Voiceprint recognition method and device based on artificial intelligence, computer equipment and medium

Abstract

The invention belongs to the technical field of artificial intelligence, and relates to a voiceprint recognition method and device based on artificial intelligence, computer equipment and a storage medium. Preprocessing the original voice based on a room impulse response strategy to obtain target voice; performing noise generation processing on the target voice based on the initial voiceprint recognition model to obtain an adversarial noise; generating an adversarial sample based on the original voice and adversarial noise; constructing sample data based on the target voice and the confrontation sample; performing confrontation training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model; and performing voiceprint recognition processing on the to-be-processed voice based on the target voiceprint recognition model to obtain a voiceprint recognition result. In addition, the invention also relates to a block chain technology, and the target voiceprint recognition model can be stored in a block chain. According to the method, the defense capability of the target voiceprint recognition model on the attack resistance is effectively improved, and the recognition accuracy of voiceprint recognition processing is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence development technology and the field of financial technology, and in particular to artificial intelligence-based voiceprint recognition methods, devices, computer equipment and storage media. Background Art

[0002] With the rapid development of science and technology, especially the continuous progress in the field of artificial intelligence, end-to-end deep neural network (DNN) technology has made significant breakthroughs in the field of voiceprint recognition. In recent years, the demand for voiceprint recognition technology in the financial field has been growing. Thanks to the significant improvement in its recognition accuracy, voiceprint recognition technology has been widely used in a variety of real-world scenarios, including but not limited to voice assistants, voiceprint-based security authentication systems, etc. The core of these applications is to accurately identify the identity of the speaker, thereby ensuring the security of information and the accuracy of services.

[0003] However, although DNN technology has shown great potential in voiceprint recognition, it also faces a challenge that cannot be ignored: the vulnerability of adversarial samples. Adversarial samples refer to samples generated by adding tiny, carefully designed perturbations to the original input data. These perturbations are almost imperceptible to human hearing, but are enough to mislead the DNN model to make wrong judgments. For example, in a voiceprint recognition system, even if the real speaker is Xiao Ming, after being attacked by adversarial samples, the system may mistakenly identify him as someone else. Therefore, the existing DNN-based voiceprint recognition methods have poor performance in defending against adversarial attacks and low accuracy of voiceprint recognition. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose a voiceprint recognition method, device, computer equipment and storage medium based on artificial intelligence to solve the technical problems that the existing DNN-based voiceprint recognition method has poor performance in defending against adversarial attacks and has low accuracy in voiceprint recognition.

[0005] In order to solve the above technical problems, the embodiment of the present application provides a voiceprint recognition method based on artificial intelligence, which adopts the following technical solutions:

[0006] Obtaining the original voice collected in advance;

[0007] Preprocessing the original speech based on a preset room impulse response strategy to obtain a corresponding target speech;

[0008] Based on the preset initial voiceprint recognition model, noise generation processing is performed on the target voice to obtain the corresponding adversarial noise;

[0009] Generate a corresponding adversarial sample based on the original speech and the adversarial noise;

[0010] Constructing corresponding sample data based on the target speech and the adversarial sample;

[0011] Performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model;

[0012] Based on the target voiceprint recognition model, voiceprint recognition processing is performed on the speech to be processed to obtain a corresponding voiceprint recognition result.

[0013] Furthermore, the step of preprocessing the original speech based on a preset room impulse response strategy to obtain a corresponding target speech specifically includes:

[0014] Calculating the room impulse response when the original speech propagates in a preset environment;

[0015] Performing a convolution operation on the original speech based on the room impulse response to obtain a corresponding first speech;

[0016] The first speech is used as the target speech.

[0017] Furthermore, the step of performing noise generation processing on the target voice based on the preset initial voiceprint recognition model to obtain the corresponding anti-noise specifically includes:

[0018] Acquire a first feature representation of the target speech;

[0019] Calling the initial voiceprint recognition model and obtaining an initial loss function of the initial voiceprint recognition model;

[0020] Calculating the gradient of the initial loss function with respect to the first feature representation based on a fast gradient sign method;

[0021] Corresponding adversarial noise is generated based on the gradient.

[0022] Furthermore, the step of generating a corresponding adversarial sample based on the original speech and the adversarial noise specifically includes:

[0023] Acquire a second feature representation of the original speech;

[0024] Adding the adversarial noise to the second feature representation to obtain a corresponding target feature representation;

[0025] The target feature representation is converted based on a preset reshaping algorithm to obtain a corresponding speech signal;

[0026] The adversarial sample is generated based on the speech signal.

[0027] Furthermore, the step of performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model specifically includes:

[0028] Dividing the sample data into a training set and a validation set;

[0029] Constructing a target loss based on the classification loss corresponding to the original speech and the pairing loss corresponding to the adversarial sample;

[0030] Based on a preset target optimization algorithm and the target loss, the initial voiceprint recognition model is trained using the training set to obtain a corresponding first voiceprint recognition model;

[0031] Performing a performance evaluation on the first voiceprint recognition model based on the verification set to obtain a corresponding performance evaluation result;

[0032] Optimizing the first voiceprint recognition model based on the performance evaluation result to obtain a second voiceprint recognition model that meets preset performance expectations;

[0033] The second voiceprint recognition model is used as the target voiceprint recognition model.

