A voiceprint recognition-based secure communication system and method

Abstract

The application relates to a security communication system and method based on voiceprint recognition, belonging to the technical field of communication security, which mainly sends a verification password to a user and a verification server through a password server, simultaneously requires the user to input voice data including a registration password and the verification password, then divides the voice data based on the number of characters of the known registration password by using endpoint detection, extracts voiceprint of the registration password audio data obtained by division to obtain input voiceprint features, performs voice recognition on the verification password audio data obtained by division to obtain input password content, and verifies the two by the registration password and the verification password respectively, so that the security check of communication is realized. For the user in the application, not only the correct user voice can be input, but also the correct registration password can be spoken to ensure the correctness of voiceprint recognition, and the correct verification password can be input to pass the user verification, so that the security is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of communication security technology, and in particular to a secure communication system and method based on voiceprint recognition. Background Technology

[0002] Communication security is becoming increasingly important in the modern information age. With technological advancements, people rely more and more on various communication methods for communication and exchange, and also depend on software clients to communicate with other devices or servers to complete tasks. However, this also makes communication channels targets for hackers and criminals. To protect personal privacy and business secrets, and to ensure the security and confidentiality of information, communication security is of paramount importance.

[0003] Voiceprint recognition technology is a biometric identification technology based on speech signal analysis. It establishes and identifies an individual's voiceprint model by analyzing their speech characteristics, such as frequency, pitch, and resonance. Voiceprint recognition offers advantages such as being non-contact, non-invasive, and real-time, with high real-time performance enabling rapid identification within a short timeframe. It is also unaffected by environment or language, allowing for accurate identification in various language and sound environments. These advantages make voiceprint recognition a crucial biometric identification technology with broad application prospects in the security field.

[0004] Therefore, how to improve communication security based on voiceprint recognition technology is a problem that people hope to solve. Summary of the Invention

[0005] Therefore, the present invention provides a secure communication system and method based on voiceprint recognition to solve the problem of how to improve communication security based on voiceprint recognition technology.

[0006] This invention provides a secure communication system based on voiceprint recognition, including a password server and a verification server, wherein the password server and the verification server are communicatively connected, and: The password server is used to send randomly generated verification passwords to users and the verification server; The verification server is used for: Obtain the user's voice input data, which includes the registration password and verification password; Endpoint detection is performed on the voice data, and the voice data is segmented based on the endpoint detection results and the number of characters in the registration password to obtain the registration password audio data and the verification password audio data. Voiceprint recognition is performed on the registration password audio data to obtain the input voiceprint features; The audio data of the verification password is used for speech recognition to obtain the input password content; The system compares the input voiceprint features with the voiceprint verification information pre-registered by the user in the verification server, compares the input password content with the verification password, and establishes a communication connection for the user based on the comparison results.

[0007] In a preferred implementation, the step of performing endpoint detection on the voice data and segmenting the voice data based on the endpoint detection results and the number of characters in the registration password to obtain registration password audio data and verification password audio data includes: Get the number of characters in the registration password; Endpoint detection is performed on speech data based on the spectral entropy method to obtain the start and end times of each character in the speech data; Based on the start and end times of each character in the voice data, and the number of characters in the registration password, the segmentation time points of the voice data are obtained; Based on the start and end times of each character in the voice data, and combined with the segmentation time points of the voice data, the voice data is segmented to obtain the registration password audio data and the verification password audio data.

[0008] In a preferred implementation, the registration password in the voice data precedes the verification password; the endpoint detection of the voice data based on the spectral entropy method to obtain the start and end times of each character in the voice data includes: The speech data is segmented, windowed, and subjected to Fourier transform to obtain multiple speech frames, and the energy probability density of each sampling point in each speech frame is calculated. Based on the energy probability density of each sampling point in each speech frame, the spectral entropy value of the speech frame is calculated sequentially according to the first preset spectral entropy calculation function, and the speech frame corresponding to the spectral entropy value is determined to be the first active frame according to the relationship between the spectral entropy value and the first preset threshold. Based on the continuity relationship between multiple known first active frames, identify the start and end times of the characters corresponding to the registration password in the voice data, and count the number of known characters; If the number of characters is known to exceed the number of characters in the registration password, then based on the energy probability density of each sampling point in each speech frame, the spectral entropy value of the speech frame is calculated sequentially according to the second preset spectral entropy calculation function. Based on the relationship between the spectral entropy value and the second preset threshold, it is determined whether the speech frame corresponding to the spectral entropy value is the second active frame. The computer operation complexity of the second preset spectral entropy calculation function is lower than that of the first preset spectral entropy calculation function. Based on the continuity relationship between multiple known second active frames, identify the start and end times of the characters corresponding to the verification password in the voice data.

