A self-supervised learning voiceprint recognition model training method, device and readable medium

Abstract

The present invention discloses a self-supervised learning voiceprint recognition model training method, device and readable medium, which is mainly based on the fact that the data obtained from the actual application scenario has no label or the label is wrong, and the real scene data and continuous model optimization are effectively utilized by means of the optimized self-supervised learning voiceprint recognition model training method. The optimized self-supervised learning voiceprint recognition model training method uses three classifications to classify the training data, and reasonably selects the training data for training according to the state of the voiceprint recognition model, so that the voiceprint recognition model obtains different capabilities at different stages. The first training and classification enables it to more accurately distinguish data with large differences in voiceprint features, the second training and classification enables it to have the characteristics of being able to distinguish people with similar voiceprints, and the third training and classification enables the model to converge and maintain a fixed classification feature network, which is conducive to the convergence and stability of the voiceprint recognition model.

Description

Technical Field

[0001] The present invention relates to the field of voiceprint recognition, and in particular to a method, device and readable medium for training a voiceprint recognition model using self-supervised learning. Background Art

[0002] With the widespread application of voiceprint recognition technology in public security, financial anti-fraud, criminal investigation and other fields, its application scenarios are increasing and the requirements for related technical indicators are becoming higher and higher. At present, among the existing models, the supervised learning model has the best effect, but there is still a lot of room for improvement. In addition, there are many unlabeled or mislabeled data in real scenes. How to use this data to train and optimize the model requires exploring the advantages of self-supervised models.

[0003] Most of the existing self-supervised models use the supervised model as a pre-trained model, and then perform pseudo-label annotation through clustering, and then perform pseudo-label self-supervised model training. However, for large-scale data, the recognition and discrimination effects of the supervised model have limitations, and it is easy for the self-supervised model effect to approach the supervised model effect. In addition, clustering is performed based on the extracted voiceprint features, which is more dependent on the performance of the model. Summary of the invention

[0004] The data obtained for the above-mentioned practical application scenarios have problems such as no labels or wrong labels, and the self-supervised model relies on the performance of the supervised model. The purpose of the embodiments of this application is to propose a self-supervised learning voiceprint recognition model training method, device and readable medium to solve the technical problems mentioned in the above background technology section.

[0005] In a first aspect, an embodiment of the present application provides a method for training a voiceprint recognition model using self-supervised learning, comprising the following steps:

[0006] S1, using the first training data to perform training in at least one generation of training of the voiceprint recognition model to obtain a trained voiceprint recognition model, and performing similarity calculation on the first training data using the trained voiceprint recognition model to obtain a first similarity result, and dividing the first training data into second training data and third training data based on the first similarity result;

[0007] S2, using the second training data to perform training in at least one generation of training of the trained voiceprint recognition model to obtain an optimized voiceprint recognition model, and performing similarity calculation on the first training data using the optimized voiceprint recognition model to obtain a second similarity result, and dividing the first training data into fourth training data and fifth training data based on the second similarity result;

[0008] S3, extracting a certain amount of data from the fifth training data and combining it with the fourth training data to form sixth training data, using the sixth training data for training in at least one generation of training of the optimized voiceprint recognition model to obtain an optimized trained voiceprint recognition model, and performing similarity calculation on the first training data using the optimized trained voiceprint recognition model to obtain a third similarity result, and dividing the first training data into seventh training data and eighth training data based on the third similarity result;

[0009] S4, repeating steps S1-S3, and using the seventh training data as the first training data to train the voiceprint recognition model until the voiceprint recognition model achieves the expected effect or the training end condition is met.

[0010] In some embodiments, the training process of the voiceprint recognition model includes:

[0011] Performing VAD processing on speech data in the training data to extract valid speech segments, wherein the training data is one of the first training data, the second training data, the third training data, the fourth training data, the fifth training data, the sixth training data, the seventh training data, and the eighth training data;

[0012] Extract at least 2 or n fixed-length data based on valid voice segments, extract voice features from at least 2 fixed-length data, input the voice features into the voiceprint recognition model, calculate the three-rank loss calculation with each voice data itself as a label, and output the voiceprint recognition result;

[0013] After back propagation, the network parameters are updated.