[0034] Furthermore, the step of performing voiceprint recognition processing on the speech to be processed based on the target voiceprint recognition model to obtain a corresponding voiceprint recognition result specifically includes:

[0035] Preprocessing the speech to be processed to obtain a corresponding second speech;

[0036] Extracting features from the second speech to obtain corresponding specified features;

[0037] Performing recognition processing on the designated feature based on the target voiceprint recognition model to obtain a corresponding voiceprint recognition result;

[0038] The voiceprint recognition result is outputted based on a preset output method.

[0039] Furthermore, after the step of performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model, the method further includes:

[0040] Call the preset specified storage medium;

[0041] Performing storage processing on the target voiceprint recognition model based on the designated storage medium;

[0042] After the target voiceprint recognition model is stored, the target voiceprint recognition model is system deployed.

[0043] In order to solve the above technical problems, the embodiment of the present application also provides a voiceprint recognition device based on artificial intelligence, which adopts the following technical solution:

[0044] An acquisition module, used to acquire the original speech collected in advance;

[0045] A preprocessing module, used for preprocessing the original speech based on a preset room impulse response strategy to obtain a corresponding target speech;

[0046] A first generation module is used to perform noise generation processing on the target voice based on a preset initial voiceprint recognition model to obtain corresponding adversarial noise;

[0047] A second generating module, configured to generate a corresponding adversarial sample based on the original speech and the adversarial noise;

[0048] A construction module, used to construct corresponding sample data based on the target speech and the adversarial sample;

[0049] A training module, used for performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model;

[0050] The recognition module is used to perform voiceprint recognition processing on the speech to be processed based on the target voiceprint recognition model to obtain a corresponding voiceprint recognition result.

[0051] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0052] Obtaining the original voice collected in advance;

[0053] Preprocessing the original speech based on a preset room impulse response strategy to obtain a corresponding target speech;

[0054] Based on the preset initial voiceprint recognition model, noise generation processing is performed on the target voice to obtain the corresponding adversarial noise;

[0055] Generate a corresponding adversarial sample based on the original speech and the adversarial noise;

[0056] Constructing corresponding sample data based on the target speech and the adversarial sample;

[0057] Performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model;

[0058] Based on the target voiceprint recognition model, voiceprint recognition processing is performed on the speech to be processed to obtain a corresponding voiceprint recognition result.

[0059] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0060] Obtaining the original voice collected in advance;

[0061] Preprocessing the original speech based on a preset room impulse response strategy to obtain a corresponding target speech;

[0062] Based on the preset initial voiceprint recognition model, noise generation processing is performed on the target voice to obtain the corresponding adversarial noise;

[0063] Generate a corresponding adversarial sample based on the original speech and the adversarial noise;

[0064] Constructing corresponding sample data based on the target speech and the adversarial sample;

[0065] Performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model;

[0066] Based on the target voiceprint recognition model, voiceprint recognition processing is performed on the speech to be processed to obtain a corresponding voiceprint recognition result.

[0067] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0068] The present application first obtains the original speech collected in advance; and pre-processes the original speech based on a preset room impulse response strategy to obtain the corresponding target speech; then performs noise generation processing on the target speech based on a preset initial voiceprint recognition model to obtain the corresponding adversarial noise; then generates a corresponding adversarial sample based on the original speech and the adversarial noise; subsequently constructs corresponding sample data based on the target speech and the adversarial sample; further performs adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model; finally, performs voiceprint recognition processing on the processed speech based on the target voiceprint recognition model to obtain the corresponding voiceprint recognition result. The present application obtains a target speech by preprocessing the pre-collected original speech based on the use of a room impulse response strategy, then performs noise generation processing on the target speech based on an initial voiceprint recognition model to obtain adversarial noise, and generates corresponding adversarial samples based on the original speech and the adversarial noise, and then performs adversarial training on the initial voiceprint recognition model based on the sample data constructed from the target speech and the adversarial samples, thereby efficiently and accurately constructing a target voiceprint recognition model that can effectively prevent adversarial sample attacks, improving the target voiceprint recognition model's defense capability against adversarial attacks, and effectively improving the recognition accuracy and robustness of the target voiceprint recognition model for the voiceprint recognition processing of the processed speech when facing adversarial attacks. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0070] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0071] Figure 2 A flowchart of an embodiment of a voiceprint recognition method based on artificial intelligence according to the present application;

[0072] Figure 3 is a structural schematic diagram of an embodiment of a voiceprint recognition device based on artificial intelligence according to the present application;

[0073] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0074] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0075] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0076] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0077] like Figure 1As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0078] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0079] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to the laptop computer 1011, tablet computer 1012 or mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0080] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0081] It should be noted that the artificial intelligence-based voiceprint recognition method provided in the embodiments of the present application is generally executed by a server / terminal device, and accordingly, the artificial intelligence-based voiceprint recognition device is generally arranged in a server / terminal device.

