Voiceprint recognition method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a voiceprint recognition method and device, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence and finance. The method comprises the following steps: acquiring sample voice data and a sample speaker tag; performing voiceprint feature distribution calculation on the sample voice data through the original voiceprint extraction model to obtain sample voiceprint Gaussian distribution data; performing voiceprint calculation on the sample voiceprint Gaussian distribution data to obtain sample voiceprint data; performing voiceprint recognition on the sample voiceprint data to obtain sample speaker identity recognition data; performing model optimization on the original voiceprint extraction model based on the sample voiceprint Gaussian distribution data, the sample speaker identity recognition data and the sample speaker tag to obtain a target voiceprint extraction model; performing voiceprint extraction on the original voice data through a target voiceprint extraction model to obtain target voiceprint data; and performing voiceprint recognition based on the target voiceprint data. According to the embodiment of the invention, the voiceprint features can be accurately obtained, and the voiceprint recognition efficiency and accuracy are improved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and financial technology, and in particular to a voiceprint recognition method and device, an electronic device, and a storage medium. Background Art

[0002] Voiceprint recognition is a biometric technology that can be applied in a variety of scenarios, such as identity authentication, anti-fraud identification, telephone banking services, etc. in the financial field.

[0003] At present, deep learning models are mainly used to perform voiceprint recognition on user's voice data. However, in practical applications, voice data is usually mixed with environmental noise, or channel noise is generated when voice data is transmitted. When the noise level is high, it is difficult for deep learning models to accurately extract the user's voiceprint features from the voice data, and thus it is impossible to accurately recognize the voiceprint features, resulting in reduced accuracy of voiceprint recognition.

[0004] Therefore, how to improve the accuracy of voiceprint recognition has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The main purpose of the embodiments of the present application is to provide a voiceprint recognition method and device, an electronic device and a storage medium, aiming to improve the accuracy of voiceprint recognition.

[0006] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a voiceprint recognition method, the method comprising:

[0007] Acquire voiceprint sample data; wherein the voiceprint sample data includes sample speech data and sample speaker labels; wherein the sample speaker labels are used to characterize the identity of the speaker of the sample speech data;

[0008] The voiceprint feature distribution of the sample speech data is calculated by using a preset original voiceprint extraction model to obtain the sample voiceprint Gaussian distribution data;

[0009] Performing voiceprint calculation on the sample voiceprint Gaussian distribution data to obtain sample voiceprint data;

[0010] Performing voiceprint recognition on the sample voiceprint data to obtain sample speaker identification data;

[0011] Optimizing the original voiceprint extraction model based on the sample voiceprint Gaussian distribution data, the sample speaker identification data and the sample speaker label to obtain a target voiceprint extraction model;

[0012] Performing voiceprint extraction on the pre-acquired original voice data through the target voiceprint extraction model to obtain target voiceprint data;

[0013] Voiceprint recognition is performed based on the target voiceprint data.

[0014] In some embodiments, the sample voiceprint Gaussian distribution data includes sample voiceprint distribution mean data and sample voiceprint distribution variance data, and performing voiceprint calculation on the sample voiceprint Gaussian distribution data to obtain the sample voiceprint data includes:

[0015] Acquire the data dimension of the sample voiceprint distribution variance data, and generate random Gaussian noise data based on the data dimension;

[0016] Performing noise processing on the sample voiceprint distribution variance data and the random Gaussian noise data to obtain noise voiceprint feature data;

[0017] The sample voiceprint distribution mean data and the noise voiceprint feature data are aggregated to obtain the sample voiceprint data.

[0018] In some embodiments, the optimizing the original voiceprint extraction model based on the sample voiceprint Gaussian distribution data, the sample speaker identification data and the sample speaker label to obtain the target voiceprint extraction model includes:

[0019] Calculating recognition loss based on the sample speaker identification data and the sample speaker label to obtain speaker recognition loss data;

[0020] Performing a divergence loss calculation based on the sample voiceprint distribution mean data and the sample voiceprint distribution variance data to obtain divergence loss data;

[0021] Performing aggregation calculation on the speaker recognition loss data and the divergence loss data to obtain target loss data;

[0022] The parameters of the original voiceprint extraction model are adjusted based on the target loss data to obtain the target voiceprint extraction model.

[0023] In some embodiments, the performing voiceprint feature distribution calculation on the sample speech data by using a preset original voiceprint extraction model to obtain the sample voiceprint Gaussian distribution data includes:

[0024] Extracting human voice from the sample speech data to obtain sample human voice data;

[0025] Performing spectrum conversion on the sample human voice data to obtain original sample spectrum data;

[0026] Performing standardization processing on the original sample spectrum data to obtain sample spectrum feature data;

[0027] The Gaussian distribution of the sample frequency spectrum feature data is calculated by the original voiceprint extraction model to obtain the Gaussian distribution data of the sample voiceprint.

[0028] In some embodiments, extracting human voice from the sample speech data to obtain the sample human voice data includes:

[0029] Converting the sample voice data into an initial sample voice data.

[0030] Segmenting the initial sample speech data to obtain sample speech segment data;

[0031] Extracting features from the sample speech segment data to obtain speech segment feature data;

[0032] Performing voice detection on the speech segment feature data to obtain the sample voice data.