[0014] In some embodiments, dividing the first training data into the second training data and the third training data based on the first similarity result in step 1 specifically includes: sorting according to the similarity of the first similarity result, and selecting the data with scores within a specified interval distance from the first training data as the second training data, and the rest as the third training data;

[0015] In step 2, dividing the first training data into fourth training data and fifth training data based on the second similarity result specifically includes: sorting according to the similarity of the second similarity result, and selecting data with scores within a specified interval distance from the first training data as the fourth training data, and the rest as the fifth training data;

[0016] In step 3, dividing the first training data into sixth training data and seventh training data based on the third similarity result specifically includes: sorting according to the similarity of the third similarity result, and selecting data with scores within a specified interval distance in the first training data as the sixth training data, and the rest as the seventh training data.

[0017] In some embodiments, each generation of training of the voiceprint recognition model includes at least one batch of training, each batch of training data consists of batch size voice data, each voice data has 2 or n voice features, the number of labels is batchsize×(2 or n), the voiceprint recognition model in step S1 is trained for 1-3 generations, and the voiceprint recognition model in steps S2 and S3 is trained for 2 generations.

[0018] In some embodiments, the sixth training data in step S3 includes one third of the fifth training data.

[0019] In some embodiments, the voiceprint recognition model includes an ecapa-tdnn network.

[0020] In some embodiments, the similarity comprises cosine similarity.

[0021] In a second aspect, an embodiment of the present application provides a self-supervised learning voiceprint recognition model training device, comprising:

[0022] A first training module is configured to use the first training data for training in at least one generation of training of the voiceprint recognition model to obtain a trained voiceprint recognition model, and perform similarity calculation on the first training data through the trained voiceprint recognition model to obtain a first similarity result, and divide the first training data into second training data and third training data based on the first similarity result;

[0023] a second training module configured to use the second training data for training in at least one generation of training of the trained voiceprint recognition model to obtain an optimized voiceprint recognition model, and to perform similarity calculation on the first training data through the optimized voiceprint recognition model to obtain a second similarity result, and to divide the first training data into fourth training data and fifth training data based on the second similarity result;

[0024] a third training module, configured to extract a certain amount of data from the fifth training data and combine it with the fourth training data to form sixth training data, use the sixth training data for training in at least one generation of training of the optimized voiceprint recognition model to obtain an optimized trained voiceprint recognition model, perform similarity calculation on the first training data through the optimized trained voiceprint recognition model to obtain a third similarity result, and divide the first training data into seventh training data and eighth training data based on the third similarity result;

[0025] The repeated training module is configured to repeatedly execute the first training module to the third training data, and use the seventh training data as the first training data to train the voiceprint recognition model until the voiceprint recognition model achieves the expected effect or until the training end condition is met.

[0026] In a third aspect, an embodiment of the present application provides an electronic device comprising one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors implement a method as described in any implementation manner in the first aspect.

[0027] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any implementation manner in the first aspect.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] (1) The self-supervised learning voiceprint recognition model training method of the present invention can effectively utilize real scene data and continuously optimize the model by means of an optimized self-supervised learning scheme according to the data obtained from the actual application scenario, where there is no label or the label is wrong.

[0030] (2) The self-supervised learning voiceprint recognition model training method of the present invention can avoid selecting speech data with similar voiceprint features for voiceprint recognition model training, but instead selects speech data with relatively large differences in voiceprint features for voiceprint recognition model training, which can improve the effectiveness of model training.

[0031] (3) The self-supervised learning voiceprint recognition model training method of the present invention can use speech data with similar voiceprint features as disturbances or mine out distinctive features after the voiceprint recognition model gradually becomes stable, which is helpful for the convergence and stability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 is an exemplary device architecture diagram to which an embodiment of the present application may be applied;

[0034] Figure 2 A flowchart of a voiceprint recognition model training method for self-supervised learning according to an embodiment of the present invention;

[0035] Figure 3 A network structure diagram of a voiceprint recognition model of a voiceprint recognition model training method using self-supervised learning according to an embodiment of the present invention;

[0036] Figure 4A schematic diagram of a voiceprint recognition model training device for self-supervised learning according to an embodiment of the present invention;

[0037] Figure 5 It is a schematic diagram of the structure of a computer device suitable for implementing the electronic device of the embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] Figure 1 An exemplary device architecture 100 is shown to which the method for training a voiceprint recognition model using self-supervised learning or the device for training a voiceprint recognition model using self-supervised learning according to an embodiment of the present application can be applied.

[0040] like Figure 1 As shown, the device architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0041] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various applications, such as data processing applications, file processing applications, etc., can be installed on terminal devices 101, 102, 103.