[0082] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0083] Continue to refer Figure 2, showing a flow chart of an embodiment of a voiceprint recognition method based on artificial intelligence according to the present application. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted. The voiceprint recognition method based on artificial intelligence provided in the embodiment of the present application can be applied to any scenario where voiceprint recognition is required, and the voiceprint recognition method based on artificial intelligence can be applied to products in these scenarios, for example, voiceprint recognition in the field of finance and insurance. The voiceprint recognition method based on artificial intelligence includes the following steps:

[0084] Step S201, obtaining the pre-collected original speech.

[0085] In this embodiment, the voiceprint recognition method based on artificial intelligence is run on the electronic device (for example Figure 1 The server / terminal device shown in the figure) can obtain the original voice through a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future. The executor of this application may be a voiceprint processing system, or simply referred to as a system. The above-mentioned original voice is the original voice signal collected in advance for attacking.

[0086] Step S202: pre-processing the original speech based on a preset room impulse response strategy to obtain a corresponding target speech.

[0087] In this embodiment, the specific implementation process of preprocessing the original speech based on the preset room impulse response strategy to obtain the corresponding target speech will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0088] Step S203: performing noise generation processing on the target speech based on the preset initial voiceprint recognition model to obtain corresponding adversarial noise.

[0089] In this embodiment, the above-mentioned specific implementation process of performing noise generation processing on the target speech based on the preset initial voiceprint recognition model to obtain the corresponding anti-noise will be further described in detail in the subsequent specific embodiments of this application, and will not be elaborated in detail here.

[0090] Step S204: Generate a corresponding adversarial sample based on the original speech and the adversarial noise.

[0091] In this embodiment, the specific implementation process of the adversarial sample corresponding to the original speech and the adversarial noise generation will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0092] Step S205: construct corresponding sample data based on the target speech and the adversarial sample.

[0093] In this embodiment, the original speech and the adversarial sample can be merged to construct the corresponding sample data. Specifically, the original speech is paired with the logit of the adversarial sample to obtain the paired original speech and adversarial sample, and used as the corresponding sample data. The subsequent adversarial training of the initial voiceprint recognition model using the paired original speech and adversarial sample can enhance the robustness of the model to adversarial samples. During the training process, the model learns how to distinguish between real speech and adversarial samples, thereby improving the defense capability against adversarial attacks. Among them, the original voiceprint recognition model will make incorrect predictions about the adversarial sample.

[0094] Step S206: performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model.

[0095] In this embodiment, the specific implementation process of performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0096] Step S207: performing voiceprint recognition processing on the speech to be processed based on the target voiceprint recognition model to obtain a corresponding voiceprint recognition result.

[0097] In this embodiment, the specific implementation process of performing voiceprint recognition processing on the processed speech based on the target voiceprint recognition model to obtain the corresponding voiceprint recognition result will be further described in detail in the subsequent specific embodiments of this application and will not be elaborated on here.

[0098] The present application first obtains the original speech collected in advance; and pre-processes the original speech based on a preset room impulse response strategy to obtain the corresponding target speech; then performs noise generation processing on the target speech based on a preset initial voiceprint recognition model to obtain the corresponding adversarial noise; then generates a corresponding adversarial sample based on the original speech and the adversarial noise; subsequently constructs corresponding sample data based on the target speech and the adversarial sample; further performs adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model; finally, performs voiceprint recognition processing on the processed speech based on the target voiceprint recognition model to obtain the corresponding voiceprint recognition result. The present application obtains a target speech by preprocessing the pre-collected original speech based on the use of a room impulse response strategy, then performs noise generation processing on the target speech based on an initial voiceprint recognition model to obtain adversarial noise, and generates corresponding adversarial samples based on the original speech and the adversarial noise, and then performs adversarial training on the initial voiceprint recognition model based on the sample data constructed from the target speech and the adversarial samples, thereby efficiently and accurately constructing a target voiceprint recognition model that can effectively prevent adversarial sample attacks, improving the target voiceprint recognition model's defense capability against adversarial attacks, and effectively improving the recognition accuracy and robustness of the target voiceprint recognition model for the voiceprint recognition processing of the processed speech when facing adversarial attacks.

[0099] In some optional implementations, step S202 includes the following steps:

[0100] The room impulse response when the original speech is propagated in a preset environment is calculated.

[0101] In this embodiment, since digital adversarial samples are most likely to lose their effectiveness during wireless transmission, modeling the sound distortion of the audio propagation process is crucial for launching practical attacks. This technology uses Room Impulse Response Estimation (RIR) for modeling, which can represent that the sound wave signal has different attenuation and delay when it propagates through different paths of the medium.

[0102] Specifically, the room impulse response (RIR) of the sound wave signal of the original speech when propagating in different environments can be calculated based on the pre-collected environmental data by using acoustic simulation software such as an acoustic library, or actual measurement. The room impulse response may specifically include effects such as attenuation, reflection, diffraction and delay of sound waves on different paths. The above environmental data includes pre-collected acoustic data of different environments (such as conference rooms, corridors, outdoors, etc.), including the size, shape, material and possible obstacles of the room.

[0103] A convolution operation is performed on the original speech based on the room impulse response to obtain a corresponding first speech.

[0104] In this embodiment, the above convolution operation refers to applying the calculated room impulse response to the original speech to simulate the propagation process of sound waves in different environments. The convolution operation includes multiplying and integrating the room impulse response and the original speech in the time domain (or equivalently multiplying in the frequency domain).