[0033] In some embodiments, performing voiceprint recognition based on the target voiceprint data includes:

[0034] Performing voiceprint extraction on the pre-acquired target speech data using the target voiceprint extraction model to obtain expected voiceprint data;

[0035] Calculating similarity between the target voiceprint data and the expected voiceprint data to obtain voiceprint similarity data;

[0036] Voiceprint recognition is performed based on the voiceprint similarity data.

[0037] In some embodiments, performing voiceprint extraction on the pre-acquired original voice data by using the target voiceprint extraction model to obtain the target voiceprint data includes:

[0038] Extracting human voice from the original voice data to obtain original human voice data;

[0039] Performing spectrum conversion on the original human voice data to obtain original human voice spectrum data;

[0040] Performing standardization processing on the original human voice spectrum data to obtain target spectrum feature data;

[0041] Performing Gaussian distribution calculation on the target frequency spectrum feature data through the target voiceprint extraction model to obtain target voiceprint Gaussian distribution data;

[0042] Perform voiceprint calculation on the target voiceprint Gaussian distribution data to obtain the target voiceprint data.

[0043] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application provides a voiceprint recognition device, the device comprising:

[0044] A sample data acquisition module, used to acquire voiceprint sample data; wherein the voiceprint sample data includes sample speech data and sample speaker labels; wherein the sample speaker labels are used to characterize the speaker identity of the sample speech data;

[0045] An original voiceprint feature distribution calculation module is used to calculate the voiceprint feature distribution of the sample speech data through a preset original voiceprint extraction model to obtain the sample voiceprint Gaussian distribution data;

[0046] An original voiceprint calculation module, used to perform voiceprint calculation on the sample voiceprint Gaussian distribution data to obtain sample voiceprint data;

[0047] An original voiceprint recognition module, used to perform voiceprint recognition on the sample voiceprint data to obtain sample speaker identification data;

[0048] A model optimization module, configured to optimize the original voiceprint extraction model based on the sample voiceprint Gaussian distribution data, the sample speaker identification data and the sample speaker label to obtain a target voiceprint extraction model;

[0049] A target voiceprint extraction module is used to extract voiceprints from the pre-acquired original voice data using the target voiceprint extraction model to obtain target voiceprint data;

[0050] The target voiceprint recognition module is used to perform voiceprint recognition based on the target voiceprint data.

[0051] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0052] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0053] The voiceprint recognition method and device, electronic device and storage medium proposed in the present application obtain voiceprint sample data; wherein the voiceprint sample data includes sample speech data and sample speaker labels; wherein the sample speaker labels are used to characterize the speaker identity of the sample speech data; voiceprint feature distribution calculation is performed on the sample speech data through a preset original voiceprint extraction model to obtain sample voiceprint Gaussian distribution data; voiceprint calculation is performed on the sample voiceprint Gaussian distribution data to obtain sample voiceprint data; voiceprint recognition is performed on the sample voiceprint data to obtain sample speaker identity recognition data; based on the sample voiceprint Gaussian distribution data, the sample speaker identity recognition data and the sample speaker label, the original voiceprint extraction model is optimized to obtain a target voiceprint extraction model; voiceprint extraction is performed on the pre-acquired original speech data through the target voiceprint extraction model to obtain the target voiceprint data; voiceprint recognition is performed based on the target voiceprint data, and the trained target voiceprint extraction model can more accurately obtain voiceprint features with high discrimination, thereby improving the efficiency and accuracy of voiceprint recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flow chart of the voiceprint recognition method provided by an embodiment of the present application;

[0055] Figure 2 yes Figure 1 Flow chart of step S102 in FIG.

[0056] Figure 3 yes Figure 2 Flow chart of step S201 in FIG.

[0057] Figure 4 yes Figure 1 Flow chart of step S103 in FIG.

[0058] Figure 5 yes Figure 1 Flow chart of step S105 in FIG.

[0059] Figure 6 yes Figure 1 Flow chart of step S106 in FIG.

[0060] Figure 7 yes Figure 1 Flow chart of step S107 in FIG.

[0061] Figure 8 is a schematic diagram of the structure of a voiceprint recognition device provided in an embodiment of the present application;

[0062] Fig. 9 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0064] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0066] First, some nouns involved in this application are analyzed:

[0067] Artificial intelligence (AI) is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems.

[0068] Voiceprint recognition is a biometric technology that authenticates identity based on voice features. It confirms the speaker's identity by analyzing and identifying the unique acoustic features (i.e., voiceprints) in an individual's voice. This technology uses the differences in pronunciation, speaking speed, intonation, etc. of each speaker to achieve efficient, contactless identity authentication by comparing the similarities between known voiceprint features and those to be verified.

[0069] Information Bottleneck Theory, Information Bottleneck Theory is a theoretical framework for analyzing and optimizing information transmission and compression. It aims to retain the key information of the input data while maximally compressing the data representation, that is, given the input and output, find an optimal intermediate representation (i.e., bottleneck) to maximize the mutual information between the input and output, while minimizing the information entropy of the intermediate representation, thereby achieving effective encoding of information. Simply put, the information bottleneck theory improves data compression efficiency by minimizing the information entropy after encoding, ensuring that the burden of storage and transmission is reduced while maintaining data quality. The information bottleneck theory has a wide range of applications in machine learning and deep learning, especially in feature selection, model simplification, and data compression.

