Privacy protection voiceprint recognition inference method and system based on full homomorphic encryption technology

Abstract

The application discloses a privacy protection voiceprint recognition inference method based on a full homomorphic encryption technology, and comprises the following steps: collecting a plurality of original voiceprint data from a preset voiceprint database, and forming a voiceprint data set after preprocessing the original voiceprint data; inputting the voiceprint data set into a voiceprint recognition model to be trained, training the voiceprint recognition model to be trained, obtaining a trained voiceprint recognition model, and generating encrypted model parameters; inputting the encrypted model parameters and voiceprint data to be recognized into an encrypter to obtain voiceprint ciphertext data; and calculating and inferring the voiceprint ciphertext data according to the homomorphic encryption characteristics. In this way, the voiceprint recognition model is used for recognizing the voiceprint data to be recognized, so that the calculation accuracy in the full homomorphic encryption environment is improved, resource consumption is reduced, and the inference efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of information security, and particularly relates to a privacy protection voiceprint recognition inference method and system based on fully homomorphic encryption technology. BACKGROUND

[0002] Under the background of rapid development of information technology, voiceprint recognition technology as a kind of biometric recognition technology has gradually attracted widespread attention. Voiceprint recognition can realize identity verification, user identification and security authentication by analyzing the acoustic characteristics of individuals. According to market research, voiceprint recognition technology is increasingly popular in the fields of finance, security, medical treatment and other fields. However, with the increasing awareness of data privacy protection, how to protect the privacy and security of users while carrying out voiceprint recognition has become an important challenge for the development of the technology.

[0003] At present, voiceprint recognition technology mainly relies on deep learning models such as convolutional neural network (CNN) and recurrent neural network (RNN) to process and analyze voiceprint data. These models have made significant progress in recognition accuracy and robustness. However, traditional voiceprint recognition systems usually face the risk of privacy leakage when processing raw voiceprint data, as the voiceprint data of users is vulnerable to attacks during storage and processing, and there is a risk of being stolen or misused. Existing encryption technologies are mostly static encryption, which cannot support direct calculation on encrypted data, resulting in frequent encryption and decryption operations, thereby increasing the complexity and delay of the system, and the data security is insufficient.

[0004] Fully homomorphic encryption technology as a new encryption method allows direct calculation on encrypted data, avoiding the privacy leakage risk caused by data decryption. Therefore, it is a powerful tool for protecting the privacy of voiceprint recognition systems. Many existing neural network architectures based on homomorphic encryption schemes such as BFV, BGV, TFHE and CKKS are mainly applied to convolutional neural networks for encrypted inference to realize privacy protection of deep learning models. Convolution operation is crucial to convolutional neural networks (CNN), but performing convolution operation on ciphertext in homomorphic encryption requires a large amount of computing resources. Each kernel shift requires a smart multiplication and addition operation on elements, which is a resource-intensive operation on ciphertext. Therefore, in the homomorphic encryption environment, the inference process of the deep learning model is usually complex and has large computational overhead, resulting in a significant reduction in inference speed. At the same time, the existing deep learning models often do not consider the characteristics of homomorphic encryption when designed, resulting in poor accuracy and effect when inferring on encrypted data. SUMMARY

[0005] In order to solve the above problems existing in the prior art, the present application provides a privacy protection voiceprint recognition inference method and system based on fully homomorphic encryption technology. The technical problem to be solved by the present application is solved by the following technical scheme:

[0006] A privacy protection voiceprint recognition inference method based on a full homomorphic encryption technology, comprising:

[0007] Collecting a plurality of original voiceprint data from a preset voiceprint database, and forming a voiceprint data set after preprocessing the original voiceprint data;

[0008] Inputting the voiceprint data set into a voiceprint recognition model to be trained, and training the voiceprint recognition model to be trained to obtain a trained voiceprint recognition model to generate encrypted model parameters;

[0009] Inputting the encrypted model parameters and voiceprint data to be recognized into an encrypter to obtain voiceprint ciphertext data;

[0010] According to the homomorphic encryption characteristic, the voiceprint ciphertext data is calculated and inferred.