[0042] Terminal devices 101, 102, 103 can be hardware or software. When terminal devices 101, 102, 103 are hardware, they can be various electronic devices, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, etc. When terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software or software modules (for example, software or software modules used to provide distributed services), or they can be implemented as a single software or software module. No specific limitation is made here.

[0043] The server 105 may be a server that provides various services, such as a background data processing server that processes files or data uploaded by the terminal devices 101, 102, and 103. The background data processing server may process the acquired files or data and generate processing results.

[0044] It should be noted that the self-supervised learning voiceprint recognition model training method provided in the embodiment of the present application can be executed by the server 105, and can also be executed by the terminal devices 101, 102, and 103. Accordingly, the self-supervised learning voiceprint recognition model training device can be set in the server 105, and can also be set in the terminal devices 101, 102, and 103.

[0045] 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 as required. In the case where the processed data does not need to be obtained remotely, the above device architecture may not include a network, but only a server or a terminal device.

[0046] Figure 2 A self-supervised learning voiceprint recognition model training method provided in an embodiment of the present application is shown, comprising the following steps:

[0047] S1, use the first training data to train in at least one generation of training of the voiceprint recognition model to obtain a trained voiceprint recognition model, and use the trained voiceprint recognition model to perform similarity calculation on the first training data to obtain a first similarity result, and divide the first training data into second training data and third training data based on the first similarity result.

[0048] In a specific embodiment, the voiceprint recognition model is an ecapa-tdnn network. Figure 3 In other optional embodiments, the voiceprint recognition model may also adopt other network structures, such as

[0049] tdnn / etdnn / cnn-ecapa / resnet, etc. One generation of training is one epoch, that is, the model is fully trained using all the data in the training set, and all the training data undergoes one forward propagation and one backward propagation in the voiceprint recognition network model. The training data of one epoch can be divided into multiple batches, and the size of the training data in each batch is the batch size.

[0050] In a specific embodiment, the training process of the voiceprint recognition model specifically includes:

[0051] Performing VAD processing on speech data in the training data to extract valid speech segments, wherein the training data is one of the first training data, the second training data, the third training data, the fourth training data, the fifth training data, the sixth training data, the seventh training data, and the eighth training data;

[0052] Extract at least 2 or n fixed-length data based on valid voice segments, extract voice features from at least 2 fixed-length data, input the voice features into the voiceprint recognition model, calculate the three-rank loss calculation with each voice data itself as a label, and output the voiceprint recognition result;

[0053] After back propagation, the network parameters are updated.

[0054] Specifically, the Kaldi method is used to perform VAD processing on the speech data, extract valid speech segments, and extract 2 or n 200-frame data from each valid speech segment. The McFF features are extracted from the extracted 200-frame data, and the McFF features are input into the Ecapa-TDNN network. Triplet Ranking Loss is used for loss calculation, mainly to construct positive samples, negative samples and anchor points. In self-supervised learning, part of a speech is used as an anchor point, the other part as a positive sample, and the rest of the speech as a negative sample. In the early calculation, the requirements for negative samples are not strict. In the later calculation, the negative sample selected is the speech with a larger distance after sorting.

[0055] In a specific embodiment, each generation of training of the voiceprint recognition model includes at least one batch of training, each batch of training data consists of batch size voice data, each voice data has 2 or n voice features, and the number of labels is batchsize×(2 or n). The network structure operation reference experienced in the training process of the voiceprint recognition model Figure 3 , its specific operation process is not the focus of this application, so it will not be repeated. In one embodiment, the similarity adopts cosine similarity. In other embodiments, similarity measurement methods such as Euclidean distance, Pearson correlation coefficient, Tanimoto coefficient, etc. can also be used. In the embodiment of this application, cosine similarity is taken as an example. In the training process of step S1, 1-3 generations of training are adopted, all the voice data of the first training data are trained, and the cosine similarity of the first training data is calculated by the trained voiceprint recognition model to obtain the first similarity result, sort according to the similarity size in the second similarity result, and select the score of the first similarity result within the specified interval distance in the first training data as the second training data, and the rest of the first training data as the third training data. The 1-3 generations of training are for the voiceprint recognition model to learn some basic features, on this basis, the first training data is classified for the first time, and the trained voiceprint recognition model is obtained. After each epoch, it is assumed that the voiceprint recognition model is moving towards the optimal classification direction. After each epoch, it is assumed that the distance between the voiceprint features of different people will be pulled apart. However, some people's voiceprint features are similar or different audio voices of the same person are similar and difficult to distinguish.