[0105] The first speech is used as the target speech.

[0106] The present application calculates the room impulse response when the original speech propagates in a preset environment; then performs a convolution operation on the original speech based on the room impulse response to obtain a corresponding first speech; and subsequently uses the first speech as the target speech. The present application calculates the room impulse response when the original speech propagates in a preset environment; and then performs a convolution operation on the original speech based on the room impulse response to achieve fast and accurate preprocessing of the original speech, effectively ensuring the accuracy of the generated target speech.

[0107] In some optional implementations of this embodiment, step S203 includes the following steps:

[0108] A first feature representation of the target speech is obtained.

[0109] In this embodiment, the first feature representation refers to the spectrum feature or MFCC feature of the speech signal.

[0110] The initial voiceprint recognition model is called, and an initial loss function of the initial voiceprint recognition model is obtained.

[0111] In this embodiment, the initial voiceprint recognition model is a pre-built voiceprint recognition model that needs to be attacked, for example, a voiceprint recognition model built based on an end-to-end deep neural network (DNN). The initial loss function is the loss function used when building the voiceprint recognition model.

[0112] The gradient of the initial loss function with respect to the first feature representation is calculated based on a fast gradient sign method.

[0113] In this embodiment, the initial loss function of the initial voiceprint recognition model can be calculated for the input feature, that is, the gradient of the first feature represented by the above-mentioned fast gradient sign method (FGSM). The gradient of the unknown parameter is iterated by using the loss function. When generating adversarial noise, the purpose is to learn small feature changes that cause relatively large changes in the loss function. Therefore, the gradient of the feature is calculated using the loss function. Generally, the adversarial noise can be obtained by calculating it only once.

[0114] Corresponding adversarial noise is generated based on the gradient.

[0115] In this embodiment, adversarial noise may be generated based on the calculated gradient, which is specifically achieved by multiplying the gradient with a small constant (called “step size” or “perturbation strength”).

[0116] The present application obtains the first feature representation of the target speech; then calls the initial voiceprint recognition model and obtains the initial loss function of the initial voiceprint recognition model; then calculates the gradient of the initial loss function for the first feature representation based on the fast gradient sign method; and subsequently generates the corresponding adversarial noise based on the gradient. The present application obtains the first feature representation of the target speech, and obtains the initial loss function of the called initial voiceprint recognition model; then calculates the gradient of the initial loss function for the first feature representation based on the use of the fast gradient sign method, and then according to the obtained gradient, the corresponding adversarial noise can be generated quickly and accurately, thereby improving the generation efficiency and accuracy of the adversarial noise.

[0117] In some optional implementations, step S204 includes the following steps:

[0118] A second feature representation of the original speech is obtained.

[0119] In this embodiment, the second feature representation may refer to features such as spectrum features and MFCC of the original speech.

[0120] The adversarial noise is added to the second feature representation to obtain a corresponding target feature representation.

[0121] In this embodiment, the generated adversarial noise can be added to the second feature representation of the original speech to obtain the feature representation of the adversarial sample, that is, the target feature representation.

[0122] The target feature representation is converted based on a preset reshaping algorithm to obtain a corresponding speech signal.

[0123] In this embodiment, the reshaping algorithm may specifically adopt algorithms such as inverse MFCC transformation and inverse spectrum transformation. The target feature representation is transformed by using the reshaping algorithm to transform the feature representation of the adversarial sample back to the time domain speech signal, that is, the speech signal.

[0124] The adversarial sample is generated based on the speech signal.

[0125] In this embodiment, the above-mentioned voice signal can be directly used as the above-mentioned adversarial sample. In addition, the generated adversarial sample can also be input into the original voiceprint recognition model to verify whether it leads to an erroneous prediction, thereby ensuring the attack effect of the generated adversarial sample.

[0126] The present application obtains the second feature representation of the original speech; then adds the adversarial noise to the second feature representation to obtain the corresponding target feature representation; then converts the target feature representation based on a preset reshaping algorithm to obtain the corresponding speech signal; and subsequently generates the adversarial sample based on the speech signal. The present application obtains the target feature representation by adding the obtained second feature representation to the second feature representation of the original speech; then converts the target feature representation based on the use of a reshaping algorithm to obtain a speech signal, and then based on the speech signal, the corresponding adversarial sample can be efficiently and accurately constructed, thereby improving the construction efficiency and accuracy of the adversarial sample.

[0127] In some optional implementations, step S206 includes the following steps:

[0128] The sample data is divided into a training set and a validation set.

[0129] In this embodiment, the sample data may be divided into a training set and a validation set according to a preset division ratio, wherein the numerical value of the division ratio is not specifically limited, for example, 7:3 may be used.

[0130] A target loss is constructed based on the classification loss corresponding to the original speech and the pairing loss corresponding to the adversarial sample.

[0131] In this embodiment, the classification loss corresponding to the original speech can specifically adopt the cross entropy loss. The pairing loss is the Logit pairing loss. The pairing loss can specifically adopt the L2 loss, KL divergence or other distance measurement methods. Among them, the corresponding target loss can be obtained by weighted summing the classification loss corresponding to the original speech and the pairing loss corresponding to the adversarial sample. In addition, Logit pairing refers to pairing the original speech with the logit of the adversarial sample (i.e., the unnormalized score of the model output layer). Logit pairing aims to enable the adversarial sample to learn the characteristics of the original sample, thereby removing the influence of adversarial noise.

