Voiceprint recognition model training method, voiceprint recognition method and related equipment

Abstract

The present application provides a voiceprint recognition model training method, a voiceprint recognition method, and related equipment. The voiceprint recognition model training method includes: inputting first voiceprint data into the teacher network and the student network, respectively, and performing N-th iterative training; the teacher network has converged based on second voiceprint data, and the duration of the second voiceprint data is greater than the duration of the first voiceprint data; calculating the first mean vector of the first eigenvector output by the teacher network and the second eigenvector output by the student network; if the output value corresponding to the first mean vector meets a first preset condition, determining the to-be-trained model after the N-th iterative training as the voiceprint recognition model, the voiceprint recognition model including the teacher network and the student network after the N-th iterative training. In this way, the accuracy of the voiceprint recognition model is improved.

Description

Technical Field

[0001] The present application relates to the field of voiceprint recognition technology, and in particular to a voiceprint recognition model training method, a voiceprint recognition method, and related equipment. Background Art

[0002] As a trusted voiceprint authentication technology, voiceprint recognition has broad application prospects in many fields and scenarios, including identity authentication and security verification. However, voice is easily affected by various external environmental factors, such as noise, emotions, and physical condition, as well as internal factors. Therefore, improving the accuracy of voiceprint recognition is of great practical significance. Currently, when training voiceprint recognition models, when the voiceprint data is short, the accuracy of the trained voiceprint recognition model is easily reduced. Summary of the Invention

[0003] The embodiments of the present application provide a voiceprint recognition model training method, a voiceprint recognition method, and related equipment to solve the problem of low accuracy of the trained voiceprint recognition model.

[0004] In order to solve the above technical problems, this application is implemented as follows:

[0005] In a first aspect, an embodiment of the present application provides a method for training a voiceprint recognition model, the method comprising:

[0006] Inputting the first voiceprint data into the teacher network and the student network included in the to-be-trained model, respectively, for the Nth iteration of training; wherein N is a positive integer, the teacher network has converged based on the training of the second voiceprint data, and the duration of the second voiceprint data is greater than the duration of the first voiceprint data;

[0007] Calculating a first mean vector of a first eigenvector output by the teacher network and a second eigenvector output by the student network;

[0008] When the output value corresponding to the first mean vector meets the first preset condition, the model to be trained after the Nth iterative training is determined as the voiceprint recognition model, and the voiceprint recognition model includes the teacher network and the student network after the Nth iterative training.

[0009] It can be seen that in the embodiment of the present application, the second voiceprint data is used to train the teacher network to converge, that is, the teacher network has a higher recognition accuracy for the second voiceprint data. When the teacher network converges, the first voiceprint data is then input into the teacher network and the student network at the same time, so that the student network learns the teacher network's method of recognizing the first voiceprint data. That is, in the student network's training process of recognizing the first voiceprint data, since the teacher network has converged, the teacher network can teach the student model to recognize the first voiceprint data in the same way as the teacher network recognizes the second voiceprint data, thereby improving the student model's recognition accuracy for the first voiceprint data, thereby improving the student network's training efficiency and recognition accuracy, and further improving the student network's recognition accuracy for the first voiceprint data.

[0010] In addition, while using the first voiceprint data to train the student network, the first voiceprint data can also be used to train the teacher network to further train the teacher network's ability to recognize the first voiceprint data, thereby improving the teacher network's accuracy in recognizing the first voiceprint data.

[0011] In a second aspect, an embodiment of the present application provides a voiceprint recognition method, which is applied to a student network included in a voiceprint recognition model, wherein the student network is connected to a target classifier, and the method includes:

[0012] Obtaining first voiceprint data of the user to be identified;

[0013] Inputting the first voiceprint data into the student network, and outputting a first feature vector corresponding to the first voiceprint data;

[0014] Input the first feature vector and the pre-stored second feature vector into the target classifier, and output a likelihood distribution value; wherein the second feature vector is output by inputting the second voiceprint data of the target user into the student network;

[0015] When the likelihood distribution value is greater than a preset threshold, it is determined that the user to be identified and the target user are the same user.

[0016] It can be seen that in the embodiment of the present application, since the teacher network has converged during the training process of the student network in recognizing the first voiceprint data, the teacher network can teach the student model to recognize the first voiceprint data in the same way as the teacher network recognizes the second voiceprint data, thereby improving the student model's accuracy in recognizing the first voiceprint data, so that the voiceprint recognition model finally trained has better recognition effect on voiceprint data with shorter duration. Therefore, in the process of recognizing voiceprint data with shorter duration (i.e., first voiceprint data), the student network in the voiceprint recognition model and the target classifier connected to the student network can be used to determine whether the user to be identified and the target user are the same user, thereby improving the accuracy of the recognition result of the voiceprint data of the user to be identified and reducing the loss caused by the inability to accurately identify the voiceprint data of the user to be identified.