[0011] In one specific embodiment, after preprocessing the original voiceprint data, a voiceprint data set is formed, comprising:

[0012] Removing the speech silence part in the original voiceprint data to form complete sample data;

[0013] Sequentially performing denoising, feature extraction and standardization processing on the complete sample data;

[0014] Converting the data format and data length of the voiceprint data after standardization processing to form a voiceprint data set to adapt to the homomorphic encryption characteristic.

[0015] In one specific embodiment, the voiceprint recognition model to be trained is trained, comprising:

[0016] Using a stochastic gradient descent optimizer to train the voiceprint recognition model to be trained.

[0017] In one specific embodiment, the encrypter uses an RNS-CKKS full homomorphic encryption algorithm.

[0018] In one specific embodiment, before calculating and inferring the voiceprint ciphertext data according to the homomorphic encryption characteristic, it further comprises:

[0019] Receiving input of a plurality of voiceprint ciphertext sub-data;

[0020] According to the structure of the voiceprint recognition model, calculating the output vector size of the plurality of voiceprint ciphertext sub-data;

[0021] According to the size of the voiceprint ciphertext sub-data, setting the number of slots for storing the voiceprint ciphertext sub-data;

[0022] The encrypted data in each slot is sequentially traversed, and each encrypted data is right-rotated according to a preset rotation bit number, so that the positions of each data in the ciphertext are not overlapped.

[0023] When the total size of the ciphertext after merging is determined to be less than the number of slots, the voiceprint ciphertext data is obtained.

[0024] In one specific embodiment, the preset rotation bit number is (i-1) * N, where i represents the order of the encrypted data, and N represents the number of slots.

[0025] The application also provides a privacy protection voiceprint recognition inference system based on a homomorphic encryption technology, comprising:

[0026] A test set generation module is configured to collect a plurality of original voiceprint data from a preset voiceprint database, and form a voiceprint data set after preprocessing the original voiceprint data.

[0027] A model training module is configured to input the voiceprint data set into a voiceprint recognition model to be trained, and train the voiceprint recognition model to be trained to obtain a trained voiceprint recognition model to generate encrypted model parameters.

[0028] An encryption module is configured to input the encrypted model parameters and voiceprint data to be recognized into an encryptor to obtain voiceprint ciphertext data.

[0029] An inference module is configured to perform calculation and inference on the voiceprint ciphertext data according to the homomorphic encryption characteristics.

[0030] In one specific embodiment, the original voiceprint data is preprocessed to form a voiceprint data set, which comprises:

[0031] The silent part of the original voiceprint data is removed to form complete sample data.

[0032] The complete sample data is sequentially denoised, feature extracted and standardized.

[0033] The voiceprint data after standardization is converted in data format and in data length to form a voiceprint data set to adapt to the homomorphic encryption characteristics.

[0034] In one specific embodiment, the voiceprint recognition model to be trained is trained, which comprises:

[0035] A random gradient descent optimizer is used to train the voiceprint recognition model to be trained.

[0036] In one specific embodiment, a batch ciphertext processing module is further included for receiving input of a plurality of voiceprint ciphertext sub-data; calculating an output vector size of the plurality of voiceprint ciphertext sub-data according to a structure of a voiceprint recognition model; setting a number of slots for storing voiceprint ciphertext sub-data according to the size of the voiceprint ciphertext sub-data; sequentially traversing encrypted data in each slot, and right-rotating each encrypted data by a preset rotation bit number to make the position of each data in the ciphertext not overlap; obtaining voiceprint ciphertext data when the total size of the ciphertext after merging is less than the number of slots.