[0009] In a preferred implementation, the first preset spectral entropy calculation function is:

[0010] In the functions above, The spectral entropy value calculated by the first preset spectral entropy calculation function. Indicates the first in the speech frame One sampling point, This represents the total number of sampling points in the speech frame. For the first speech frame Energy probability density of each sampling point.

[0011] In a preferred implementation, the step of performing voiceprint recognition on the registration password audio data to obtain input voiceprint features includes: Based on the registration password, the audio of the registration password is identified using a voiceprint recognition algorithm based on text-related types to obtain the input voiceprint features.

[0012] In a preferred implementation, the step of performing speech recognition on the verification password audio data to obtain the input password content includes: The number of characters in the input verification password is obtained based on the start and end times of the characters corresponding to the verification password in the voice data. Feature extraction is performed on the audio data of the verification password to obtain feature data. The input data is obtained by combining the feature data and the number of characters in the input verification password. Input data is fed into a preset neural network speech recognition model to obtain the input password.

[0013] In a preferred implementation, the step of extracting features from the verification password audio data to obtain feature data, and combining the feature data with the number of characters in the input verification password to obtain input data, includes: The audio data for verifying the password is divided into multiple audio segments in the time domain, with the number of audio segments being the same as the number of characters in the input verification password. MFCC feature extraction is performed on each speech segment to obtain the waveform feature vector of each speech segment; The input matrix is ​​obtained as input data based on the waveform feature vector of each speech segment.

[0014] In a preferred implementation, the steps of comparing the input voiceprint features with the user's pre-registered voiceprint verification information in the verification server, comparing the input password content with the verification password, and establishing a communication connection for the user based on the comparison results include: The first similarity is obtained by comparing the input voiceprint features with the voiceprint verification information pre-registered by the user in the verification server. Compare the input password and the verification password to obtain the second similarity score; Based on the preset similarity calculation formula, the overall similarity is calculated according to the first similarity and the second similarity. Compare the overall similarity with a preset similarity threshold; If the overall similarity exceeds the preset similarity threshold, the user verification is deemed successful and a communication connection with the user is established. If the overall similarity does not exceed the preset similarity threshold, the user verification is deemed to have failed. The preset similarity calculation formula is as follows:

[0015] in, For overall similarity, The first similarity score, For the second similarity, and They are weighted differently. Greater than , and These are different variable control parameters, Greater than .

[0016] The present invention also provides a secure communication method based on voiceprint recognition, comprising: The password server is used to send randomly generated verification passwords to users and the verification server; The system obtains the user's voice input data through a verification server. The voice data includes the registration password and the verification password. Endpoint detection is performed on the voice data, and the voice data is segmented based on the endpoint detection results and the number of characters in the registration password to obtain the registration password audio data and the verification password audio data. Voiceprint recognition is performed on the registration password audio data to obtain the input voiceprint features; The audio data of the verification password is used for speech recognition to obtain the input password content; The system compares the input voiceprint features with the voiceprint verification information pre-registered by the user in the verification server, compares the input password content with the verification password, and establishes a communication connection for the user based on the comparison results.