[0017] In a third aspect, an embodiment of the present application further provides a voiceprint recognition model training device, comprising:

[0018] A first training module is configured to input the first voiceprint data into a teacher network and a student network included in a to-be-trained model, respectively, for Nth iteration of training; wherein N is a positive integer, the teacher network has converged based on the training of the second voiceprint data, and the duration of the second voiceprint data is greater than the duration of the first voiceprint data;

[0019] A calculation module, configured to calculate a first mean vector of a first eigenvector output by the teacher network and a second eigenvector output by the student network;

[0020] The first determination module is used to determine the model to be trained after the Nth iterative training as the voiceprint recognition model when the output value corresponding to the first mean vector meets the first preset condition, and the voiceprint recognition model includes the teacher network and the student network after the Nth iterative training.

[0021] In a fourth aspect, an embodiment of the present application further provides a voiceprint recognition device, which is applied to a student network included in a voiceprint recognition model, wherein the student network is connected to a target classifier, and the voiceprint recognition device includes:

[0022] An acquisition module, configured to acquire first voiceprint data of a user to be identified;

[0023] a first output module, configured to input the first voiceprint data into the student network and output a first feature vector corresponding to the first voiceprint data;

[0024] A second output module is configured to input the first feature vector and a pre-stored second feature vector into the target classifier and output a likelihood distribution value; wherein the second feature vector is output by inputting the second voiceprint data of the target user into the student network;

[0025] The second determining module is configured to determine that the user to be identified and the target user are the same user if the likelihood distribution value is greater than a preset threshold.

[0026] In a fifth aspect, an embodiment of the present application further provides an electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps in the above-mentioned voiceprint recognition model training method or voiceprint recognition method when executing the computer program.

[0027] In a sixth aspect, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned voiceprint recognition model training method or voiceprint recognition method are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0029] Figure 1 This is a flow chart of a voiceprint recognition model training method provided in an embodiment of the present application;

[0030] Figure 2 This is a flowchart of another voiceprint recognition model training method provided by an embodiment of the present application;

[0031] Figure 3 This is a schematic diagram of the structure of a model to be trained provided in an embodiment of the present application;

[0032] Figure 4 This is a flow chart of a voiceprint recognition method provided by an embodiment of the present application;

[0033] Figure 5 This is a structural diagram of a voiceprint recognition model training device provided in an embodiment of the present application;

[0034] Figure 6 This is a schematic diagram of the structure of a voiceprint recognition device provided in an embodiment of the present application;

[0035] Figure 7 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0037] Since in many application scenarios, the duration of voiceprint data is relatively short, but accurate voiceprint recognition is required, there is a lack of a voiceprint recognition model with high recognition accuracy for voiceprint data with shorter duration. The voiceprint recognition models trained using voiceprint data with shorter duration in the existing technology have low accuracy.

[0038] In order to solve the above problems, the embodiment of the present application considers pre-training the teacher network with long-time voiceprint data (i.e., the second voiceprint data), so that the teacher network can maintain the model effect of the long-time voiceprint data, and then input short-time data (i.e., the first voiceprint data) into the teacher network and perform fine-tuning during the training process of the short-time data. At the same time, short-time data is also input into the student network for training. That is, in the training process of the student network recognizing the first voiceprint data, the teacher network has converged, so the teacher network can teach the student model to recognize the first voiceprint data in the same way as the teacher network recognizes the second voiceprint data, thereby improving the accuracy of the student model in recognizing the first voiceprint data and the convergence rate, that is, improving the training rate of the student model.

[0039] In addition, the first eigenvector output by the teacher model and the second eigenvector output by the student network are averaged to obtain a first mean vector. Based on the first mean vector, it can be determined whether a first preset condition is met. When the first preset condition is met, a voiceprint recognition model can be determined to be obtained. When the first mean vector does not meet the first preset condition, the training parameters of the student network can be adjusted based on the first mean vector, ultimately achieving student network convergence to obtain a voiceprint recognition model. In this way, the student model ultimately trained combines the parameters of the model trained on long-term data with fine-tuning of short-term data, resulting in improved performance on short-term data without excessively compromising the accuracy of the model on long-term data. This improves the recognition accuracy of the voiceprint recognition model for voiceprint data. For details, please refer to the following specific embodiments.

[0040] See also Figure 1 , Figure 1 This is a flow chart of the voiceprint recognition model training method provided by the embodiment of the present application. Figure 1 As shown, the following steps are included:

[0041] Step 101: Input the first voiceprint data into the teacher network and student network included in the model to be trained at the same time, and perform the Nth iterative training; wherein, N is a positive integer, the teacher network has converged based on the training of the second voiceprint data, and the duration of the second voiceprint data is greater than the duration of the first voiceprint data.

[0042] Among them, the time when the first voiceprint data is input into the teacher network and the student network can be the same or different. For example, the first voiceprint data can be input into the teacher network and the student network respectively at the same time. Of course, the first voiceprint data can also be input into the teacher network at the first moment and into the student network at the second moment. The first moment and the second moment can be different moments, and the time difference between the first moment and the second moment is small, that is, the first moment and the second moment can be determined to be closer moments.