[0070] Kullback-Leibler divergence, also known as KL divergence or relative entropy, is an asymmetric measure of the difference between two probability distributions P and Q, indicating the amount of information loss when using Q to approximate P. In short, KL divergence quantifies the degree of difference between one probability distribution and another, but it is not a true distance measure because it does not satisfy symmetry and triangle inequality. In machine learning and statistics, KL divergence is often used to evaluate the difference between the model's predicted distribution and the true distribution to guide the model's training so that the predicted distribution gradually approaches the target distribution. It is an important tool for optimizing and evaluating model performance.

[0071] In the financial field, voiceprint recognition technology is often used for payment verification, identity verification, anti-fraud identification, telephone banking services, etc., which can enhance the security and convenience of business operations.

[0072] At present, deep learning models are mainly used to perform voiceprint recognition on user voice data, such as I-Vector model, TDNN model, ResNet model, etc. However, in practical applications, voice data is usually mixed with environmental noise, or channel noise is generated when voice data is transmitted. When the noise level is high, it is difficult for deep learning models to accurately extract user voiceprint features from voice data, and thus it is impossible to accurately recognize voiceprint features, resulting in reduced accuracy of voiceprint recognition.

[0073] Based on this, the embodiments of the present application provide a voiceprint recognition method and device, an electronic device and a storage medium, aiming to improve the accuracy of voiceprint recognition.

[0074] The voiceprint recognition method and device, electronic device and storage medium provided in the embodiments of the present application are specifically described through the following embodiments. First, the voiceprint recognition method in the embodiments of the present application is described.

[0075] 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.

[0076] 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.

[0077] The voiceprint recognition method provided in the embodiment of the present application relates to the field of artificial intelligence and financial technology. The voiceprint recognition method provided in the embodiment of the present application can be applied to a terminal, can be applied to a server, or can be software running in a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the voiceprint recognition method, etc., but is not limited to the above forms.

[0078] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0079] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0080] Figure 1 is an optional flow chart of the voiceprint recognition method provided in the embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S107.

[0081] Step S101, obtaining voiceprint sample data; wherein the voiceprint sample data includes sample speech data and sample speaker labels; wherein the sample speaker labels are used to characterize the speaker identity of the sample speech data;

[0082] Step S102, calculating the voiceprint feature distribution of the sample speech data using a preset original voiceprint extraction model to obtain Gaussian distribution data of the sample voiceprint;

[0083] Step S103, performing voiceprint calculation on the sample voiceprint Gaussian distribution data to obtain sample voiceprint data;

[0084] Step S104, performing voiceprint recognition on the sample voiceprint data to obtain sample speaker identification data;

[0085] Step S105, optimizing the original voiceprint extraction model based on the sample voiceprint Gaussian distribution data, the sample speaker identification data and the sample speaker label to obtain a target voiceprint extraction model;

[0086] Step S106, performing voiceprint extraction on the pre-acquired original voice data through a target voiceprint extraction model to obtain target voiceprint data;

[0087] Step S107: performing voiceprint recognition based on the target voiceprint data.

[0088] In steps S101 to S107 shown in the embodiment of the present application, the original voiceprint extraction model is optimized by using sample speech data and corresponding sample speaker labels, thereby improving the model's step capability and recognition accuracy of voiceprint features, and obtaining a target voiceprint extraction model. Therefore, when voiceprint extraction is performed on the original speech data, voiceprint features with high discrimination can be obtained more accurately, thereby improving the efficiency and accuracy of voiceprint recognition.

[0089] In some embodiments, the voiceprint sample data is pre-collected by technical personnel. During the collection process, a combination of automatic machine verification and manual spot checks is used to ensure that the voice quality of the voiceprint sample data meets the preset quality requirements.

[0090] Each voiceprint sample data appears in pairs, including sample voice data and sample speaker labels, wherein the sample speaker labels are used to characterize the speaker identity of the sample voice data; for example: the sample voice data is voice 1, and the sample speaker label is the name of user A.

[0091] See also Figure 2 In some embodiments, step S102 may include but is not limited to steps S201 to S204:

[0092] Step S201, extracting human voice from sample speech data to obtain sample human voice data;

[0093] Step S202, performing spectrum conversion on the sample human voice data to obtain original sample spectrum data;

[0094] Step S203, normalizing the original sample spectrum data to obtain sample spectrum feature data;

[0095] Step S204, performing Gaussian distribution calculation on the sample frequency spectrum feature data through the original voiceprint extraction model to obtain Gaussian distribution data of the sample voiceprint.

[0096] In steps S201 to S204 shown in the embodiment of the present application, sample spectrum feature data is obtained by successively performing human voice extraction, spectrum conversion, and standardization on the sample speech data, and then Gaussian distribution calculation is performed on the sample spectrum feature data through the original voiceprint extraction model to obtain sample voiceprint Gaussian distribution data. This can effectively capture and characterize the probability distribution of voiceprint features, not only retain the key information of the voiceprint, but also facilitate the subsequent voiceprint matching and recognition algorithms to perform efficient and accurate calculations, thereby improving the accuracy of voiceprint recognition.