[0037] Advantages of the present application:

[0038] The present application provides a privacy protection voiceprint recognition inference method based on a homomorphic encryption technology, which comprises the following steps: collecting a plurality of original voiceprint data from a preset voiceprint database, and forming a voiceprint data set after preprocessing the original voiceprint data; inputting the voiceprint data set into a voiceprint recognition model to be trained, and training the voiceprint recognition model to be trained to obtain a trained voiceprint recognition model to generate encrypted model parameters; inputting the encrypted model parameters and voiceprint data to be recognized into an encrypter to obtain voiceprint ciphertext data; and performing calculation and inference on the voiceprint ciphertext data according to the homomorphic encryption characteristics. In this way, the voiceprint recognition model is used to recognize the voiceprint data to be recognized, so as to improve the calculation accuracy in the homomorphic encryption environment, reduce resource consumption, and improve the inference efficiency.

[0039] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a flowchart of a privacy protection voiceprint recognition inference method based on a homomorphic encryption technology provided by an embodiment of the present application;

[0041] Figure 2 is a block diagram of a privacy protection voiceprint recognition inference system module based on a homomorphic encryption technology provided by an embodiment of the present application. DETAILED DESCRIPTION

[0042] The present application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0043] Embodiment one

[0044] Please refer to Figure 1 , Figure 1 is a flowchart of a privacy protection voiceprint recognition inference method based on a homomorphic encryption technology provided by an embodiment of the present application, which comprises:

[0045] S1, collect a plurality of original voiceprint data from a preset voiceprint database, and form a voiceprint data set after preprocessing the original voiceprint data; specifically, the voiceprint data set includes a training set and a test set.

[0046] In this step, the original voiceprint data is preprocessed to form a voiceprint data set, including:

[0047] Removing the speech silence part in the original voiceprint data to form complete sample data;

[0048] The complete sample data is sequentially denoised, feature extracted and standardized;

[0049] The voiceprint data after standardization is converted in data format and data length to form a voiceprint data set to adapt to the homomorphic encryption characteristics.

[0050] This process is to denoise, feature extract and standardize the voiceprint data to improve the accuracy of subsequent recognition. At the same time, the test data is preprocessed to ensure that its format and features meet the requirements of the homomorphic encryption algorithm. Homomorphic encryption algorithms usually have strict requirements for the type and range of input data. For example, some encryption algorithms only support integer operations, so it is necessary to ensure that the data input by the model can be converted to an appropriate format.

[0051] S2, input the voiceprint data set into a voiceprint recognition model to be trained, and train the voiceprint recognition model to be trained to obtain a trained voiceprint recognition model to generate encrypted model parameters;

[0052] It should be noted that the process of model training is carried out in plaintext state using the voiceprint data set, the purpose is to ensure that the model can effectively identify and classify voiceprint signals after optimization, so as to obtain a model for subsequent inference. The voiceprint recognition model of the embodiment uses, for example, a MobioleNetV2 model, uses a stochastic gradient descent (SGD) optimizer to perform deep training on the MobilNetV2, and uses a cross-entropy loss function or an AAM loss function to calculate the training cost.

[0053] S3, input the encrypted model parameters and the voiceprint data to be identified into an encrypter to obtain voiceprint ciphertext data;

[0054] Because the calculation of convolution and other operations in the homomorphic environment is relatively complex, the inference time is also relatively long. In order to reduce the ciphertext inference time, the embodiment adopts a combined ciphertext batch processing technology, which combines a plurality of encrypted input data into a single ciphertext for batch processing, greatly reducing the overall calculation time, and having a significant advantage in large-scale data processing.

[0055] The specific processing steps further include:

[0056] receiving an input of a plurality of voiceprint ciphertext sub-data;

[0057] calculating an output vector size of the plurality of voiceprint ciphertext sub-data according to a structure of a voiceprint recognition model;

[0058] setting a number of slots for storing voiceprint ciphertext sub-data according to a size of the voiceprint ciphertext sub-data;

[0059] sequentially traversing encrypted data in each slot, and right-rotating each encrypted data according to a preset rotation bit number to make positions of each data in ciphertext not overlap;

[0060] when the total size of the ciphertext after merging is less than the number of slots, obtaining voiceprint ciphertext data. Preferably, the preset rotation bit number is (i-1) x N, where i represents the order of the encrypted data, and N represents the number of slots.