[0043] The content of each copy of the first voiceprint data can be identical, and the number of copies of the first voiceprint data input into the teacher network and the student network can be the same or different. For example, one copy of the first voiceprint data can be input into the teacher network and one copy of the first voiceprint data can be input into the student network. Alternatively, one copy of the first voiceprint data can be input into the teacher network and three copies of the first voiceprint data can be input into the student network.

[0044] Among them, the first voiceprint data and the second voiceprint data can also be called speech voiceprint feature vectors or voiceprint feature vectors, and the method of obtaining the first voiceprint data and the second voiceprint data is not limited here. For example, the first voiceprint data and the second voiceprint data can be directly obtained from the sample pool, that is, the first voiceprint data and the second voiceprint data are pre-existing voiceprint data in the sample pool; in addition, the first voiceprint data and the second voiceprint data can also be extracted from the audio.

[0045] In addition, data amplification can be performed on the sample data in the sample pool, and the data amplification includes at least one of adding noise, speeding up the speech speed, increasing data disturbance, etc.

[0046] In addition, the sample data after the above data amplification can be divided into a training set and a test set, and the above first voiceprint data and second voiceprint data can be data in the training set, wherein the ratio of the training set to the test set is not limited here, for example: it can be 8:2.

[0047] It should be noted that the data in the training set can be divided into two data sets, one data set is used to store the first voiceprint data, that is, the voiceprint data with shorter length, and the other data set can be used to store the second voiceprint data, that is, the voiceprint data with longer length.

[0048] The specific structures of the teacher network and student network can be found in Figure 3The difference between the teacher network and the student network is that the number of network layers they include is different. For example, the teacher network may include 34 network layers, while the student network may include 18 network layers. The above network layers may refer to resnet-fpn network layers.

[0049] See also Figure 3 , Figure 3 It can be used to represent the specific structure of the teacher network and the student network, where the input layer is used to input vector features, and the output layer is used to output vector features. Conv&bn is a fusion layer that can be used to reduce the dimension of vector features (for example, reducing a two-dimensional vector to a one-dimensional vector), and ResNetBlock is used to represent different convolutional layers. Among them, c3, c4, c5 and p6 included in ResNetBlock can be used to represent the features extracted by the corresponding ResNetBlock convolutional layer, and ResNetBlock makes the number of network layers deeper and more trainable, and also greatly reduces the network parameters, thereby improving network performance while effectively improving network efficiency. Fc-TDNN is used to represent the time-delay neural network. Connecting Fc-TDNN after ResNetBlock can enable the model to better capture audio timing information. P7, P6, P5, P4, and P3 are used to represent the extracted features, where the output layer can output the above-mentioned P7, P6, P5, P4, and P3 features at the same time, while Temp2 is used to combine P5 to obtain P4, and Temp3 is used to combine P4 to obtain P3.

[0050] It should be noted that if Figure 3 As shown in Figure 2, the difference between the teacher network and the student network is the number of layers in ResNetBlock. In addition, Figure 3 The structure shown can be called a feature pyramid (FPN) structure, because the underlying features (i.e., the basic features such as Figure 3 In the example above, P4 in the image (e.g., P3) has strong appearance information, while high-level features (e.g., P3, which is derived from P4, is a low-level feature relative to P3, and P3 is a high-level feature relative to P4) have strong semantic information. Stacking multiple types of information to form a feature pyramid can effectively improve the ability to represent various types of information. Applying this to voiceprint recognition can effectively enhance the model's ability to map voiceprint information at multiple scales.

[0051] As an optional implementation, both the teacher network and the student network are ResNet-FPN network models, and the number of network layers of the student network is smaller than that of the teacher network. This can achieve better training effects for the teacher network and the student network. At the same time, the number of network layers of the student network is smaller than that of the teacher network, which reduces the number of network layers of the student network and the number of parameters of the student network, facilitating the convergence of student network training. At the same time, since the first voiceprint data can also be referred to as short-term voiceprint data, the impact of short-term voiceprint data on student network training can be reduced.

[0052] Among them, if the duration of the second voiceprint data is greater than that of the first voiceprint data, the second voiceprint data can be called long-term voiceprint data or long-term voice data, and the first voiceprint data can be called short-term voiceprint data or short-term voice data.

[0053] As an optional embodiment, the duration of the first voiceprint data is less than or equal to a first duration, and the duration of the second voiceprint data is greater than the first duration. The first duration can be, for example, 3 seconds, 4 seconds, or another value. This facilitates distinguishing the first and second voiceprint data, thereby improving the voiceprint data screening rate.

[0054] In addition, the content of the first voiceprint data and the second voiceprint data may be related. For example, the first voiceprint data and the second voiceprint data may both be voiceprint data of the same speaker, or the first voiceprint data and the second voiceprint data may be voiceprint data of the same scene. The difference between the first voiceprint data and the second voiceprint data may only be the difference in duration.