[0017] The beneficial effects of using the above embodiments are: This invention provides a secure communication system and method based on voiceprint recognition. It primarily involves a password server sending verification passwords to both the user and the verification server, while simultaneously requiring the user to input voice data including a registration password and a verification password. Endpoint detection is then used to segment the voice data based on the known character count of the registration password. Voiceprint extraction is performed on the segmented registration password audio data to obtain input voiceprint features, and speech recognition is performed on the segmented verification password audio data to obtain the input password content. Finally, the registration password and verification password are used to verify both, thereby achieving secure communication verification. For the user in this invention, they must not only input the correct user voice but also the correct registration password to ensure accurate voiceprint recognition, and further, the correct verification password to pass user verification, significantly improving security. Furthermore, while endpoint detection is typically used as a preprocessing step for voice data in existing technologies, this invention cleverly utilizes the endpoint detection results, using the character count of the registration password as a reference, to segment the voice data into the registration password and verification password parts, reducing the overall complexity of the solution. Attached Figure Description

[0018] Figure 1 The system architecture diagram of the secure communication system based on voiceprint recognition provided by the present invention; Figure 2 for Figure 1 Flowchart of the method executed by the verification server; Figure 3 for Figure 2 Flowchart of the method for step S202; Figure 4 for Figure 3 Flowchart of step S302; Figure 5 for Figure 2 The flowchart of step S204. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Combination Figure 1 and Figure 2As shown in the figure, a specific embodiment of the present invention discloses a secure communication system based on voiceprint recognition, including a password server 100 and a verification server 200, wherein the password server and the verification server are communicatively connected, and: Password server 100 is used to send randomly generated verification passwords to user 300 and verification server 200; Verification server 200 is used for: S201. Obtain the voice data input by the user. The voice data includes the registration password and the verification password. S202. Perform endpoint detection on the voice data, and segment the voice data according to the endpoint detection results and the number of characters in the registration password to obtain the registration password audio data and the verification password audio data. S203. Perform voiceprint recognition on the registered password audio data to obtain the input voiceprint features; S204. Perform speech recognition on the audio data of the verification password to obtain the input password content; S205. Compare the input voiceprint features with the voiceprint verification information pre-registered by the user in the verification server, compare the input password content with the verification password, and establish a communication connection for the user based on the comparison results.

[0021] This invention provides a secure communication system and method based on voiceprint recognition. It primarily involves a password server sending verification passwords to both the user and the verification server, while simultaneously requiring the user to input voice data including a registration password and a verification password. Endpoint detection is then used to segment the voice data based on the known character count of the registration password. Voiceprint extraction is performed on the segmented registration password audio data to obtain input voiceprint features, and speech recognition is performed on the segmented verification password audio data to obtain the input password content. Finally, the registration password and verification password are used to verify both, thereby achieving secure communication verification. For the user in this invention, they must not only input the correct user voice but also the correct registration password to ensure accurate voiceprint recognition, and further, the correct verification password to pass user verification, significantly improving security. Furthermore, while endpoint detection is typically used as a preprocessing step for voice data in existing technologies, this invention cleverly utilizes the endpoint detection results, using the character count of the registration password as a reference, to segment the voice data into the registration password and verification password parts, reducing the overall complexity of the solution.

[0022] The registration password is the password that the user pre-registers with the verification server, and the voiceprint verification information pre-registered by the user with the verification server is also extracted based on the registration password. The verification password, on the other hand, is a dynamically generated password that is randomly generated in real time. The registration password audio data is the content related to the registration password extracted from the speech data after endpoint detection. Similarly, the verification password audio data is the content related to the verification password extracted from the speech data after endpoint detection. The input voiceprint feature is the voiceprint feature of the user's actual input voice, and the input password content is the content of the verification password actually entered by the user.

[0023] The present invention also provides a more detailed embodiment to clearly illustrate the above process: For example, the user's voice input could be "blue sky and white clouds 2345", where "blue sky and white clouds" is the registration password the user pre-registered with the verification server, and voiceprint verification information is pre-registered with the verification server based on this password. "2345" is a randomly generated verification password that can be sent to the user as a verification code. In addition to inputting the correct voice, the user needs to know both "blue sky and white clouds" and "2345" to pass the verification.

[0024] The user's local client requires no processing beyond sending and receiving data, offering flexibility and convenience. For the verification server, it only needs to identify the position of each character in each voice data segment through endpoint detection. After pre-defining the voice data structure, it can segment the voice data simply by counting the number of characters in the registration password. For example, if the phrase "blue sky and white clouds" has four characters, the verification server can directly use the last time of the fourth character as the segmentation point after endpoint detection, accurately extracting the registration and verification password portions of the voice data. Of course, the voice data could also be "2345 blue sky and white clouds," which the user needs to agree upon beforehand with the verification server (this can be determined during software development).