[0097] Before step S201 in some embodiments, the voiceprint recognition method further includes but is not limited to the following steps:

[0098] Perform data augmentation processing on the voiceprint sample data.

[0099] The data augmentation processing includes, but is not limited to, adding noise to the sample speech data, doubling the speed, and random mosaic of the feature graph.

[0100] In step S201 of some embodiments, human voice extraction can be performed on sample voice data through a voice activation detection method to obtain sample voice data; wherein the voice activation detection method can be a Silero-VAD algorithm, or a VAD model of WebRTC, etc., but is not limited thereto.

[0101] See also Figure 3 In some embodiments, step S201 may include but is not limited to steps S301 to S304:

[0102] Step S301, converting the format of the sample voice data to obtain initial sample voice data;

[0103] Step S302, segmenting the initial sample speech data to obtain sample speech segment data;

[0104] Step S303, extracting features from the sample speech segment data to obtain speech segment feature data;

[0105] Step S304: perform voice detection on the speech segment feature data to obtain sample voice data.

[0106] In steps S301 to S304 shown in the embodiment of the present application, by converting the format of the sample voice data before voiceprint recognition, it is possible to ensure that the sample voice data remains consistent in subsequent processing and reduce errors caused by format differences; segmentation processing helps to decompose long voice data into segments that are easier to process, so as to capture the voiceprint features in each segment. Feature extraction further refines key information useful for voiceprint recognition and eliminates redundant data. Voice detection ensures that only the valid part containing the human voice is used for voiceprint recognition, eliminating interference from non-human voice parts such as background noise and silent segments. This series of preprocessing steps work together to significantly improve the accuracy and efficiency of voiceprint recognition, laying a solid foundation for subsequent voiceprint matching and recognition.

[0107] In step S301 of some embodiments, the format of the sample speech data is converted, for example, MP3, FLAC and other formats are converted into WAV format, and the sampling rate is unified (for example, 16000 Hz), which helps to unify the input of the model, reduce errors caused by format differences, and improve processing efficiency.

[0108] In step S302 of some embodiments, the continuous initial sample voice data is divided into a plurality of shorter voice segment data, for example, each segment is 600 ms long, which can reduce the computing resource occupation of a single processing and improve processing efficiency.

[0109] In step S303 of one embodiment, the sample speech segment data is converted into a time-frequency representation, and feature extraction is performed to facilitate further determination of whether a human voice exists in the speech segment data.

[0110] Specifically, feature extraction can be performed through a pre-trained neural network model, such as a RoBERTa model, an ALBERT model, etc., but not limited thereto.

[0111] In step S304 of some embodiments, each voice segment feature data is analyzed to obtain confidence data indicating whether the voice segment feature data contains voice activity, and the confidence data is compared with a preset confidence threshold to obtain the voice segment type. Specifically, the voice segment feature data whose confidence data exceeds the confidence threshold is determined as a non-silent segment, and the voice segment feature data whose confidence data is lower than the confidence threshold is determined as a silent segment.

[0112] Then, based on the voice segment type, the sample voice segment data is screened to screen out all non-silent segments to obtain human voice sample data.

[0113] In step S202 of some embodiments, the original sample spectrum data can be extracted from the sample human voice data by using the Fbank method; wherein the Fbank method can simulate the nonlinear response of the human ear to the sound spectrum and process the audio to improve the performance of voiceprint recognition. Specifically, the original sample spectrum data is Mel spectrum data of a preset dimension.

[0114] In some other embodiments, the spectrum conversion of the sample human voice data can also be achieved by fast Fourier transform (FFT), short-time Fourier transform (STFT), wavelet transform, Mel-frequency cepstral coefficient (MFCC), etc. Specifically, it is necessary to make a selection in combination with the actual application scenario, but it is not limited thereto.

[0115] In step S203 of some embodiments, the original sample spectrum data can be standardized by maximum-minimum standardization, Z-score standardization, normalization, etc., and the original sample spectrum data can be scaled to a specific range to obtain sample spectrum feature data.

[0116] In step S204 of some embodiments, the sample voiceprint Gaussian distribution data includes sample voiceprint distribution mean data and sample voiceprint distribution variance data.

[0117] For example, assuming that the sample spectrum feature data is a vector with a dimension of 256, the data of the 0th to 127th dimensions can be used as the sample voiceprint distribution mean data, and the data of the 128th to 155th dimensions can be used as the sample voiceprint distribution variance data.

[0118] See also Figure 4 In some embodiments, step S103 may include but is not limited to steps S401 to S403:

[0119] Step S401, obtaining the data dimension of the sample voiceprint distribution variance data, and generating random Gaussian noise data based on the data dimension;

[0120] Step S402, performing noise processing on the sample voiceprint distribution variance data and the random Gaussian noise data to obtain the noised voiceprint feature data;

[0121] Step S403, performing aggregation processing on the sample voiceprint distribution mean data and the noise voiceprint feature data to obtain the sample voiceprint data.