[0055] It should be noted that the process of training the teacher network using the second voiceprint data can be described as follows. That is, the implementation method of converging the teacher network according to the second voiceprint data training may include the following methods:

[0056] Inputting the second voiceprint data into the teacher network and performing an M-th iterative training, where M is a positive integer less than N;

[0057] When the third eigenvector output by the teacher network satisfies the second preset condition, it is determined that the teacher network converges after the M-th iterative training.

[0058] Among them, the second voiceprint data can be subjected to feature extraction to obtain a third feature vector, and the process of extracting the second voiceprint data to obtain the third feature vector can be called x-vector feature extraction, that is: the third feature vector can also be called x-vector feature.

[0059] In this embodiment, the second voiceprint data is used to train the teacher network for convergence. In this way, the teacher network can have a higher accuracy in classification prediction of long-term voiceprint data (long-term voiceprint data is similar to the second voiceprint data and has a longer duration), that is, the accuracy in identifying long-term voiceprint data is higher.

[0060] The features included in the second voiceprint data are not limited here. As an optional embodiment, the second voiceprint data includes at least one of speech features, first-order difference features, and timbre information features. This allows the second voiceprint data to include a wider variety of features, thereby enhancing the teacher network's ability to identify features during training.

[0061] It should be noted that when the second voiceprint data includes speech features, first-order difference features and timbre information features, the speech features, first-order difference features and timbre information features can be combined to form a three-channel feature map, and the three-channel feature map can be stored in a database or sample pool, making it more convenient to save the above-mentioned speech features, first-order difference features and timbre information features.

[0062] In addition, during the training process of the teacher network, the teacher network can first obtain the above three-channel feature map, and then extract the third eigenvector based on the above three-channel feature map, so that the extraction of the third eigenvector is more convenient.

[0063] The specific content of the second preset condition is not limited here. As an optional implementation, the second preset condition includes: the difference between the output value corresponding to the eigenvector and the actual value input to the teacher network during the current training is less than the second preset value. In other words, when the difference between the output value of the third eigenvector and the actual value of the second voiceprint data is less than the second preset value, the third eigenvector meets the second preset condition.

[0064] Among them, the specific expression of the second preset condition can refer to the expression of the first preset condition in the following text, the output value of the third eigenvector in this embodiment can refer to the relevant expression of the output value of the first mean vector in the following text, and the actual value of the second voiceprint data can refer to the relevant expression of the actual value of the first voiceprint data.

[0065] Among them, the output value of the third eigenvector can be converted in the classifier, that is, the third eigenvector can be input into the classifier, and the classifier outputs the above output value. The classifier can be embedded in the teacher network or connected to the teacher network.

[0066] In this embodiment, when the difference between the output value corresponding to the third eigenvector and the actual value of the second voiceprint data is less than the second preset value, it can be considered that the classification prediction probability of the teacher network for the second voiceprint data is close to the actual classification value, and the teacher network is available, that is, it can be judged that the teacher network converges after the Mth iterative training, and then the trained teacher network can have a higher accuracy in classification prediction of long-term voiceprint data (long-term voiceprint data is similar to the second voiceprint data and has a longer duration), that is, the accuracy in the recognition of long-term voiceprint data is higher.

[0067] As another optional embodiment, the second preset condition includes: the output value corresponding to the eigenvector is less than a preset value. In other words, when the output value corresponding to the third eigenvector is less than the preset value, it can be determined that the third eigenvector meets the second preset condition. In other words, it can be considered that the output value of the third eigenvector meets the second preset condition, i.e., the teacher network has converged.

[0068] Step 102: Calculate a first mean vector of a first eigenvector output by the teacher network and a first mean vector of a second eigenvector output by the student network.

[0069] The number of first voiceprint data input into the student network is not limited here. For example, the number of first voiceprint data input into the student network can be 1, 2 or more.

[0070] As an optional implementation manner, the number of first voiceprint data input into the student network is L, where L is a positive integer greater than 1;

[0071] The calculating of a first mean vector of a first eigenvector output by the teacher network and a second eigenvector output by the student network comprises:

[0072] Obtain L second eigenvectors output by the student network, wherein the second eigenvectors correspond one-to-one to the first voiceprint data input into the student network;

[0073] Calculate the second mean vector of the L second eigenvectors;

[0074] Obtain a first eigenvector output by the teacher network, and calculate a mean vector of the first eigenvector and the second mean vector to obtain the first mean vector, wherein the first eigenvector corresponds to the first voiceprint data input into the teacher network.

[0075] In this embodiment, multiple copies of the first voiceprint data can be input into the student network at the same time. In this way, inputting multiple copies of the first voiceprint data at the same time can improve the training efficiency of the student network. At the same time, inputting multiple copies of the first voiceprint data into the student network at the same time allows the student network to output multiple second eigenvectors, and calculate the second mean vector based on the multiple second eigenvectors, and then calculate the mean vector of the second mean vector and the first eigenvector to obtain the first mean vector. In this way, compared with the method of inputting multiple copies of the first voiceprint data in sequence and allowing the student network to undergo multiple trainings, this embodiment reduces the number of training times while ensuring that the training intensity of the student network is better, thereby improving the accuracy of the first mean vector.