[0056] It should be noted that the first training data, the second training data, the third training data, the fourth training data, the fifth training data, the sixth training data, the seventh training data and the eighth training data all use themselves as labels. For example, the label of the speech feature extracted from the first training data is the first training data itself.

[0057] S2, using the second training data to train in at least one generation of training of the trained voiceprint recognition model to obtain an optimized voiceprint recognition model, and performing similarity calculation on the first training data through the optimized voiceprint recognition model to obtain a second similarity result, and dividing the first training data into fourth training data and fifth training data based on the second similarity result.

[0058] In a specific embodiment, the training process of the trained voiceprint recognition model refers to the training process of the voiceprint recognition model described above, and the second training data generated by the first classification is used for the second generation of training, so that the voiceprint recognition model can more accurately distinguish individuals with large differences in voiceprint features, that is, the voiceprint recognition model can converge quickly. On this basis, the second classification is performed, and the resulting voiceprint recognition model can classify individuals with large differences in features. In the first few epochs, try to avoid speech with similar voiceprint features, and select data with large differences in voiceprint features for model training.

[0059] After some epochs, the voiceprint recognition model gradually stabilizes, and then considers people with similar voiceprint features or the voice of the same person. During the training process, the fifth training data is used as a disturbance or to discover new distinguishing features. This self-supervised training data selection strategy helps the convergence and stability of the voiceprint recognition model.

[0060] In the training process of step S2, two generations of training are adopted to train all the speech data of the second training data, and the cosine similarity calculation is performed on the first training data through the optimized voiceprint recognition model to obtain the second similarity result. The second similarity result is sorted according to the similarity size, and the score of the second similarity result within the specified interval distance is selected from the first training data as the fourth training data, and the rest of the first training data is used as the fifth training data.

[0061] S3, extracting a certain amount of data from the fifth training data and combining it with the fourth training data as the sixth training data, using the sixth training data for training in at least one generation of training of the optimized voiceprint recognition model to obtain the optimized trained voiceprint recognition model, and performing similarity calculation on the first training data through the optimized trained voiceprint recognition model to obtain a third similarity result, and dividing the first training data into seventh training data and eighth training data based on the third similarity result.

[0062] In a specific embodiment, the training process of the optimized voiceprint recognition model refers to the training process of the voiceprint recognition model described above, and the fourth training data and part of the fifth training data generated by the second classification are used for two generations of training. Specifically, the sixth training data in step S3 includes 1 / 3 of the fifth training data. The fourth training data and 1 / 3 of the fifth training data are used as the sixth training data so that the voiceprint recognition model can have the characteristics of distinguishing people with similar voiceprint features, adding some disturbances to the model training and discovering new distinguishing features.

[0063] In a specific embodiment, two generations of training are used in the training process of step S3, all speech data of the sixth training data are trained, and the cosine similarity calculation is performed on the first training data by optimizing the trained voiceprint recognition model to obtain a third similarity result. The third similarity results are sorted according to the similarity size, and the scores of the third similarity results within the specified interval distance are selected from the first training data as the seventh training data, and the remaining first training data are used as the eighth training data.

[0064] S4, repeating steps S1-S3, and using the seventh training data as the first training data to train the voiceprint recognition model until the voiceprint recognition model achieves the expected effect or the training end condition is met.

[0065] Specifically, the seventh training data generated by the third classification is used as the first training data to train the voiceprint recognition model in order to converge the voiceprint recognition model and maintain a fixed classification feature network. Repeat the above steps to train the voiceprint recognition model until the network parameters of the voiceprint recognition model are stable.

[0066] The above self-supervised learning voiceprint recognition model training method can be effectively applied to situations where data is unlabeled or mislabeled in real application scenarios, which is conducive to the continuous optimization of the model.