[0061] S4, performing calculation and reasoning on the voiceprint ciphertext data according to the homomorphic encryption characteristic. Specifically, the encrypted voiceprint ciphertext data is transmitted to a forward method of a homomorphic encryption architecture, and the architecture performs convolution, activation and pooling operations and the like according to the layer name or identifier of the pre-trained model, and performs reasoning calculation on the encrypted data by using the characteristic of homomorphic encryption, so as to realize voiceprint recognition and return an encrypted result. Then, the reasoning result is decrypted, the encrypted output is converted into plaintext, and the final voiceprint recognition result is output. Subsequently, the voiceprint recognition result in the plaintext state is analyzed and compared. Thus, by establishing such an encrypted reasoning framework, the voiceprint can be accurately recognized while protecting the privacy of the voiceprint data.

[0062] Preferably, the RNS-CKKS full homomorphic encryption algorithm is used in the embodiment to encrypt and decrypt data and to perform operation on ciphertext, so as to facilitate encryption of real or complex vectors. Specifically, the calculation function of the homomorphic encryption algorithm is realized by using the SEAL-Python encryption calculation library. The encryption scheme uses parameters determined according to the training model architecture, so as to ensure that all security parameters meet the 128-bit security level. By adjusting the MobioleNetV2 model to be trained based on the homomorphic encryption algorithm and the encryption calculation library, an encryption model architecture suitable for the voiceprint recognition task is adapted.

[0063] In one specific example, first, the system uses the TIMIT dataset for voiceprint data collection and preprocessing. In the preprocessing stage, the voiceprint data is preprocessed, the speech silence part is removed, and the complete sample is segmented for direct output of the model for special audio feature extraction, and then the voiceprint recognition model is trained. Next, the optimized MobileNetV2 model is selected for voiceprint recognition task, the model is trained using preprocessed data, and the model performance is improved by adjusting hyperparameters, while the accuracy and robustness of the model are evaluated by cross-validation and other methods. After the model training is completed, the system sets the RNS-CKKS encryption context, generates encryption parameters, including key generation, encoder and encrypter. Then, the trained model parameters and voiceprint data are input into the encrypter to generate encrypted ciphertext data. Next, use the encrypted voiceprint features as input, pass through the MobileNetV2 model for inference, and use the homomorphic encryption feature to calculate directly on the ciphertext to ensure data privacy. Finally, the system decrypts the output of the model to restore the voiceprint recognition result, and outputs the decrypted result to the user. Through this series of steps, the voiceprint recognition privacy protection system based on homomorphic encryption effectively protects user privacy while realizing efficient voiceprint recognition function.

[0064] For the multi-ciphertext batch processing method, first, receive multiple encrypted voiceprint data for batch processing. Then, according to the structure of the MobileNetV2 model, calculate the size of the intermediate or final output vector to ensure that there is no overlap when merging. Next, initialize the slot, set the number of slots for storing encrypted data, for example, 800. Traverse each encrypted data and process each encrypted data one by one. Rotate the encrypted data to the right for the ith encrypted data, and the number of bits rotated is (i-1)×800(i-1)×800 to ensure that the position of each data in the ciphertext does not overlap. Next, insert the rotated encrypted data into the ciphertext. Then check that the total size of the merged ciphertext does not exceed the initialized slot number. Output the merged ciphertext and return the merged ciphertext to complete the batch processing.

[0065] See Figure 2 , Figure 2 is a kind of private protection voiceprint recognition inference system module block diagram based on full homomorphic encryption technology provided by the embodiment of the application, comprising:

[0066] The test set generation module is configured to collect a plurality of original voiceprint data from a pre-set voiceprint database, and form a voiceprint data set after preprocessing the original voiceprint data.

[0067] The model training module is configured to input the voiceprint data set into a voiceprint recognition model to be trained, and train the voiceprint recognition model to be trained to obtain a trained voiceprint recognition model to generate encrypted model parameters.