[0132] Based on a preset target optimization algorithm and the target loss, the initial voiceprint recognition model is trained using the training set to obtain a corresponding first voiceprint recognition model.

[0133] In this embodiment, the above-mentioned target optimization algorithm can specifically adopt any one of stochastic gradient descent (SGD), Adam, RMSprop, etc. The target optimization algorithm will be used to update the parameters of the model to minimize the loss function. Specifically, the initial voiceprint recognition model is trained by using the training set and the selected target optimization algorithm. During the model training process, the gradient of the target loss is iteratively calculated, and the parameters of the initial voiceprint recognition model are updated according to the gradient. At the same time, indicators such as loss value and accuracy during the training process are monitored to ensure that the model is learning correctly. And according to the performance during the training process, hyperparameters such as learning rate, batch size, regularization parameter, etc. are adjusted to optimize the performance of the model, and finally a trained first voiceprint recognition model is obtained.

[0134] A performance evaluation is performed on the first voiceprint recognition model based on the verification set to obtain a corresponding performance evaluation result.

[0135] In this embodiment, the trained first voiceprint recognition model is evaluated by using a validation set, the recognition accuracy, false recognition rate, rejection rate and other indicators of the first voiceprint recognition model are calculated, and the performance of the first voiceprint recognition model when facing adversarial attacks is observed to obtain corresponding performance evaluation results.

[0136] The first voiceprint recognition model is optimized based on the performance evaluation result to obtain a second voiceprint recognition model that meets preset performance expectations.

[0137] In this embodiment, the first voiceprint recognition model can be adjusted and optimized according to the obtained performance evaluation results, including modifying the model architecture, adding regularization terms, adjusting hyperparameters, etc. And use the adjusted model and training set for retraining. During the training process, continue to monitor indicators such as loss value and accuracy, and adjust hyperparameters in a timely manner. And use the validation set to evaluate the retrained model. Compare the model performance before and after the adjustment to verify the effectiveness of the improvement. And then decide whether to continue iterative training based on the evaluation results. If there is still room for improvement in the model performance, continue to adjust the model and retrain; if the model performance has stabilized or reached the predetermined performance expectation, stop iterative training and obtain the corresponding second voiceprint recognition model, and use it as the final target voiceprint recognition model.

[0138] The second voiceprint recognition model is used as the target voiceprint recognition model.

[0139] In this embodiment, after generating the adversarial sample, the initial voiceprint recognition model is trained adversarially by using the sample data constructed based on the target speech and the adversarial sample, thereby achieving the purpose of preventing the adversarial sample attack. The core idea of ​​adversarial training is to use the prediction result of the original sample as a noise-free reference, so that the adversarial sample learns the characteristics of the original sample to achieve the purpose of denoising, and then the constructed target voiceprint recognition model can effectively prevent the attack of the adversarial sample.

[0140] The present application divides the sample data into a training set and a validation set; then constructs a target loss based on the classification loss corresponding to the original speech and the pairing loss corresponding to the adversarial sample; then, based on a preset target optimization algorithm and the target loss, uses the training set to train the initial voiceprint recognition model to obtain a corresponding first voiceprint recognition model; subsequently, based on the validation set, a performance evaluation is performed on the first voiceprint recognition model to obtain a corresponding performance evaluation result; further based on the performance evaluation result, the first voiceprint recognition model is optimized to obtain a second voiceprint recognition model that meets preset performance expectations; finally, the second voiceprint recognition model is used as the target voiceprint recognition model. The present application divides the sample data into a training set and a validation set, and then uses the training set to train the initial voiceprint recognition model based on the target optimization algorithm and the target loss constructed according to the classification loss corresponding to the original speech and the pairing loss corresponding to the adversarial sample to obtain a first voiceprint recognition model. Thereafter, based on the use of the validation set, the performance of the first voiceprint recognition model is evaluated to obtain a performance evaluation result, and then the first voiceprint recognition model is optimized based on the performance evaluation result, so as to obtain a target voiceprint recognition model that meets the preset performance expectations, thereby effectively improving the construction efficiency of the target voiceprint recognition model and ensuring the model effect of the obtained target voiceprint recognition model.

[0141] In some optional implementations of this embodiment, step S207 includes the following steps:

[0142] The speech to be processed is preprocessed to obtain a corresponding second speech.

[0143] In this embodiment, the above-mentioned preprocessing includes: filtering, denoising, endpoint detection, etc. The purpose of preprocessing is to extract clean, impurity-free speech signals from a noisy speech environment to provide high-quality input for subsequent feature extraction and recognition. Specifically, filtering and denoising include: using digital signal processing technology to filter the speech signal to remove background noise and interference and improve the quality of the speech signal. Endpoint detection Boake: Analyze the input audio stream, automatically delete invalid parts such as silence or non-human voice in the audio, and retain valid speech.

[0144] Perform feature extraction on the second speech to obtain corresponding specified features.