[0076] For example: See Figure 2 , Figure 2 It can be used to represent the various processes of the embodiments of the present application, such as Figure 2 As shown, the first voiceprint data is input into the teacher network and the first eigenvector is output, while three copies of the first voiceprint data are input into the student network at the same time, namely the first voiceprint data 1, the first voiceprint data 2 and the first voiceprint data 3. At the same time, after passing through the student network, the second eigenvector 1, the second eigenvector 2 and the second eigenvector 3 can be output, and the first voiceprint data 1 corresponds to the second eigenvector 1, the first voiceprint data 2 corresponds to the second eigenvector 2, and the first voiceprint data 3 corresponds to the second eigenvector 3. Then the second mean vector of the second eigenvector 1, the second eigenvector 2 and the second eigenvector 3 can be calculated, and then the second mean vector and the first mean vector of the first eigenvector are calculated, and the first mean vector is input to the output layer (the output layer can be used to connect with the classifier, for details, please refer to the above-mentioned relevant statements about the classifier).

[0077] In addition, if Figure 2 As shown, while the teacher network outputs the first eigenvector, a soft label can also be attached to the first eigenvector to distinguish and mark the first eigenvector.

[0078] Step 103: When the output value corresponding to the first mean vector satisfies a first preset condition, the model to be trained after the Nth iterative training is determined as a voiceprint recognition model, and the voiceprint recognition model includes a teacher network and a student network after the Nth iterative training.

[0079] Among them, the specific content of the first preset condition is not limited here. As an optional implementation method, the first preset condition includes: the difference between the output value corresponding to the mean vector and the actual value of the voiceprint data input into the model to be trained during this training is less than the first preset value, that is: when the difference between the output value corresponding to the first mean vector and the actual value of the first voiceprint data is less than the first preset value, it is determined that the output value corresponding to the first mean vector meets the first preset condition.

[0080] Among them, the output value corresponding to the first mean vector can be a specific numerical value, and the output value has a one-to-one correspondence with the first mean vector. The above output value can be converted by the classifier. The specific steps can be: inputting the first mean vector into the classifier, and outputting the output value corresponding to the first mean vector. The above output value can be understood as: the classification prediction probability of the model to be trained and the classifier for the first mean vector.

[0081] It should be noted that the above-mentioned classifier can be embedded in the voiceprint recognition model (for example, embedded in the student network), that is, the classifier is part of the voiceprint recognition model. Of course, the above-mentioned classifier can also be connected to the voiceprint recognition model (such as the student network), that is, it can serve as the next-level network layer structure connected to the voiceprint recognition model.

[0082] It should be noted that when the classifier is embedded in the voiceprint recognition model, a classifier can be embedded in both the teacher network and the student network. When the classifier is embedded in the voiceprint recognition model, the classifier can be connected to the teacher network and the student network respectively, or the teacher network and the student network can be connected to a classifier respectively.

[0083] According to actual practice, the first voiceprint data can also correspond to an actual classification value, and the actual classification value can also be called an actual value. Therefore, the actual value of the first mean vector can be obtained according to the actual value of the first voiceprint data (for example, it can be the actual value of the first voiceprint data), and the actual value of the first voiceprint data can exist in the identification information, and the identification information can be the identification information of the first voiceprint data.

[0084] In this embodiment, when the difference between the output value corresponding to the first mean vector and the actual value of the first voiceprint data is less than the first preset value, it can be considered that the classification prediction probability of the trained model for the first voiceprint data is close to the actual classification value, and the trained model is usable, that is, it can be judged that the trained model has converged, so that the trained model after the Nth iterative training can be determined as the voiceprint recognition model, and the accuracy of the trained voiceprint recognition model can be made higher.

[0085] In addition, the above-mentioned voiceprint recognition model can be used in subsequent use to classify the target voiceprint data and determine whether the speaker of the target voiceprint data is the same user as the speaker pre-stored in the database.

[0086] As another optional implementation, the first preset condition includes: the output value corresponding to the mean vector is less than the preset value, that is, when the output value corresponding to the first mean vector is less than the preset value, it is determined that the output value corresponding to the first mean vector meets the first preset condition, that is, the student network has converged.

[0087] In this embodiment, through steps 101 to 104, the second voiceprint data is used to train the teacher network to converge, so that the teacher network has a higher recognition accuracy for the second voiceprint data. At the same time, when the teacher network converges, the first voiceprint data is input into the teacher network and the student network at the same time, so that the student network learns the teacher network's way of recognizing the first voiceprint data, thereby improving the training efficiency and recognition accuracy of the student network, and further improving the student network's recognition accuracy and fitting ability for the first voiceprint data.

[0088] In addition, while using the first voiceprint data to train the student network, the first voiceprint data can also be used to train the teacher network to further train the teacher network's ability to recognize the first voiceprint data, thereby improving the teacher network's accuracy in recognizing the first voiceprint data.