[0068] an encryption module configured to input the encrypted model parameters and voiceprint data to be identified into an encryptor to obtain voiceprint ciphertext data;

[0069] an inference module configured to perform computational inference on the voiceprint ciphertext data according to homomorphic encryption characteristics.

[0070] In one specific embodiment, the original voiceprint data is preprocessed to form a voiceprint data set, including:

[0071] removing speech silence portions in the original voiceprint data to form complete sample data;

[0072] sequentially performing denoising, feature extraction and standardization processing on the complete sample data;

[0073] performing data format conversion and data length on the voiceprint data after standardization processing to form a voiceprint data set to adapt to homomorphic encryption characteristics.

[0074] In one specific embodiment, the voiceprint recognition model to be trained is trained, including:

[0075] using a stochastic gradient descent optimizer to train the voiceprint recognition model to be trained.

[0076] In one specific embodiment, further comprising a batch ciphertext processing module configured to receive input of a plurality of voiceprint ciphertext sub-data; calculate an output vector size of the plurality of voiceprint ciphertext sub-data according to a structure of a voiceprint recognition model; set a number of slots for storing voiceprint ciphertext sub-data according to the size of the voiceprint ciphertext sub-data; sequentially traverse encrypted data in each slot, and right-rotate each encrypted data according to a preset rotation bit number to make the position of each data in ciphertext not overlap; after merging the ciphertext, if the total size of the ciphertext does not exceed the number of slots, obtain voiceprint ciphertext data; wherein the preset rotation bit number is (i-1) x N, where i represents the order of the encrypted data, and N represents the number of slots.

[0077] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in the present specification.

[0078] Although the present application is described in conjunction with various embodiments, it will be understood by those skilled in the art that other variations of the disclosed embodiments can be understood and effected by a worker in the art from the disclosures provided herein, the drawings, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite articles "a" or "an" do not exclude a plurality. A single processor or other unit can fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to an advantage.

[0079] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, apparatus (device), or computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects all generally referred to herein as a "module" or "system." Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, and the like) embodying computer readable program code. The computer program product can be for use in implementing the techniques described herein, and the various embodiments can be implemented as part of the computer program product, which in turn is provided on one or more computer-usable media. The computer-usable media can have code embodied therein or thereon, and the code can cause a computer to implement the techniques described herein.

[0080] The present application is described in terms of exemplary embodiments, apparatus (devices), and computer program products thereof, which together can implement methods, in flowcharts and / or block diagrams. It will be understood by those within the art that each block of the flowcharts and / or block diagrams, and combinations thereof, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowcharts and / or block diagrams block or blocks. Figure 1 Each flowchart and / or block diagram block can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). Figure 1 The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowcharts and / or block diagrams block or blocks.

[0081] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowcharts and / or block diagrams block or blocks. Figure 1 Each flowchart and / or block diagram block can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). Figure 1 The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowcharts and / or block diagrams block or blocks.

[0082] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable data processing devices to generate a computer implemented process, so that the instructions executed on the computer or other programmable data processing devices provide a process for implementing the functions specified in the flowchart Figure 1 one flow or multiple flows and / or the functions specified in the block Figure 1 one flow or multiple flows and / or the functions specified in the block

[0083] The above is a further detailed description of the present application in conjunction with specific preferred embodiments, and cannot be deemed to limit the specific implementation of the present application to these descriptions. For those of ordinary skill in the art to which the present application belongs, without departing from the concept of the present application, a number of simple deductions or substitutions can also be made, which should be considered to fall within the protection scope of the present application.