[0145] In this embodiment, the feature extraction refers to extracting spectral feature parameters that can characterize the speaker's specific organ structure or behavioral habits from the preprocessed speech signal, that is, the second speech. These feature parameters are relatively stable for the same speaker, do not change with time or environmental changes, are consistent for different speeches of the same speaker, are not easy to imitate and have strong noise resistance. Specifically, linear prediction coefficients (LPC), Mel-frequency cepstral coefficients (MFCC), perceptual linear prediction coefficients (PLP) and the like can be used to extract distinguishing feature parameters from the second speech. After extracting a large number of feature parameters, further feature selection is performed based on the application of feature dimensionality reduction and feature selection algorithms to screen out the most valuable features for voiceprint recognition, that is, to obtain the above-mentioned specified features.

[0146] The designated feature is recognized based on the target voiceprint recognition model to obtain a corresponding voiceprint recognition result.

[0147] In this embodiment, by inputting the above-mentioned specified features into the trained target voiceprint recognition model, the target voiceprint recognition model will perform matching calculations on the input specified features and output the corresponding voiceprint recognition results. The voiceprint recognition results usually include the speaker's identity label and the corresponding confidence level.

[0148] The voiceprint recognition result is outputted based on a preset output method.

[0149] In this embodiment, the generated voiceprint recognition result can be presented to the user in an appropriate output form (such as text, voice, etc.). At the same time, the voiceprint recognition result can be further verified to ensure its accuracy and reliability.

[0150] The present application obtains the corresponding second speech by preprocessing the speech to be processed; then extracts features from the second speech to obtain corresponding designated features; then recognizes and processes the designated features based on the target voiceprint recognition model to obtain corresponding voiceprint recognition results; and subsequently outputs and processes the voiceprint recognition results based on a preset output method. After preprocessing and extracting features from the speech to be processed to obtain designated features, the present application obtains voiceprint recognition results by recognizing and processing the designated features based on the target voiceprint recognition model that has undergone adversarial training, and outputs and processes the voiceprint recognition results based on the output method, thereby effectively improving the recognition accuracy of voiceprint recognition.

[0151] In some optional implementations of this embodiment, after step S206, the electronic device may further perform the following steps:

[0152] Recall the preset specified storage medium.

[0153] In this embodiment, there is no specific limitation on the selection of the designated storage medium. For example, any one of a local database, a cloud server, a network disk, a blockchain, etc. can be used.

[0154] The target voiceprint recognition model is stored based on the designated storage medium.

[0155] In this embodiment, blockchain is preferably used as the above-mentioned designated storage medium. By using blockchain to store and manage the target voiceprint recognition model, the security and non-tamperability of the target voiceprint recognition model can be effectively guaranteed.

[0156] After the target voiceprint recognition model is stored, the target voiceprint recognition model is system deployed.

[0157] In this embodiment, the above-mentioned target voiceprint recognition model can be deployed in an actual generation environment to obtain a voiceprint recognition system that can prevent adversarial sample attacks, and ensure that the voiceprint recognition system can correctly process the original speech and adversarial samples and accurately identify the identity of the speaker.

[0158] This application calls a preset designated storage medium; then stores and processes the target voiceprint recognition model based on the designated storage medium; after completing the storage of the target voiceprint recognition model, the target voiceprint recognition model is system deployed. This application uses a designated storage medium to store and process the target voiceprint recognition model, which can effectively ensure the security and non-tamperability of the target voiceprint recognition model and improve the storage intelligence of the target voiceprint recognition model. In addition, the target voiceprint recognition model will also be intelligently system deployed to achieve rapid and accurate construction of a voiceprint recognition system that can prevent adversarial sample attacks, thereby improving the construction intelligence of the voiceprint recognition system.

[0159] In some optional implementations, the user information obtained is subject to the user's consent and complies with relevant laws and policies.

[0160] In addition, the non-Company software tools or components that appear in the embodiments of the present application are merely examples and do not represent actual use.

[0161] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0162] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned voiceprint recognition results, the above-mentioned voiceprint recognition results can also be stored in a node of a blockchain.

[0163] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0164] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0165] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0166] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0167] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0168] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a voiceprint recognition device based on artificial intelligence, and the device embodiment is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0169] like Figure 3 As shown, the voiceprint recognition device 300 based on artificial intelligence described in this embodiment includes: an acquisition module 301, a preprocessing module 302, a first generation module 303, a second generation module 304, a construction module 305, a training module 306 and a recognition module 307. Among them:

[0170] An acquisition module 301 is used to acquire the original speech collected in advance;

[0171] A preprocessing module 302 is used to preprocess the original speech based on a preset room impulse response strategy to obtain a corresponding target speech;

[0172] The first generating module 303 is used to perform noise generation processing on the target speech based on a preset initial voiceprint recognition model to obtain corresponding adversarial noise;

[0173] A second generating module 304, configured to generate a corresponding adversarial sample based on the original speech and the adversarial noise;

[0174] A construction module 305, configured to construct corresponding sample data based on the target speech and the adversarial sample;

[0175] A training module 306 is used to perform adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model;

[0176] The recognition module 307 is used to perform voiceprint recognition processing on the speech to be processed based on the target voiceprint recognition model to obtain a corresponding voiceprint recognition result.