[0122] Steps S401 to S403 shown in the embodiment of the present application are to obtain the data dimensions of the sample voiceprint distribution variance data, and generate random Gaussian noise data based on the data dimensions; perform noise processing on the sample voiceprint distribution variance data and the random Gaussian noise data to obtain the noisy voiceprint feature data; perform aggregation processing on the sample voiceprint distribution mean data and the noisy voiceprint feature data to obtain the sample voiceprint data, which can simulate the uncertainty in the real environment, so that when faced with new and unseen voiceprint samples, it can maintain a high recognition accuracy, thereby improving the robustness and generalization ability of the voiceprint extraction model.

[0123] In some embodiments, the data dimension of the sample voiceprint distribution variance data is determined, and then random sampling is performed from high-dimensional Gaussian distribution data according to the data dimension to obtain random Gaussian noise data.

[0124] It can be understood that the data dimension of the random Gaussian noise data is the same as the data dimension of the sample voiceprint distribution variance data;

[0125] In some embodiments, the calculation process of the sample voiceprint data can be calculated as follows:

[0126] z=μ+σ×∈,∈~N(0,1) (1);

[0127] Among them, z is the sample voiceprint data, μ is the sample voiceprint distribution mean data, σ is the sample voiceprint distribution variance data, and ∈ is the random Gaussian noise data.

[0128] In step S104 of some embodiments, voiceprint recognition is performed on the sample voiceprint data using the original speaker recognition model to obtain sample speaker identification data;

[0129] Among them, the original voiceprint extraction model and the original speaker recognition model communicate to form a complete original voiceprint recognition model. The original voiceprint extraction model is first used to process the voiceprint data, and then the voiceprint data is input into the original speaker recognition model for voiceprint recognition, thereby obtaining the speaker's identity recognition data, that is, the speaker identity.

[0130] See also Figure 5 In some embodiments, step S105 may also include but is not limited to steps S501 to S504:

[0131] Step S501, calculating recognition loss based on sample speaker identification data and sample speaker labels to obtain speaker recognition loss data;

[0132] Step S502, performing a divergence loss calculation based on the sample voiceprint distribution mean data and the sample voiceprint distribution variance data to obtain divergence loss data;

[0133] Step S503, performing aggregation calculation on the speaker recognition loss data and the divergence loss data to obtain target loss data;

[0134] Step S504: adjusting parameters of the original voiceprint extraction model based on the target loss data to obtain a target voiceprint extraction model.

[0135] In the steps S501 to S504 shown in the embodiment of the present application, the information bottleneck theory is introduced when training the original voiceprint extraction model, and in addition to calculating the speaker recognition loss data that can characterize the prediction difference, the process of calculating the loss data of Kullback-Leibler divergence is added to limit the complexity of the model output information, forcing it to retain only the features that are critical to speaker identification and remove redundant information, so that the model can learn a more accurate voiceprint representation. This can not only improve the generalization ability of the model and reduce the risk of overfitting, but also make the voiceprint features more concise and efficient while maintaining high recognition accuracy, which is conducive to subsequent storage, transmission and processing.

[0136] It can be understood that the model optimization step involved in the training process of the original voiceprint recognition model will optimize the original voiceprint extraction model and the original speaker recognition model at the same time. Therefore, the specific steps in step S105 should be:

[0137] Based on the sample voiceprint Gaussian distribution data, sample speaker identification data and sample speaker labels, the original voiceprint extraction model and the original speaker identification model are optimized to obtain the target voiceprint extraction model.

[0138] Specifically, the basic assumption of the information bottleneck theory is:

[0139] First, the voiceprint data compresses the voice data as much as possible;

[0140] Second, the voiceprint data should have as much information as possible for the predicted speaker identification data. In other words, this hypothesis hopes that the extracted voiceprint data will only retain the speaker's identity information as much as possible.

[0141] Therefore, mutual information can be used to describe the amount of information between two random variables, and the above assumptions can be modeled as maximizing the following objective function:

[0142] R IB =I(z,y;θ)-λI(z,x;θ) (2);

[0143] Among them, R IB is the objective function, I(·,·) is the amount of information, x is the speech data, z is the voiceprint data, y is the speaker identification data, θ is the parameters of the original voiceprint recognition model (including the parameters of the original voiceprint extraction model and the parameters of the original speaker identification model); λ is the hyperparameter of the amount of information throughput.

[0144] In order to facilitate model optimization, the objective function R IB Transformed into:

[0145] L IB =L CLS +λL KL (3);

[0146] Among them, L IB is the target loss data, L CLS Loss data for speaker recognition, L KL is the divergence loss data.

[0147] Speaker recognition loss data L CLS It can be calculated through cross entropy loss function, mean square error loss function, etc.

[0148] Divergence loss data L KL The calculation method is as follows:

[0149]

[0150] Among them, N is the number of voiceprint sample data, μ is the mean data of the sample voiceprint distribution, and σ is the variance data of the sample voiceprint distribution.

[0151] In step S504 of some embodiments, parameters of the original voiceprint extraction model and the original speaker recognition model are adjusted based on the target loss data to obtain a target voiceprint extraction model.

[0152] Specifically, the Adam algorithm may be used to perform gradient descent to complete step S504.

[0153] In some embodiments, the original voiceprint extraction model can be understood as an "encoder", and models such as ResNet and TDNN can be selected; and the original speaker recognition model can be a CNN, RNN, SVM or other models, or a fully connected layer, etc., but is not limited to this.

[0154] In some embodiments, the target voiceprint extraction model includes a target voiceprint extraction model and a target speaker recognition model.

