequency segmentation ratio for each data point based on the voiceprint characterization degree, wherein the frequency segmentation ratio of the data point is positively correlated with the voiceprint characterization degree; randomly arrange the frequency segmentation ratios of all data points at a moment, and segment the frequency interval according to the order of any arrangement result to obtain a segmentation result of that arrangement; analyze the frequency offset degree of each segmentation result, wherein the frequency offset degree reflects the difference between the center of the interval corresponding to the frequency segmentation ratio of each data point and the original frequency value of the data point; select the segmentation result of the arrangement corresponding to the maximum frequency offset degree as the target segmentation result; and randomly select an integer in each interval of the target segmentation result as the frequency encryption data of the corresponding data point; Complete the amplitude encryption of each data point to achieve user voice signal encryption.

2. A user voice data security encryption method according to claim 1, characterized in that: The method for obtaining the change model includes: A preset area is obtained with each data point as the center, a Gaussian mixture model is constructed based on all data points in the preset area, and the Gaussian mixture model of the preset area is used as the change model of each data point.

3. A user voice data security encryption method according to claim 2, characterized in that: The method for obtaining the variation law reference model includes: The similarity of the change model of each two data points is obtained and recorded as the change similarity. The change similarities of all two data points are clustered into two categories. The data points corresponding to the change similarity in the category with the largest change similarity mean are recorded as regular reference data points. The regular reference data points are clustered according to the change similarity, and the regular reference data points in a category are recorded as regular data points of the same type. The mean value of the corresponding parameter of the change model of the regular data points of the same type is recorded as the comprehensive parameter, and the change model composed of the comprehensive parameters is used as the regular model. The regular model with the smallest difference from the variation model of the data point is obtained as the reference model of the variation pattern of the data point.

4. A user voice data security encryption method according to claim 1, characterized in that: The step of setting the frequency division ratio of each data point according to the voiceprint representation degree includes: The voiceprint representation degree of the data point is divided by the cumulative sum of the voiceprint representation degrees of all data points at a moment, and the obtained quotient is used as the frequency division ratio of the data point.

5. A user voice data security encryption method according to claim 1, characterized in that: The segmentation process of the frequency interval according to the arrangement order of any arrangement result to obtain the segmentation result of the arrangement includes: According to the arrangement order of any arrangement result, intervals of corresponding proportions are intercepted from the frequency interval based on the frequency division ratio in sequence to obtain a plurality of intervals.

6. A user voice data security encryption method according to claim 1, characterized in that: The analyzing the frequency shift degree of the segmentation results of each arrangement includes: In the segmentation results of any arrangement, the frequency value at the center point of each interval is obtained, and the absolute value of the difference between the frequency value at the center point of the data point and the interval corresponding to the frequency segmentation ratio is recorded as the frequency deviation value of the data point; the voiceprint representation degree of the data point is used as the weight, and the frequency deviation values of all data points at a moment are weighted and summed to obtain the frequency deviation degree of the segmentation result of this arrangement.

7. A user voice data security encryption method according to claim 1, characterized in that: The step of completing the amplitude encryption of each data point includes: Get the pre-obtained amplitude interval; According to the method of obtaining the target segmentation result of the frequency interval, the target segmentation result of the amplitude interval is obtained, and an integer is randomly selected in each interval of the target segmentation result of the amplitude interval as the amplitude encrypted data of the corresponding data point.

8. A user voice data security encryption method according to claim 1, characterized in that: The method of realizing user voice signal encryption includes: The spectrogram composed of the frequency encrypted data and the amplitude encrypted data is used as the encrypted spectrogram, and the encrypted spectrogram is inversely transformed to obtain the encrypted signal of the user voice signal.

9. A user voice data security encryption system, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a user voice data security encryption method according to any one of claims 1 to 8 is implemented.

Images