[0067] Further references Figure 4 As an implementation of the methods shown in the above figures, the present application provides an embodiment of a self-supervised learning voiceprint recognition model training device. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0068] The present application embodiment provides a self-supervised learning voiceprint recognition model training device, comprising:

[0069] A first training module 1 is configured to use the first training data for training in at least one generation of training of the voiceprint recognition model to obtain a first voiceprint recognition model, perform similarity calculation on the first training data through the first voiceprint recognition model to obtain a first similarity result, and divide the first training data into second training data and third training data based on the first similarity result;

[0070] The second training module 2 is configured to use the second training data for training in at least one generation of training of the first voiceprint recognition model to obtain a trained voiceprint recognition model, and perform similarity calculation on the first training data through the trained voiceprint recognition model to obtain a second similarity result, and divide the first training data into fourth training data and fifth training data based on the second similarity result;

[0071] The third training module 3 is configured to extract a certain amount of data from the fifth training data and combine it with the fourth training data to form the sixth training data, use the sixth training data to perform training in at least one generation of training of the trained voiceprint recognition model to obtain a hybrid trained voiceprint recognition model, and perform similarity calculation on the first training data through the hybrid trained voiceprint recognition model to obtain a third similarity result, and divide the first training data into seventh training data and eighth training data based on the third similarity result;

[0072] The repeated training module 4 is configured to repeatedly execute the first training module to the third training data, and use the seventh training data as the first training data to train the voiceprint recognition model until the voiceprint recognition model achieves the expected effect or until the training end condition is met.

[0073] Reference below Figure 5 , which shows an electronic device (eg Figure 1 A schematic diagram of the structure of a computer device 500 (a server or terminal device as shown). Figure 5 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0074] like Figure 5 As shown, the computer device 500 includes a central processing unit (CPU) 501 and a graphics processing unit (GPU) 502, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 503 or the program loaded from the storage part 509 to the random access memory (RAM) 504. In the RAM 504, various programs and data required for the operation of the device 500 are also stored. The CPU 501, GPU 502, ROM 503 and RAM 504 are connected to each other through a bus 505. An input / output (I / O) interface 506 is also connected to the bus 505.

[0075] The following components are connected to the I / O interface 506: an input section 507 including a keyboard, a mouse, etc.; an output section 508 including a display such as a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 509 including a hard disk, etc.; and a communication section 510 including a network interface card such as a LAN card, a modem, etc. The communication section 510 performs communication processing via a network such as the Internet. A drive 511 can also be connected to the I / O interface 506 as needed. A removable medium 512, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 511 as needed, so that a computer program read therefrom is installed into the storage section 509 as needed.

[0076] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part 510, and / or installed from a removable medium 512. When the computer program is executed by a central processing unit (CPU) 501 and a graphics processing unit (GPU) 502, the above-mentioned functions defined in the method of the present application are executed.

[0077] It should be noted that the computer-readable medium of the present application may be a computer-readable signal medium or a computer-readable medium or any combination of the above two. The computer-readable medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor device, device or component, or any combination of the above. More specific examples of computer-readable media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution device, device or component. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium may also be any computer-readable medium other than a computer-readable medium that can send, propagate or transmit a program for use by or in conjunction with an instruction execution device, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0078] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0079] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the device, method and computer program product according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based device that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0080] The modules involved in the embodiments described in this application may be implemented by software or hardware, and the modules described may also be set in a processor.

[0081] As another aspect, the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or it may exist independently and not be assembled into the electronic device. The above computer-readable medium carries one or more programs. When the above one or more programs are executed by the electronic device, the electronic device: uses the first training data for training in at least one generation of training of the voiceprint recognition model to obtain a trained voiceprint recognition model, and calculates the similarity of the first training data through the trained voiceprint recognition model to obtain a first similarity result, and divides the first training data into second training data and third training data based on the first similarity result; uses the second training data for training in at least one generation of training of the trained voiceprint recognition model to obtain an optimized voiceprint recognition model, and calculates the similarity of the first training data through the optimized voiceprint recognition model to obtain a second similarity result. , based on the second similarity result, the first training data is divided into fourth training data and fifth training data; a certain amount of data is extracted from the fifth training data and combined with the fourth training data as sixth training data, and the sixth training data is used for training in at least one generation of training of the optimized voiceprint recognition model to obtain the optimized trained voiceprint recognition model, and the first training data is calculated for similarity by the optimized trained voiceprint recognition model to obtain a third similarity result, and the first training data is divided into seventh training data and eighth training data based on the third similarity result; repeat the above steps, and use the seventh training data as the first training data to train the voiceprint recognition model until the voiceprint recognition model achieves the expected effect or until the training end condition is met.