[0155] See also Figure 6 In some embodiments, step S106 includes but is not limited to steps S601 to S605:

[0156] Step S601, extracting human voice from original speech data to obtain original human voice data;

[0157] Step S602, performing spectrum conversion on the original human voice data to obtain original human voice spectrum data;

[0158] Step S603, normalizing the original human voice spectrum data to obtain target spectrum feature data;

[0159] Step S604, performing Gaussian distribution calculation on the target frequency spectrum feature data through the target voiceprint extraction model to obtain Gaussian distribution data of the target voiceprint;

[0160] Step S605, performing voiceprint calculation on the target voiceprint Gaussian distribution data to obtain the target voiceprint data.

[0161] Steps S601 to S605 shown in the embodiment of the present application are to extract human voice, convert spectrum, and standardize the original voice data in sequence to obtain the original human voice spectrum data, and then use the trained target voiceprint extraction model to perform Gaussian distribution calculation on the target spectrum feature data to obtain the target voiceprint Gaussian distribution data, which can fully utilize the effective feature extraction and expression capabilities learned by the target voiceprint extraction model to automatically, quickly, and accurately extract the target voiceprint Gaussian distribution data rich in recognition information. Further, voiceprint calculation is performed on the target voiceprint Gaussian distribution data to obtain accurate target voiceprint data, which significantly improves the accuracy of voiceprint recognition.

[0162] In some embodiments, step S601 may include but is not limited to the following steps:

[0163] Convert the format of the original voice data to obtain initial voice data;

[0164] Segmenting the initial speech data to obtain target speech segment data;

[0165] Extracting features of the target speech segment data to obtain target segment feature data;

[0166] Perform voice detection on the target segment feature data to obtain original voice data.

[0167] Specifically, the specific implementation scheme composed of the sub-steps included in step S601 is the same as the specific implementation schemes of the above-mentioned steps S301 to S304, and will not be repeated here.

[0168] In step S602 of some embodiments, the original human voice data is spectrally converted by the same method as step S202, and the obtained original human voice spectrum data is the same as the original sample spectrum data in form and data dimension, thereby ensuring the consistency of the data format and reducing errors caused by format differences.

[0169] In step S603 of some embodiments, the original human voice spectrum data is standardized by the same method as step S203, and the obtained target spectrum feature data is the same as the sample spectrum feature data in form and data dimension, thereby ensuring the consistency of the data format and reducing errors caused by format differences.

[0170] In step S604 of some embodiments, the target voiceprint Gaussian distribution data includes target voiceprint distribution mean data and target voiceprint distribution variance data.

[0171] Moreover, the target voiceprint Gaussian distribution data has the same data dimension as the sample spectrum feature data, and the division method of the target voiceprint distribution mean data and the target voiceprint distribution variance data is the same as the division method of the sample voiceprint distribution mean data and the sample voiceprint distribution variance data, which will not be repeated here.

[0172] In step S605 of an embodiment, voiceprint calculation can be performed based on the target voiceprint distribution mean data and the target voiceprint distribution variance data to obtain the target voiceprint data. The specific calculation method is the same as the specific implementation of steps S401 to S403, and will not be repeated here.

[0173] In step S605 of another embodiment, the target voiceprint distribution mean data may be directly used as the target voiceprint data.

[0174] It should be noted that the method for voiceprint calculation needs to be selected according to the actual application scenario and is not limited to this.

[0175] See also Figure 7 In some embodiments, in a one-to-one voiceprint verification scenario, step S107 may include but is not limited to steps S701 to S703:

[0176] Step S701, performing voiceprint extraction on the pre-acquired target speech data through a target voiceprint extraction model to obtain expected voiceprint data;

[0177] Step S702, calculating the similarity between the target voiceprint data and the expected voiceprint data to obtain voiceprint similarity data;

[0178] Step S703: Perform voiceprint recognition based on the voiceprint similarity data.

[0179] In the step S701 to step S703 shown in the embodiment of the present application, in a one-to-one voiceprint verification scenario, the target voiceprint extraction model is used to extract the target voiceprint data to obtain the expected voiceprint data, and then the similarity between the target voiceprint data and the expected voiceprint data is calculated to obtain the voiceprint similarity data; thus, the speaker of the original voice data and the speaker of the target voice data are judged to be the same speaker based on the voiceprint similarity data. This not only simplifies the complexity of identity authentication and improves the verification speed, but also enhances the accuracy and reliability of voiceprint verification, effectively resisting security risks such as impersonation.

[0180] In some embodiments, one-to-one voiceprint verification scenarios such as payment verification, identity verification, anti-fraud identification, telephone banking services, etc. in the financial field can enhance the security and convenience of business operations.

[0181] In some embodiments, the target voice data is pre-recorded voice data of the target user, which is used for voiceprint comparison when identity authentication is required, wherein the pre-processing process of the target voice data is the same as the pre-processing process of the original voice data and will not be repeated.

[0182] In step S702 of some embodiments, the probabilistic linear discriminant analysis method (PLDA) can be used to calculate the similarity, or the linear discriminant analysis method (LDA), Gaussian mixture model (GMM), etc. can be used to calculate the similarity. The specific selection needs to be made in combination with the actual application scenario, but is not limited to this.