[0089] That is to say, the second voiceprint data is used to pre-train the teacher network so that the teacher network can maintain the model effect of long-term data. When the teacher network has converged, the first voiceprint data is then used to simultaneously train the teacher network and the student network. The student network can learn the parameters of the teacher network, that is, the teacher network can guide the student network for training, thereby improving the training efficiency of the student network. At the same time, the teacher network can also be fine-tuned under short-term data (that is, the first voiceprint data).

[0090] In addition, since the first eigenvector output by the teacher network and the first mean vector of the second eigenvector output by the student network are calculated, and whether the model to be trained has converged is judged based on the first mean vector, the voiceprint recognition model finally trained can integrate the parameters of the training model on long-term data (i.e., the second voiceprint data) and the fine-tuning of short-term data, so that the model performs better on short-term data without excessively damaging the accurate effect on long-term data.

[0091] It should be noted that when the teacher network is trained using the first voiceprint data, a small learning rate can be used. At the same time, when the student network is trained using the first voiceprint data, a large learning rate can be used. Since the teacher network has been pre-trained with a large amount of second voiceprint data and its parameters have tended to converge, when training using the first voiceprint data, the amount of short-term data (i.e., the first voiceprint data) is relatively small. If a large learning rate is given to the teacher network, the teacher network will easily overfit the short-term data during training, resulting in poor learning results for the teacher network and, in turn, even worse learning results for the student network. Using a small learning rate for the teacher network results in smaller changes in the teacher network's network parameters, while using a large learning rate for the student network allows it to learn the short-term data distribution and learn the pre-trained parameters of the teacher network, thereby improving the training efficiency of the student network.

[0092] See also Figure 4 , Figure 4 This is a flow chart of the voiceprint recognition method provided by an embodiment of the present application. The method is applied to the student network included in the voiceprint recognition model, and the student network is connected to the target classifier, such as Figure 4 As shown, the following steps are included:

[0093] Step 401: Acquire the first voiceprint data of the user to be identified.

[0094] Step 402: Input the first voiceprint data into the student network, and output a first feature vector corresponding to the first voiceprint data.

[0095] Step 403: Input the first feature vector and the pre-stored second feature vector into the target classifier, and output a likelihood distribution value; wherein the second feature vector is output by inputting the second voiceprint data of the target user into the student network.

[0096] Step 404: When the likelihood distribution value is greater than a preset threshold, determine that the user to be identified and the target user are the same user.

[0097] It should be noted that the first eigenvector and the second eigenvector can both be x-vector features, and the x-vector features are neural network features extracted by the deep neural network.

[0098] The second feature vector may be understood as a feature vector corresponding to the second voiceprint data of the target user collected in advance, and may be stored on a server corresponding to the database.

[0099] Among them, the type of target classifier is not limited here. For example, the target classifier can be a PLDA classifier, that is, the PLDA algorithm can be run in the PLDA classifier. In this way, the accuracy of the likelihood distribution value output by the PLDA classifier is higher, thereby making the accuracy of the judgment result of whether the user to be identified and the target user are the same user higher.

[0100] The likelihood distribution value can also be understood as similarity, that is, the larger the likelihood distribution value is, the higher the possibility that the user to be identified and the target user are the same user.

[0101] In an embodiment of the present application, the student network in the voiceprint recognition model and the target classifier connected to the student network can be used to determine whether the user to be identified and the target user are the same user, thereby improving the accuracy of the recognition results of the voiceprint data of the user to be identified and reducing the loss caused by the inability to accurately identify the voiceprint data of the user to be identified.

[0102] See also Figure 5 , Figure 5 This is a structural diagram of the voiceprint recognition model training device provided in the embodiment of the present application, which can implement the details of the voiceprint recognition model training method in the above embodiment and achieve the same effect. The voiceprint recognition model training device is applied to the model to be trained, and the model to be trained includes a teacher network and a student network, such as Figure 5 As shown, the voiceprint recognition model training device 500 includes:

[0103] The first training module 501 is configured to input the first voiceprint data into the training model, including the teacher network and the student network, for Nth iteration training; wherein N is a positive integer, the teacher network has converged based on the second voiceprint data, and the duration of the second voiceprint data is greater than the duration of the first voiceprint data;

[0104] A calculation module 502 is used to calculate a first mean vector of a first eigenvector output by the teacher network and a second eigenvector output by the student network;

[0105] The first determination module 503 is used to determine the model to be trained after the Nth iterative training as the voiceprint recognition model when the output value corresponding to the first mean vector meets the first preset condition. The voiceprint recognition model includes the teacher network and the student network after the Nth iterative training.

[0106] Optionally, the first preset condition includes: a difference between an output value corresponding to the mean vector and an actual value of the voiceprint data input into the model to be trained during this training is less than a first preset value.

[0107] Optionally, the teacher network is trained and converged according to the second voiceprint data in the following ways:

[0108] A second training module, configured to input the second voiceprint data into the teacher network and perform an M-th iterative training, where M is a positive integer less than N;

[0109] The third determining module is used to determine whether the teacher network converges after the M-th iterative training when the third eigenvector output by the teacher network meets the second preset condition.