[0025] Specifically, in combination Figure 3 As shown, in a preferred embodiment, step S202, which involves performing endpoint detection on the voice data and segmenting the voice data based on the endpoint detection results and the number of characters in the registration password, to obtain registration password audio data and verification password audio data, specifically includes: S301. Obtain the number of characters in the registration password; S302. Based on the spectral entropy method, perform endpoint detection on the speech data to obtain the start and end times of each character in the speech data; S303. Based on the start and end times of each character in the voice data, and the number of characters in the registration password, obtain the segmentation time points of the voice data; S304. Based on the start and end times of each character in the voice data, and combined with the segmentation time points of the voice data, segment the voice data to obtain the registration password audio data and the verification password audio data.

[0026] Spectral entropy is a commonly used endpoint detection method that analyzes the spectrum of a speech signal to determine the boundary between speech and non-speech activities. In spectral entropy, the entropy value on the spectrum is calculated to measure the complexity and information content of the spectrum. During speech activity, the spectrum of the speech signal fluctuates significantly, resulting in a higher entropy value; conversely, during non-speech activity, the spectrum changes less, leading to a lower entropy value. Based on the changes in entropy value, the start and end points of speech can be identified.

[0027] Spectral entropy has several important advantages. First, it exhibits high robustness, accurately detecting the start and end points of speech even in noisy environments or with poor speech quality. This makes it highly adaptable and reliable in practical applications. Second, spectral entropy is computationally fast, meeting real-time requirements and suitable for speech communication and speech recognition scenarios demanding rapid response. Furthermore, it is applicable to various speech materials and languages, demonstrating broad applicability. It is not limited by specific speech features and can adapt to different speech signal characteristics. Finally, spectral entropy offers parameter adjustability, allowing for parameter adjustments based on specific application scenarios to achieve better detection results and accuracy.

[0028] It should be noted that, in addition to segmenting a voice data into two parts—the registration password and the verification password—the above process also preprocesses endpoint detection by determining the start and end times of each character. This includes removing silence before and after the start of speech, and removing the blank space between the end of one character's speech and the beginning of the next. This reduces noise, decreases data volume, and improves recognition accuracy. Furthermore, the registration and verification password audio data obtained in this step are processed by endpoint detection. In subsequent voiceprint and speech recognition processes, preprocessing steps such as framing and windowing can be repeated, but endpoint detection is unnecessary, reducing redundancy and improving efficiency.

[0029] Furthermore, in combination Figure 4 As shown, in a preferred embodiment, the registration password in the voice data precedes the verification password; the above step S302, performing endpoint detection on the voice data based on the spectral entropy method to obtain the start and end times of each character in the voice data, specifically includes: S401. Perform frame segmentation, windowing, and Fourier transform on the speech data to obtain multiple speech frames, and calculate the energy probability density of each sampling point in each speech frame. S402. Based on the energy probability density of each sampling point in each speech frame, calculate the spectral entropy value of the speech frame in sequence according to the first preset spectral entropy calculation function, and determine whether the speech frame corresponding to the spectral entropy value is the first active frame according to the relationship between the spectral entropy value and the first preset threshold. S403. Based on the continuous relationship between multiple known first active frames, identify the start and end times of the characters corresponding to the registration password in the voice data, and count the number of known characters; S404. If the number of characters is known to exceed the number of characters in the registration password, then based on the energy probability density of each sampling point in each speech frame, the spectral entropy value of the speech frame is calculated sequentially according to the second preset spectral entropy calculation function. Based on the relationship between the spectral entropy value and the second preset threshold, it is determined whether the speech frame corresponding to the spectral entropy value is the second active frame. The computer operation complexity of the second preset spectral entropy calculation function is lower than that of the first preset spectral entropy calculation function. S405. Based on the continuity relationship between multiple known second active frames, identify the start and end times of the characters corresponding to the verification password in the voice data.