[0082] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above invention concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

Claims

1. A self-supervised learning voiceprint recognition model training method, characterized in that: The following steps are involved: S1, using first training data to train in at least one generation of training of a voiceprint recognition model to obtain a trained voiceprint recognition model, and performing similarity calculation on the first training data through the trained voiceprint recognition model to obtain a first similarity result, dividing the first training data into second training data and third training data based on the first similarity result, sorting the first similarity result according to the similarity, and selecting data with scores within a specified interval distance from the first training data as second training data, and the rest as third training data; S2, using the second training data to train in at least one generation of training of the trained voiceprint recognition model to obtain an optimized voiceprint recognition model, and performing similarity calculation on the first training data through the optimized voiceprint recognition model to obtain a second similarity result, dividing the first training data into fourth training data and fifth training data based on the second similarity result, sorting the second similarity results according to their similarity, and selecting data with scores within a specified interval distance from the first training data as fourth training data, and the rest as fifth training data; S3, extracting a certain amount of data from the fifth training data and combining it with the fourth training data to form the sixth training data, the sixth training data including one third of the fifth training data, using the sixth training data for training in at least one generation of training of the optimized voiceprint recognition model to obtain the optimized trained voiceprint recognition model, and performing similarity calculation on the first training data by using the optimized trained voiceprint recognition model to obtain a third similarity result, dividing the first training data into seventh training data and eighth training data based on the third similarity result, sorting the data according to the similarity of the third similarity result, and selecting the data with scores within a specified interval distance from the first training data as the sixth training data, and the rest as the seventh training data; S4, repeating steps S1-S3, using the seventh training data as the first training data to train the voiceprint recognition model, until the voiceprint recognition model achieves the expected effect or until the training end condition is met.

2. The method for training a voiceprint recognition model using self-supervised learning according to claim 1, characterized in that: The training process of the voiceprint recognition model includes: Performing VAD processing on speech data in the training data to extract valid speech segments, wherein the training data is one of the first training data, the second training data, the third training data, the fourth training data, the fifth training data, the sixth training data, the seventh training data, and the eighth training data; Extracting at least two fixed-length data based on the valid voice segment, extracting voice features from the at least two fixed-length data, inputting the voice features into the voiceprint recognition model, calculating the three-rank loss calculation with each voice data itself as a label, and outputting the voiceprint recognition result; After back propagation, the network parameters are updated.

3. The self-supervised learning voiceprint recognition model training method according to claim 2 is characterized in that: Each generation of training of the voiceprint recognition model includes at least one batch of training, each batch of training data consists of batch size voice data, each voice data has n voice features, and the number of labels is batch size×n. The voiceprint recognition model in step S1 is trained for 1-3 generations, and the voiceprint recognition model in steps S2 and S3 is trained for 2 generations.

4. The method for training a voiceprint recognition model using self-supervised learning according to claim 1, characterized in that: The voiceprint recognition model includes an ecapa-tdnn network.

5. The method for training a voiceprint recognition model using self-supervised learning according to claim 1, characterized in that: The similarity includes cosine similarity.

6. A self-supervised learning voiceprint recognition model training device, characterized in that: include: A first training module is configured to use the first training data for training in at least one generation of training of the voiceprint recognition model to obtain a trained voiceprint recognition model, and perform similarity calculation on the first training data through the trained voiceprint recognition model to obtain a first similarity result, divide the first training data into second training data and third training data based on the first similarity result, sort the first similarity result according to the similarity, and select the data with scores within a specified interval distance from the first training data as the second training data, and the rest as the third training data; a second training module configured to use the second training data for training in at least one generation of training of the trained voiceprint recognition model to obtain an optimized voiceprint recognition model, and to perform similarity calculation on the first training data through the optimized voiceprint recognition model to obtain a second similarity result, and to divide the first training data into fourth training data and fifth training data based on the second similarity result, and to sort the first training data according to the similarity of the second similarity result, and to select the data with scores within a specified interval distance from the first training data as the fourth training data, and the rest as the fifth training data; a third training module, configured to extract a certain amount of data from the fifth training data and combine it with the fourth training data to form sixth training data, wherein the sixth training data includes one third of the fifth training data, and use the sixth training data to perform training in at least one generation of training of the optimized voiceprint recognition model to obtain an optimized trained voiceprint recognition model, and perform similarity calculation on the first training data by using the optimized trained voiceprint recognition model to obtain a third similarity result, and divide the first training data into seventh training data and eighth training data based on the third similarity result, and sort the data according to the similarity of the third similarity result, and select the data with scores within a specified interval distance from the first training data as the sixth training data, and the rest as the seventh training data; The repeated training module is configured to repeatedly execute the first training module to the third training module, and use the seventh training data as the first training data to train the voiceprint recognition model until the voiceprint recognition model achieves the expected effect or until the training end condition is met.

7. An electronic device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images