[0110] Optionally, the second preset condition includes: a difference between an output value corresponding to the feature vector and an actual value input into the teacher network during this training is less than a second preset value.

[0111] Optionally, the second voiceprint data includes at least one of speech features, first-order difference features and timbre information features.

[0112] Optionally, the number of first voiceprint data input into the student network is L, where L is a positive integer greater than 1;

[0113] The calculation module 502 includes:

[0114] an acquisition submodule, configured to acquire L second feature vectors output by the student network, wherein the second feature vectors correspond one-to-one to the first voiceprint data input into the student network;

[0115] A first calculation submodule, configured to calculate a second mean vector of L second eigenvectors;

[0116] The second calculation submodule is used to obtain the first eigenvector output by the teacher network and calculate the mean vector of the first eigenvector and the second mean vector to obtain the first mean vector, wherein the first eigenvector corresponds to the first voiceprint data input into the teacher network.

[0117] Optionally, both the teacher network and the student network are ResNet-FPN network models, and the number of network layers of the student network is smaller than the number of network layers of the teacher network.

[0118] The voiceprint recognition model training device provided in the embodiment of the present application can achieve Figure 1 To avoid repetition, the various processes implemented by the voiceprint recognition model training device in the method embodiment will not be described here.

[0119] In this way, the second voiceprint data is used to train the teacher network to converge, so that the teacher network has a higher recognition accuracy for the second voiceprint data. At the same time, when the teacher network converges, the first voiceprint data is input into the teacher network and the student network at the same time, so that the student network learns the teacher network's way of recognizing the first voiceprint data, thereby improving the training efficiency and recognition accuracy of the student network, and further improving the student network's recognition accuracy of the first voiceprint data.

[0120] In addition, while using the first voiceprint data to train the student network, the first voiceprint data can also be used to train the teacher network to further train the teacher network's ability to recognize the first voiceprint data, thereby improving the teacher network's accuracy in recognizing the first voiceprint data.

[0121] See also Figure 6 , Figure 6 This is a schematic diagram of the structure of a voiceprint recognition device provided in an embodiment of the present application. The voiceprint recognition device is applied to a student network included in a voiceprint recognition model, and the student network is connected to a target classifier, such as Figure 6 As shown, the voiceprint recognition device 600 includes:

[0122] An acquisition module 601 is used to acquire first voiceprint data of a user to be identified;

[0123] A first output module 602 is configured to input the first voiceprint data into the student network and output a first feature vector corresponding to the first voiceprint data;

[0124] The second output module 603 is configured to input the first feature vector and a pre-stored second feature vector into the target classifier and output a likelihood distribution value; wherein the second feature vector is output by inputting the second voiceprint data of the target user into the student network;

[0125] The second determining module 604 is configured to determine that the user to be identified and the target user are the same user if the likelihood distribution value is greater than a preset threshold.

[0126] The voiceprint recognition device provided in the embodiment of the present application can achieve Figure 4 To avoid repetition, the various processes implemented by the voiceprint recognition device in the method embodiment will not be described here.

[0127] In this way, the student network in the voiceprint recognition model and the target classifier connected to the student network can be used to determine whether the user to be identified and the target user are the same user, thereby improving the accuracy of the recognition results of the voiceprint data of the user to be identified and reducing the loss caused by the inability to accurately identify the voiceprint data of the user to be identified.

[0128] Figure 7 A schematic diagram of the hardware structure of an electronic device for implementing various embodiments of the present application.

[0129] The electronic device 700 includes but is not limited to: a radio frequency unit 701, a network module 702, an audio output unit 703, an input unit 704, a sensor 705, a display unit 706, a user input unit 707, an interface unit 708, a memory 709, a processor 710, and a power supply 711. It will be understood by those skilled in the art that Figure 7 The electronic device structure shown in the figure does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or arrange the components differently. In the embodiments of this application, the electronic device includes but is not limited to mobile phones, tablet computers, laptop computers, PDAs, vehicle-mounted terminals, wearable devices, and pedometers.

[0130] When the electronic device is used to execute the steps in the voiceprint recognition model training method, the processor 710 is used to perform the following operations:

[0131] Inputting the first voiceprint data into both the teacher network and the student network of the to-be-trained model for the Nth iteration of training; wherein N is a positive integer, the teacher network has converged based on the training of the second voiceprint data, and the duration of the second voiceprint data is greater than the duration of the first voiceprint data;

[0132] Calculating a first mean vector of a first eigenvector output by the teacher network and a second eigenvector output by the student network;

[0133] When the output value corresponding to the first mean vector meets the first preset condition, the model to be trained after the Nth iterative training is determined as the voiceprint recognition model, and the voiceprint recognition model includes the teacher network and the student network after the Nth iterative training.

[0134] Optionally, the first preset condition includes: a difference between an output value corresponding to the mean vector and an actual value of the voiceprint data input into the model to be trained during this training is less than a first preset value.