[0030] The above process is an improvement on the existing spectral entropy method. In this embodiment, the registration password precedes the verification password in the voice data. Therefore, the spectral entropy method first calculates the spectral entropy of the voice frame corresponding to the characters in the registration password. In this embodiment, only the registration password portion needs voiceprint detection, requiring precise segmentation of the voice's starting point to ensure correct extraction of the input voiceprint features. For the portion including the verification password, a more robust speech recognition model can be used to extract the text content and obtain the input password. Therefore, while calculating the spectral entropy value using the first preset spectral entropy calculation function, this embodiment also counts the number of detected characters in real time. If the number of characters exceeds the number of characters in the registration password, the voice frame for which the spectral entropy value is to be calculated is considered to belong to the verification password portion of the voice data. In this case, the less complex second preset spectral entropy calculation function can be used to calculate the spectral entropy value. Although this may sacrifice some accuracy, it results in faster processing speed and less computational pressure.

[0031] Specifically, in a preferred embodiment, the first preset spectral entropy calculation function is:

[0032] In the above functions, The spectral entropy value calculated by the first preset spectral entropy calculation function. Indicates the first in the speech frame One sampling point, This represents the total number of sampling points in the speech frame. For the first speech frame Energy probability density of each sampling point.

[0033] Compared to the first preset spectral entropy function mentioned above, the second preset spectral entropy function approximates the complex logarithmic function with a simple linear function, thereby reducing the computational burden on the computer.

[0034] Furthermore, in a preferred embodiment, step S203, performing voiceprint recognition on the registered password audio data to obtain input voiceprint features, specifically includes: Based on the registration password, the audio of the registration password is identified using a voiceprint recognition algorithm based on text-related types to obtain the input voiceprint features.

[0035] Voiceprint recognition algorithms can be divided into two categories: text-related and text-independent. Text-related voiceprint recognition algorithms are based on the features of specific speech content. These algorithms can extract voiceprint features relatively accurately, but obviously require the user's pronunciation to match the preset text, making them inflexible in practice. Text-independent voiceprint recognition algorithms, while not dependent on speech content, are affected by factors such as environment, the speaker's emotional state, and speaking style, making it relatively difficult to build a recognition model and resulting in lower feature extraction accuracy.

[0036] In this embodiment, the registration password is a known text that exists as a key. Therefore, using a text-related algorithm for voiceprint extraction is clearly more appropriate and can improve the accuracy of voiceprint recognition. As for the verification password, since it is randomly generated, this embodiment uses speech recognition to extract the content rather than voiceprint recognition, eliminating the shortcomings of text-independent voiceprint recognition algorithms and improving the feasibility and reliability of the entire solution.

[0037] Furthermore, in combination Figure 5 As shown, in a preferred embodiment, step S204, performing speech recognition on the verification password audio data to obtain the input password content, specifically includes: S501. Based on the start and end times of the characters corresponding to the verification password in the voice data, obtain the number of characters in the input verification password; S502. Extract features from the audio data of the verification password to obtain feature data, and combine the feature data with the number of characters in the input verification password to obtain the input data; S503. Input the input data into the preset neural network speech recognition model to obtain the input password content.

[0038] The above process further utilizes the processing results of the endpoint detection preprocessing step. Based on the start and end times of the detected characters, the number of input verification password characters in the verification password audio data is determined. The number of input verification password characters and feature data are then used as known information input data and input into the preset neural network speech recognition model to improve the speed and accuracy of speech recognition.

[0039] The number of characters in the input verification password can be displayed and recorded in the input data, serving as a known parameter in the calculation of the preset neural network speech recognition model. Alternatively, it can be implicitly incorporated into the input data as a constraint on the input data format. For example, in a preferred embodiment, step S502, which involves extracting features from the verification password audio data to obtain feature data, and combining the feature data with the number of characters in the input verification password to obtain the input data, specifically includes: The audio data for verifying the password is divided into multiple audio segments in the time domain, with the number of audio segments being the same as the number of characters in the input verification password. MFCC feature extraction is performed on each speech segment to obtain the waveform feature vector of each speech segment; The input matrix is ​​obtained as input data based on the waveform feature vector of each speech segment.

[0040] Assuming each waveform feature vector is a row vector in the input matrix, then the number of rows in the input matrix is ​​the number of characters in the input verification password. Although the input matrix does not explicitly record the number of characters in the input verification password, it implicitly contains this information.