Claims

1. A privacy-preserving voiceprint recognition inference method based on full homomorphic encryption technology, characterized in that, The method comprises the following steps: Collecting a plurality of original voiceprint data from a preset voiceprint database, and forming a voiceprint data set after preprocessing the original voiceprint data; Inputting the voiceprint data set into a voiceprint recognition model to be trained, and training the voiceprint recognition model to be trained to obtain a trained voiceprint recognition model to generate encrypted model parameters; Inputting the encrypted model parameters and voiceprint data to be identified into an encrypter to obtain voiceprint ciphertext data; According to the homomorphic encryption characteristic, the voiceprint ciphertext data is calculated and inferred; Before the voiceprint ciphertext data is calculated and inferred according to the homomorphic encryption characteristic, the method further comprises the following steps: Receiving input of a plurality of voiceprint ciphertext sub-data; According to the structure of the voiceprint recognition model, the output vector size of the plurality of voiceprint ciphertext sub-data is calculated; According to the size of the voiceprint ciphertext sub-data, the number of slots for storing the voiceprint ciphertext sub-data is set; The encrypted data in each slot is sequentially traversed, and each encrypted data is right-rotated according to a preset rotation bit number to make the position of each data in the ciphertext not overlap; When the total size of the ciphertext after merging is not more than the number of slots, the voiceprint ciphertext data is obtained; The preset rotation bit number is (i-1)×N(i-1)×N, wherein i represents the order of the encrypted data, and N represents the number of slots.

2. The privacy-preserving voiceprint recognition inference method based on homomorphic encryption technology according to claim 1, characterized in that, After preprocessing the original voiceprint data, the voiceprint data set is formed, which comprises the following steps: Removing the speech silence part in the original voiceprint data to form complete sample data; The complete sample data is sequentially subjected to denoising, feature extraction and standardization processing; The voiceprint data after standardization processing is subjected to data format conversion and data length to form a voiceprint data set to adapt to the homomorphic encryption characteristic.

3. The privacy-preserving voiceprint recognition inference method based on homomorphic encryption technology according to claim 1, characterized in that, The voiceprint recognition model to be trained is trained, which comprises the following steps: Using a stochastic gradient descent optimizer to train the voiceprint recognition model to be trained.

4. The privacy-preserving voiceprint recognition inference method based on homomorphic encryption technology according to claim 3, characterized in that, The encrypter adopts an RNS-CKKS full homomorphic encryption algorithm.

5. A privacy-preserving voiceprint recognition inference system based on a fully homomorphic encryption technique, characterized in that, The method comprises the following steps: A test set generation module is configured to collect a plurality of original voiceprint data from a preset voiceprint database, and form a voiceprint data set after preprocessing the original voiceprint data; A model training module is configured to input the voiceprint data set into a voiceprint recognition model to be trained, and train the voiceprint recognition model to be trained to obtain a trained voiceprint recognition model to generate encrypted model parameters; An encryption module is configured to input the encrypted model parameters and voiceprint data to be identified into an encrypter to obtain voiceprint ciphertext data; An inference module is configured to calculate and infer the voiceprint ciphertext data according to the homomorphic encryption characteristic; A batch ciphertext processing module is further included, which is configured to receive input of a plurality of voiceprint ciphertext sub-data; According to the structure of the voiceprint recognition model, the output vector size of a plurality of voiceprint ciphertext sub-data is calculated; the number of slots for storing voiceprint ciphertext sub-data is set according to the size of the voiceprint ciphertext sub-data; each encrypted data in each slot is sequentially traversed, and each encrypted data is right-rotated according to a preset rotation bit number so that the position of each data in the ciphertext does not overlap; when the total size of the ciphertext after merging is less than the number of slots, the voiceprint ciphertext data is obtained; wherein the preset rotation bit number is (i-1) x N (i-1) x N, wherein i represents the order of the encrypted data, and N represents the number of slots.

6. The privacy-preserving voiceprint recognition inference system based on homomorphic encryption technology according to claim 5, characterized in that, After preprocessing the original voiceprint data, a voiceprint data set is formed, including: Removing the speech silence part in the original voiceprint data to form complete sample data; The complete sample data is sequentially subjected to denoising, feature extraction and standardization processing; The voiceprint data after standardization processing is subjected to data format conversion and data length to form a voiceprint data set to adapt to the homomorphic encryption characteristics.

7. The privacy-preserving voiceprint recognition inference system based on homomorphic encryption technology according to claim 5, characterized in that, Training the voiceprint recognition model to be trained, including: Using a stochastic gradient descent optimizer to train the voiceprint recognition model to be trained.

Images