[0183] In step S703 of some embodiments, the voiceprint similarity data is compared with a preset similarity threshold. If the voiceprint similarity data exceeds the similarity threshold, it is considered that they belong to the same speaker and the voiceprint verification passes; if the voiceprint similarity data is less than the similarity threshold, it is considered that they do not belong to the same speaker and the voiceprint verification fails.

[0184] In some embodiments, in a one-to-many voiceprint verification scenario, for example, voiceprint verification is applied to access control.

[0185] Step S107 may include but is not limited to the following steps:

[0186] Acquire a target voiceprint database; wherein the target voiceprint database includes predetermined voiceprint data;

[0187] Voiceprint recognition is performed using the target speaker recognition model and the established voiceprint data to obtain the target speaker identity recognition data.

[0188] It should be noted that each predetermined voiceprint data in the target voiceprint database is extracted by the target voiceprint extraction model proposed in the embodiment of the present application, and the specific implementation process will not be repeated here.

[0189] The target speaker recognition model calculates the similarity between the target voiceprint data and each predetermined voiceprint data to obtain a voiceprint similarity sequence, and the voiceprint similarity sequence is screened based on a similarity threshold to obtain a potential similarity data series, wherein the number of data in the potential similarity data series is greater than or equal to zero;

[0190] If the amount of data in the potential similarity data series is greater than zero, the established voiceprint data with the highest similarity is selected from the potential similarity data series to obtain the speaker identity corresponding to the original voice data, that is, the target speaker identity recognition data.

[0191] If the number of data in the potential similarity data series is equal to zero, the speaker identity corresponding to the original speech data cannot be found, that is, the target speaker identity recognition data is empty.

[0192] See also Figure 8 The embodiment of the present application also provides a voiceprint recognition device, which can implement the above-mentioned voiceprint recognition method, and the device includes:

[0193] The sample data acquisition module 801 is used to acquire voiceprint sample data; wherein the voiceprint sample data includes sample speech data and sample speaker labels; wherein the sample speaker labels are used to characterize the speaker identity of the sample speech data;

[0194] The original voiceprint feature distribution calculation module 802 is used to calculate the voiceprint feature distribution of the sample speech data through a preset original voiceprint extraction model to obtain the sample voiceprint Gaussian distribution data;

[0195] The original voiceprint calculation module 803 is used to perform voiceprint calculation on the sample voiceprint Gaussian distribution data to obtain the sample voiceprint data;

[0196] The original voiceprint recognition module 804 is used to perform voiceprint recognition on the sample voiceprint data to obtain the sample speaker identification data;

[0197] A model optimization module 805 is used to optimize the original voiceprint extraction model based on the sample voiceprint Gaussian distribution data, the sample speaker identification data and the sample speaker label to obtain a target voiceprint extraction model;

[0198] The target voiceprint extraction module 806 is used to extract the voiceprint of the original speech data acquired in advance by using the target voiceprint extraction model to obtain the target voiceprint data;

[0199] The target voiceprint recognition module 807 is used to perform voiceprint recognition based on the target voiceprint data.

[0200] The specific implementation of the voiceprint recognition device is basically the same as the specific embodiment of the above-mentioned voiceprint recognition method, and will not be repeated here.

[0201] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned voiceprint recognition method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0202] See also Fig. 9 , Fig. 9 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0203] The processor 901 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0204] The memory 902 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 calls and executes the voiceprint recognition method of the embodiment of this application;

[0205] Input / output interface 903, used to implement information input and output;

[0206] Communication interface 904, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0207] A bus 905 that transmits information between various components of the device (e.g., the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);

[0208] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .

[0209] The embodiment of the present application further provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the above-mentioned voiceprint recognition method is implemented.

[0210] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0211] The voiceprint recognition method and device, electronic device and storage medium provided in the embodiments of the present application obtain voiceprint sample data; wherein the voiceprint sample data includes sample speech data and sample speaker labels; wherein the sample speaker labels are used to characterize the speaker identity of the sample speech data; voiceprint feature distribution calculation is performed on the sample speech data through a preset original voiceprint extraction model to obtain sample voiceprint Gaussian distribution data; voiceprint calculation is performed on the sample voiceprint Gaussian distribution data to obtain sample voiceprint data; voiceprint recognition is performed on the sample voiceprint data to obtain sample speaker identity recognition data; based on the sample voiceprint Gaussian distribution data, the sample speaker identity recognition data and the sample speaker label, the original voiceprint extraction model is optimized to obtain a target voiceprint extraction model; voiceprint extraction is performed on the pre-acquired original speech data through the target voiceprint extraction model to obtain the target voiceprint data; voiceprint recognition is performed based on the target voiceprint data, and the trained target voiceprint extraction model can more accurately obtain voiceprint features with high discrimination, thereby improving the efficiency and accuracy of voiceprint recognition.

[0212] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0213] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0214] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0215] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0216] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0217] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0218] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0219] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0220] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0221] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0222] The preferred embodiments of the present application are described above with reference to the accompanying drawings, but the scope of the rights of the present application is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present application should be within the scope of the rights of the present application.