[0041] The present invention also provides a more detailed embodiment to clearly illustrate step S502 above: Assuming the user's speech rate is stable and uniform during voice input, and endpoint detection indicates that the verification password audio data to be recognized contains 10 characters, the verification password audio data can be divided into 10 small speech segments, and MFCC features (Mel frequency cepstral coefficients, which are existing techniques that can be understood by those skilled in the art and will not be explained in detail here) can be extracted from each segment.

[0042] For each speech segment, a fixed time window (e.g., 25 milliseconds) can be selected, and MFCC feature extraction is performed using the sound waveform within this time window. Assuming the sampling rate of the speech segment is 16kHz, then a 25-millisecond time window corresponds to 400 sampling points. Therefore, for each speech segment, the input data can be a waveform feature vector of size 400. For the entire verification password audio data, the input data is an input matrix composed of multiple waveform feature vectors, with each input matrix containing 10 waveform feature vectors.

[0043] Furthermore, in a preferred embodiment, step S205, comparing the input voiceprint features with the voiceprint verification information pre-registered by the user in the verification server, comparing the input password content with the verification password, and establishing a communication connection for the user based on the comparison result, specifically includes: The first similarity is obtained by comparing the input voiceprint features with the voiceprint verification information pre-registered by the user in the verification server. Compare the input password and the verification password to obtain the second similarity score; Based on the preset similarity calculation formula, the overall similarity is calculated according to the first similarity and the second similarity. Compare the overall similarity with a preset similarity threshold; If the overall similarity exceeds the preset similarity threshold, the user verification is deemed successful and a communication connection with the user is established. If the overall similarity does not exceed the preset similarity threshold, the user verification is deemed to have failed. The preset similarity calculation formula is as follows:

[0044] in, For overall similarity, The first similarity score, For the second similarity, and They are weighted differently. Greater than , and These are different variable control parameters, Greater than .

[0045] In the above formula, the first similarity, which represents the similarity between the input voiceprint features and the pre-registered voiceprint verification information, is taken as the main determining factor, while the second similarity is used as a secondary factor. By correctly setting the weights and controlling the variation parameters, the following effect can be achieved: When the first similarity is high, it indicates that the user's voice is accurate and the probability of it being the user is extremely high, so the overall similarity will also increase. At this time, the higher the second similarity, the more it indicates that the user has received the correct verification password, and the overall similarity will also increase, allowing the user to pass the verification. Conversely, if the second similarity is low, it indicates that the user may not have received the verification password, and the overall similarity will be appropriately reduced. However, this situation may be due to signal delay, noise, or data transmission errors. The overall similarity calculated by the above formula will not be too low, so the user still has a chance to pass the verification.

[0046] When the first similarity score is low, it indicates that the user's voice is inaccurate and the probability that the user is not the actual person is extremely high. At this time, the risk is high and should be handled with caution. The calculated overall similarity score will be low. Even if the second similarity score is high, the calculated overall similarity score may not reach a high level. This prevents criminals from passing the verification by stealing the verification password.

[0047] Furthermore, the present invention also provides a secure communication method based on voiceprint recognition, comprising: The password server is used to send randomly generated verification passwords to users and the verification server; The system obtains the user's voice input data through a verification server. The voice data includes the registration password and the verification password. Endpoint detection is performed on the voice data, and the voice data is segmented based on the endpoint detection results and the number of characters in the registration password to obtain the registration password audio data and the verification password audio data. Voiceprint recognition is performed on the registration password audio data to obtain the input voiceprint features; The audio data of the verification password is used for speech recognition to obtain the input password content; The system compares the input voiceprint features with the voiceprint verification information pre-registered by the user in the verification server, compares the input password content with the verification password, and establishes a communication connection for the user based on the comparison results.

[0048] The principles and beneficial effects of the above methods can be found in the previous description, and will not be explained further here.