[0177] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the voiceprint recognition method based on artificial intelligence in the aforementioned implementation mode, and will not be repeated here.

[0178] In some optional implementations of this embodiment, the preprocessing module 302 includes:

[0179] A first calculation submodule is used to calculate the room impulse response when the original speech propagates in a preset environment;

[0180] An operation submodule, configured to perform a convolution operation on the original speech based on the room impulse response to obtain a corresponding first speech;

[0181] The first determination submodule is configured to use the first speech as the target speech.

[0182] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the voiceprint recognition method based on artificial intelligence in the aforementioned implementation mode, and will not be repeated here.

[0183] In some optional implementations of this embodiment, the first generating module 303 includes:

[0184] A first acquisition submodule, used to acquire a first feature representation of the target speech;

[0185] A second acquisition submodule is used to call the initial voiceprint recognition model and obtain an initial loss function of the initial voiceprint recognition model;

[0186] A second calculation submodule, used for calculating the gradient of the initial loss function with respect to the first feature representation based on a fast gradient sign method;

[0187] The first generating submodule is used to generate corresponding adversarial noise based on the gradient.

[0188] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the voiceprint recognition method based on artificial intelligence in the aforementioned implementation mode, and will not be repeated here.

[0189] In some optional implementations of this embodiment, the second generating module 304 includes:

[0190] A third acquisition submodule, used to acquire a second feature representation of the original speech;

[0191] An adding submodule, used for adding the adversarial noise to the second feature representation to obtain a corresponding target feature representation;

[0192] A conversion submodule, used to convert the target feature representation based on a preset reshaping algorithm to obtain a corresponding speech signal;

[0193] The second generating submodule is used to generate the adversarial sample based on the speech signal.

[0194] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the voiceprint recognition method based on artificial intelligence in the aforementioned implementation mode, and will not be repeated here.

[0195] In some optional implementations of this embodiment, the training module 306 includes:

[0196] A division submodule, used for dividing the sample data into a training set and a validation set;

[0197] A construction submodule, configured to construct a target loss based on the classification loss corresponding to the original speech and the pairing loss corresponding to the adversarial sample;

[0198] A training submodule, configured to train the initial voiceprint recognition model using the training set based on a preset target optimization algorithm and the target loss to obtain a corresponding first voiceprint recognition model;

[0199] An evaluation submodule, configured to perform a performance evaluation on the first voiceprint recognition model based on the verification set to obtain a corresponding performance evaluation result;

[0200] An optimization submodule, configured to optimize the first voiceprint recognition model based on the performance evaluation result to obtain a second voiceprint recognition model that meets preset performance expectations;

[0201] The second determining submodule is used to use the second voiceprint recognition model as the target voiceprint recognition model.

[0202] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the voiceprint recognition method based on artificial intelligence in the aforementioned implementation mode, and will not be repeated here.

[0203] In some optional implementations of this embodiment, the identification module 307 includes:

[0204] A preprocessing submodule, used for preprocessing the speech to be processed to obtain a corresponding second speech;

[0205] An extraction submodule, used for performing feature extraction on the second speech to obtain corresponding specified features;

[0206] An identification submodule, configured to perform identification processing on the specified feature based on the target voiceprint recognition model to obtain a corresponding voiceprint recognition result;

[0207] The output submodule is used to output the voiceprint recognition result based on a preset output method.

[0208] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the voiceprint recognition method based on artificial intelligence in the aforementioned implementation mode, and will not be repeated here.

[0209] In some optional implementations of this embodiment, the voiceprint recognition device based on artificial intelligence further includes:

[0210] A calling module, used to call a preset designated storage medium;

[0211] A storage module, used for storing the target voiceprint recognition model based on the designated storage medium;

[0212] A deployment module is used to perform system deployment processing on the target voiceprint recognition model after completing the storage of the target voiceprint recognition model.

[0213] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the voiceprint recognition method based on artificial intelligence in the aforementioned implementation mode, and will not be repeated here.

[0214] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0215] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0216] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0217] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 4. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of the voiceprint recognition method based on artificial intelligence, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0218] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions stored in the memory 41 or process data, such as computer-readable instructions for running the artificial intelligence-based voiceprint recognition method.

[0219] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0220] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0221] In an embodiment of the present application, the pre-collected original speech is preprocessed based on the use of a room impulse response strategy to obtain a target speech, and then the target speech is subjected to noise generation processing based on the initial voiceprint recognition model to obtain adversarial noise, and corresponding adversarial samples are generated based on the original speech and the adversarial noise, and then the initial voiceprint recognition model is adversarially trained based on the sample data constructed from the target speech and the adversarial samples. This can efficiently and accurately construct a target voiceprint recognition model that can effectively prevent adversarial sample attacks, improve the target voiceprint recognition model's defense capability against adversarial attacks, and effectively improve the recognition accuracy and robustness of the target voiceprint recognition model for the voiceprint recognition processing of the processed speech when facing adversarial attacks.

[0222] The present application also provides another implementation, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the above-mentioned artificial intelligence-based voiceprint recognition method.