Claims

1. A voiceprint recognition method, characterized in that: The method comprises: Acquire voiceprint sample data; wherein the voiceprint sample data includes sample speech data and sample speaker labels; wherein the sample speaker labels are used to characterize the identity of the speaker of the sample speech data; The voiceprint feature distribution of the sample speech data is calculated by using a preset original voiceprint extraction model to obtain the sample voiceprint Gaussian distribution data; Performing voiceprint calculation on the sample voiceprint Gaussian distribution data to obtain sample voiceprint data; Performing voiceprint recognition on the sample voiceprint data to obtain sample speaker identification data; Optimizing the original voiceprint extraction model based on the sample voiceprint Gaussian distribution data, the sample speaker identification data and the sample speaker label to obtain a target voiceprint extraction model; Performing voiceprint extraction on the pre-acquired original voice data through the target voiceprint extraction model to obtain target voiceprint data; Voiceprint recognition is performed based on the target voiceprint data.

2. The method according to claim 1, characterized in that The sample voiceprint Gaussian distribution data includes sample voiceprint distribution mean data and sample voiceprint distribution variance data, and performing voiceprint calculation on the sample voiceprint Gaussian distribution data to obtain the sample voiceprint data includes: Acquire the data dimension of the sample voiceprint distribution variance data, and generate random Gaussian noise data based on the data dimension; Performing noise processing on the sample voiceprint distribution variance data and the random Gaussian noise data to obtain noise voiceprint feature data; The sample voiceprint distribution mean data and the noise voiceprint feature data are aggregated to obtain the sample voiceprint data.

3. The method according to claim 2, characterized in that The optimizing the original voiceprint extraction model based on the sample voiceprint Gaussian distribution data, the sample speaker identification data and the sample speaker label to obtain a target voiceprint extraction model includes: Calculating recognition loss based on the sample speaker identification data and the sample speaker label to obtain speaker recognition loss data; Performing a divergence loss calculation based on the sample voiceprint distribution mean data and the sample voiceprint distribution variance data to obtain divergence loss data; Performing aggregation calculation on the speaker recognition loss data and the divergence loss data to obtain target loss data; The parameters of the original voiceprint extraction model are adjusted based on the target loss data to obtain the target voiceprint extraction model.

4. The method according to claim 1, characterized in that The method of performing voiceprint feature distribution calculation on the sample speech data by using a preset original voiceprint extraction model to obtain sample voiceprint Gaussian distribution data includes: Extracting human voice from the sample speech data to obtain sample human voice data; Performing spectrum conversion on the sample human voice data to obtain original sample spectrum data; Performing standardization processing on the original sample spectrum data to obtain sample spectrum feature data; The Gaussian distribution of the sample frequency spectrum feature data is calculated by the original voiceprint extraction model to obtain the Gaussian distribution data of the sample voiceprint.

5. The method according to claim 4, characterized in that The extracting human voice from the sample voice data to obtain the sample human voice data comprises: Converting the sample voice data into an initial sample voice data. Segmenting the initial sample speech data to obtain sample speech segment data; Extracting features from the sample speech segment data to obtain speech segment feature data; Performing voice detection on the speech segment feature data to obtain the sample voice data.

6. The method according to any one of claims 1 to 5, characterized in that: The performing voiceprint recognition based on the target voiceprint data includes: Performing voiceprint extraction on the pre-acquired target speech data using the target voiceprint extraction model to obtain expected voiceprint data; Calculating similarity between the target voiceprint data and the expected voiceprint data to obtain voiceprint similarity data; Voiceprint recognition is performed based on the voiceprint similarity data.

7. The method according to any one of claims 1 to 5, characterized in that: The method of performing voiceprint extraction on the pre-acquired original voice data by using the target voiceprint extraction model to obtain target voiceprint data includes: Extracting human voice from the original voice data to obtain original human voice data; Performing spectrum conversion on the original human voice data to obtain original human voice spectrum data; Performing standardization processing on the original human voice spectrum data to obtain target spectrum feature data; Performing Gaussian distribution calculation on the target frequency spectrum feature data through the target voiceprint extraction model to obtain target voiceprint Gaussian distribution data; Perform voiceprint calculation on the target voiceprint Gaussian distribution data to obtain the target voiceprint data.

8. A voiceprint recognition device, characterized in that: The device comprises: A sample data acquisition module, used to acquire voiceprint sample data; wherein the voiceprint sample data includes sample speech data and sample speaker labels; wherein the sample speaker labels are used to characterize the speaker identity of the sample speech data; An original voiceprint feature distribution calculation module is used to calculate the voiceprint feature distribution of the sample speech data through a preset original voiceprint extraction model to obtain the sample voiceprint Gaussian distribution data; An original voiceprint calculation module, used to perform voiceprint calculation on the sample voiceprint Gaussian distribution data to obtain sample voiceprint data; An original voiceprint recognition module, used to perform voiceprint recognition on the sample voiceprint data to obtain sample speaker identification data; A model optimization module, configured to optimize the original voiceprint extraction model based on the sample voiceprint Gaussian distribution data, the sample speaker identification data and the sample speaker label to obtain a target voiceprint extraction model; A target voiceprint extraction module is used to extract voiceprints from the pre-acquired original voice data using the target voiceprint extraction model to obtain target voiceprint data; The target voiceprint recognition module is used to perform voiceprint recognition based on the target voiceprint data.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images