[0049] This invention provides a secure communication system and method based on voiceprint recognition. It primarily involves a password server sending verification passwords to both the user and the verification server, while simultaneously requiring the user to input voice data including a registration password and a verification password. Endpoint detection is then used to segment the voice data based on the known character count of the registration password. Voiceprint extraction is performed on the segmented registration password audio data to obtain input voiceprint features, and speech recognition is performed on the segmented verification password audio data to obtain the input password content. Finally, the registration password and verification password are used to verify both, thereby achieving secure communication verification. For the user in this invention, they must not only input the correct user voice but also the correct registration password to ensure accurate voiceprint recognition, and further, the correct verification password to pass user verification, significantly improving security. Furthermore, while endpoint detection is typically used as a preprocessing step for voice data in existing technologies, this invention cleverly utilizes the endpoint detection results, using the character count of the registration password as a reference, to segment the voice data into the registration password and verification password parts, reducing the overall complexity of the solution.

[0050] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0051] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A secure communication system based on voiceprint recognition, characterized in that, This includes a password server and a verification server, and the communication connection between the password server and the verification server, wherein: The password server is used to send randomly generated verification passwords to users and the verification server; The verification server is used for: Obtain the user's voice input data, which includes the registration password and verification password; Endpoint detection is performed on the voice data, and the voice data is segmented based on the endpoint detection results and the number of characters in the registration password to obtain the registration password audio data and the verification password audio data. Voiceprint recognition is performed on the registration password audio data to obtain the input voiceprint features; The audio data of the verification password is used for speech recognition to obtain the input password content; Compare the input voiceprint features with the voiceprint verification information pre-registered by the user in the verification server, compare the input password content with the verification password, and establish a communication connection for the user based on the comparison results; This process involves endpoint detection of the voice data, followed by segmentation of the voice data based on the endpoint detection results and the number of characters in the registration password, resulting in registration password audio data and verification password audio data, including: Get the number of characters in the registration password; Endpoint detection is performed on speech data based on the spectral entropy method to obtain the start and end times of each character in the speech data; Based on the start and end times of each character in the voice data, and the number of characters in the registration password, the segmentation time points of the voice data are obtained; Based on the start and end times of each character in the voice data, and combined with the segmentation time points of the voice data, the voice data is segmented to obtain the registration password audio data and the verification password audio data. In this context, the registration password in the voice data precedes the verification password; the endpoint detection of the voice data based on the spectral entropy method to obtain the start and end times of each character in the voice data includes: The speech data is segmented, windowed, and subjected to Fourier transform to obtain multiple speech frames, and the energy probability density of each sampling point in each speech frame is calculated. Based on the energy probability density of each sampling point in each speech frame, the spectral entropy value of the speech frame is calculated sequentially according to the first preset spectral entropy calculation function, and the speech frame corresponding to the spectral entropy value is determined to be the first active frame according to the relationship between the spectral entropy value and the first preset threshold. Based on the continuity relationship between multiple known first active frames, identify the start and end times of the characters corresponding to the registration password in the voice data, and count the number of known characters; If the number of characters is known to exceed the number of characters in the registration password, then based on the energy probability density of each sampling point in each speech frame, the spectral entropy value of the speech frame is calculated sequentially according to the second preset spectral entropy calculation function. Based on the relationship between the spectral entropy value and the second preset threshold, it is determined whether the speech frame corresponding to the spectral entropy value is the second active frame. The computer operation complexity of the second preset spectral entropy calculation function is lower than that of the first preset spectral entropy calculation function. Based on the continuity relationship between multiple known second active frames, identify the start and end times of the characters corresponding to the verification password in the voice data.

2. The secure communication system based on voiceprint recognition according to claim 1, characterized in that, The first preset spectral entropy calculation function is: In the above functions, The spectral entropy value calculated by the first preset spectral entropy calculation function. Indicates the first in the speech frame One sampling point, This represents the total number of sampling points in the speech frame. For the first speech frame Energy probability density of each sampling point.

3. The secure communication system based on voiceprint recognition according to claim 1, characterized in that, The process of performing voiceprint recognition on the registered password audio data to obtain input voiceprint features includes: Based on the registration password, the audio of the registration password is identified using a voiceprint recognition algorithm based on text-related types to obtain the input voiceprint features.

4. The secure communication system based on voiceprint recognition according to claim 1, characterized in that, The step of performing speech recognition on the verification password audio data to obtain the input password content includes: The number of characters in the input verification password is obtained based on the start and end times of the characters corresponding to the verification password in the voice data. Feature extraction is performed on the audio data of the verification password to obtain feature data. The input data is obtained by combining the feature data and the number of characters in the input verification password. Input data is fed into a preset neural network speech recognition model to obtain the input password.