[0223] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0224] In an embodiment of the present application, the pre-collected original speech is preprocessed based on the use of a room impulse response strategy to obtain a target speech, and then the target speech is subjected to noise generation processing based on the initial voiceprint recognition model to obtain adversarial noise, and corresponding adversarial samples are generated based on the original speech and the adversarial noise, and then the initial voiceprint recognition model is adversarially trained based on the sample data constructed from the target speech and the adversarial samples. This can efficiently and accurately construct a target voiceprint recognition model that can effectively prevent adversarial sample attacks, improve the target voiceprint recognition model's defense capability against adversarial attacks, and effectively improve the recognition accuracy and robustness of the target voiceprint recognition model for the voiceprint recognition processing of the processed speech when facing adversarial attacks.

[0225] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0226] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A voiceprint recognition method based on artificial intelligence, characterized in that: The steps include: Obtaining the original voice collected in advance; Preprocessing the original speech based on a preset room impulse response strategy to obtain a corresponding target speech; Based on the preset initial voiceprint recognition model, noise generation processing is performed on the target voice to obtain the corresponding adversarial noise; Generate a corresponding adversarial sample based on the original speech and the adversarial noise; Constructing corresponding sample data based on the target speech and the adversarial sample; Performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model; Based on the target voiceprint recognition model, voiceprint recognition processing is performed on the speech to be processed to obtain a corresponding voiceprint recognition result.

2. The voiceprint recognition method based on artificial intelligence according to claim 1 is characterized in that: The step of preprocessing the original speech based on a preset room impulse response strategy to obtain a corresponding target speech specifically includes: Calculating the room impulse response when the original speech propagates in a preset environment; Performing a convolution operation on the original speech based on the room impulse response to obtain a corresponding first speech; The first speech is used as the target speech.

3. The voiceprint recognition method based on artificial intelligence according to claim 1 is characterized in that: The step of performing noise generation processing on the target voice based on the preset initial voiceprint recognition model to obtain the corresponding anti-noise specifically includes: Acquire a first feature representation of the target speech; Calling the initial voiceprint recognition model and obtaining an initial loss function of the initial voiceprint recognition model; Calculating the gradient of the initial loss function with respect to the first feature representation based on a fast gradient sign method; Corresponding adversarial noise is generated based on the gradient.

4. The voiceprint recognition method based on artificial intelligence according to claim 1, characterized in that: The step of generating a corresponding adversarial sample based on the original speech and the adversarial noise specifically includes: Acquire a second feature representation of the original speech; Adding the adversarial noise to the second feature representation to obtain a corresponding target feature representation; The target feature representation is converted based on a preset reshaping algorithm to obtain a corresponding speech signal; The adversarial sample is generated based on the speech signal.

5. The voiceprint recognition method based on artificial intelligence according to claim 1, characterized in that: The step of performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model specifically includes: Dividing the sample data into a training set and a validation set; Constructing a target loss based on the classification loss corresponding to the original speech and the pairing loss corresponding to the adversarial sample; Based on a preset target optimization algorithm and the target loss, the initial voiceprint recognition model is trained using the training set to obtain a corresponding first voiceprint recognition model; Performing a performance evaluation on the first voiceprint recognition model based on the verification set to obtain a corresponding performance evaluation result; Optimizing the first voiceprint recognition model based on the performance evaluation result to obtain a second voiceprint recognition model that meets preset performance expectations; The second voiceprint recognition model is used as the target voiceprint recognition model.

6. The voiceprint recognition method based on artificial intelligence according to claim 1, characterized in that: The step of performing voiceprint recognition processing on the processed speech based on the target voiceprint recognition model to obtain the corresponding voiceprint recognition result specifically includes: Preprocessing the speech to be processed to obtain a corresponding second speech; Extracting features from the second speech to obtain corresponding specified features; Performing recognition processing on the designated feature based on the target voiceprint recognition model to obtain a corresponding voiceprint recognition result; The voiceprint recognition result is outputted based on a preset output method.

7. The voiceprint recognition method based on artificial intelligence according to claim 1, characterized in that: After the step of performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model, the method further includes: Call the preset specified storage medium; Performing storage processing on the target voiceprint recognition model based on the designated storage medium; After the target voiceprint recognition model is stored, the target voiceprint recognition model is system deployed.

8. A voiceprint recognition device based on artificial intelligence, characterized in that: include: An acquisition module, used to acquire the original speech collected in advance; A preprocessing module, used for preprocessing the original speech based on a preset room impulse response strategy to obtain a corresponding target speech; A first generation module is used to perform noise generation processing on the target voice based on a preset initial voiceprint recognition model to obtain corresponding adversarial noise; A second generating module, configured to generate a corresponding adversarial sample based on the original speech and the adversarial noise; A construction module, used to construct corresponding sample data based on the target speech and the adversarial sample; A training module, used for performing adversarial training on the initial voiceprint recognition model based on the sample data to obtain a trained target voiceprint recognition model; The recognition module is used to perform voiceprint recognition processing on the speech to be processed based on the target voiceprint recognition model to obtain a corresponding voiceprint recognition result.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the voiceprint recognition method based on artificial intelligence as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the voiceprint recognition method based on artificial intelligence as described in any one of claims 1 to 7 are implemented.

Images