5. The secure communication system based on voiceprint recognition according to claim 4, characterized in that, The process of extracting features from the audio data of the verification password to obtain feature data, and combining the feature data with the number of characters in the input verification password to obtain input data, includes: The audio data for verifying the password is divided into multiple audio segments in the time domain, with the number of audio segments being the same as the number of characters in the input verification password. MFCC feature extraction is performed on each speech segment to obtain the waveform feature vector of each speech segment; The input matrix is ​​obtained as input data based on the waveform feature vector of each speech segment.

6. The secure communication system based on voiceprint recognition according to claim 1, characterized in that, The process of comparing the input voiceprint features with the user's pre-registered voiceprint verification information in the verification server, comparing the input password content with the verification password, and establishing a communication connection for the user based on the comparison results includes: The first similarity is obtained by comparing the input voiceprint features with the voiceprint verification information pre-registered by the user in the verification server. Compare the input password and the verification password to obtain the second similarity score; Based on the preset similarity calculation formula, the overall similarity is calculated according to the first similarity and the second similarity. Compare the overall similarity with a preset similarity threshold; If the overall similarity exceeds the preset similarity threshold, the user verification is deemed successful and a communication connection with the user is established. If the overall similarity does not exceed the preset similarity threshold, the user verification is deemed to have failed. The preset similarity calculation formula is as follows: in, For overall similarity, The first similarity score, For the second similarity, and Each with different weights, Greater than , and These are different control parameters for different changes. Greater than .

7. A secure communication method based on voiceprint recognition, characterized in that, include: The password server is used to send randomly generated verification passwords to users and the verification server; The system obtains the user's voice input data through a verification server. The voice data includes the registration password and the verification password. Endpoint detection is performed on the voice data, and the voice data is segmented based on the endpoint detection results and the number of characters in the registration password to obtain the registration password audio data and the verification password audio data. Voiceprint recognition is performed on the registration password audio data to obtain the input voiceprint features; The audio data of the verification password is used for speech recognition to obtain the input password content; Compare the input voiceprint features with the voiceprint verification information pre-registered by the user in the verification server, compare the input password content with the verification password, and establish a communication connection for the user based on the comparison results; This process involves endpoint detection of the voice data, followed by segmentation of the voice data based on the endpoint detection results and the number of characters in the registration password, resulting in registration password audio data and verification password audio data, including: Get the number of characters in the registration password; Endpoint detection is performed on speech data based on the spectral entropy method to obtain the start and end times of each character in the speech data; Based on the start and end times of each character in the voice data, and the number of characters in the registration password, the segmentation time points of the voice data are obtained; Based on the start and end times of each character in the voice data, and combined with the segmentation time points of the voice data, the voice data is segmented to obtain the registration password audio data and the verification password audio data. In this context, the registration password in the voice data precedes the verification password; the endpoint detection of the voice data based on the spectral entropy method to obtain the start and end times of each character in the voice data includes: The speech data is segmented, windowed, and subjected to Fourier transform to obtain multiple speech frames, and the energy probability density of each sampling point in each speech frame is calculated. Based on the energy probability density of each sampling point in each speech frame, the spectral entropy value of the speech frame is calculated sequentially according to the first preset spectral entropy calculation function, and the speech frame corresponding to the spectral entropy value is determined to be the first active frame according to the relationship between the spectral entropy value and the first preset threshold. Based on the continuity relationship between multiple known first active frames, identify the start and end times of the characters corresponding to the registration password in the voice data, and count the number of known characters; If the number of characters is known to exceed the number of characters in the registration password, then based on the energy probability density of each sampling point in each speech frame, the spectral entropy value of the speech frame is calculated sequentially according to the second preset spectral entropy calculation function. Based on the relationship between the spectral entropy value and the second preset threshold, it is determined whether the speech frame corresponding to the spectral entropy value is the second active frame. The computer operation complexity of the second preset spectral entropy calculation function is lower than that of the first preset spectral entropy calculation function. Based on the continuity relationship between multiple known second active frames, identify the start and end times of the characters corresponding to the verification password in the voice data.

Images