[0135] Optionally, the teacher network is trained and converged according to the second voiceprint data in the following ways:

[0136] Inputting the second voiceprint data into the teacher network and performing an M-th iterative training, where M is a positive integer less than N;

[0137] When the third eigenvector output by the teacher network satisfies the second preset condition, it is determined that the teacher network converges after the M-th iterative training.

[0138] Optionally, the second preset condition includes: a difference between an output value corresponding to the feature vector and an actual value input into the teacher network during this training is less than a second preset value.

[0139] Optionally, the second voiceprint data includes at least one of speech features, first-order difference features and timbre information features.

[0140] Optionally, the number of first voiceprint data input into the student network is L, where L is a positive integer greater than 1;

[0141] The calculation of the first mean vector of the first eigenvector output by the teacher network and the second eigenvector output by the student network performed by the processor 710 includes:

[0142] Obtain L second eigenvectors output by the student network, wherein the second eigenvectors correspond one-to-one to the first voiceprint data input into the student network;

[0143] Calculate the second mean vector of the L second eigenvectors;

[0144] Obtain a first eigenvector output by the teacher network, and calculate a mean vector of the first eigenvector and the second mean vector to obtain the first mean vector, wherein the first eigenvector corresponds to the first voiceprint data input into the teacher network.

[0145] Optionally, both the teacher network and the student network are ResNet-FPN network models, and the number of network layers of the student network is smaller than the number of network layers of the teacher network.

[0146] When the electronic device is used to perform the steps in the voiceprint recognition method, the processor 710 is used to perform the following operations:

[0147] Obtaining first voiceprint data of the user to be identified;

[0148] Inputting the first voiceprint data into the student network, and outputting a first feature vector corresponding to the first voiceprint data;

[0149] Input the first feature vector and the pre-stored second feature vector into the target classifier, and output a likelihood distribution value; wherein the second feature vector is output by inputting the second voiceprint data of the target user into the student network;

[0150] When the likelihood distribution value is greater than a preset threshold, it is determined that the user to be identified and the target user are the same user.

[0151] It should be understood that in the embodiments of the present application, the RF unit 701 may be used to receive and transmit signals during information transmission or calls. Specifically, it receives downlink data from the base station and transmits it to the processor 710 for processing; in addition, it transmits uplink data to the base station. Typically, the RF unit 701 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier, a duplexer, and the like. Furthermore, the RF unit 701 may communicate with the network and other devices via a wireless communication system.

[0152] The electronic device provides users with wireless broadband Internet access through the network module 702, such as helping users to send and receive emails, browse web pages, and access streaming media.

[0153] The audio output unit 703 can convert audio data received by the RF unit 701 or the network module 702 or stored in the memory 709 into an audio signal and output it as sound. In addition, the audio output unit 703 can also provide audio output related to a specific function performed by the electronic device 700 (for example, a call signal reception sound, a message reception sound, etc.). The audio output unit 703 includes a speaker, a buzzer, a receiver, etc.

[0154] The input unit 704 is used to receive audio or video signals. The input unit 704 may include a graphics processing unit (GPU) 7041 and a microphone 7042. The graphics processor 7041 processes image data of a still picture or video obtained by an image capture device (such as a camera) in a video capture mode or an image capture mode. The processed image frames can be displayed on the display unit 706. The image frames processed by the graphics processor 7041 can be stored in the memory 709 (or other storage medium) or transmitted via the radio frequency unit 701 or the network module 702. The microphone 7042 can receive sound and process such sound into audio data. The processed audio data can be converted into a format that can be sent to a mobile communication base station via the radio frequency unit 701 in the case of a telephone call mode.

[0155] The electronic device 700 also includes at least one sensor 705, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor includes an ambient light sensor and a proximity sensor, wherein the ambient light sensor can adjust the brightness of the display panel 7061 according to the brightness of the ambient light, and the proximity sensor can turn off the display panel 7061 and / or the backlight when the electronic device 700 is moved to the ear. As a type of motion sensor, the accelerometer sensor can detect the magnitude of acceleration in all directions (generally three axes), and can detect the magnitude and direction of gravity when stationary. It can be used to identify the posture of the electronic device (such as horizontal and vertical screen switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc.; the sensor 705 can also include a fingerprint sensor, a pressure sensor, an iris sensor, a molecular sensor, a gyroscope, a barometer, a hygrometer, a thermometer, an infrared sensor, etc., which will not be repeated here.

[0156] The display unit 706 is used to display information input by the user or information provided to the user. The display unit 706 may include a display panel 7061, which may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), or the like.

[0157] The user input unit 707 can be used to receive input digital or character information, and to generate key signal input related to the user settings and function control of the electronic device. Specifically, the user input unit 707 includes a touch panel 7071 and other input devices 7072. The touch panel 7071, also known as a touch screen, can collect user touch operations on or near it (such as operations performed by the user using any suitable object or accessory such as a finger, stylus, etc. on or near the touch panel 7071). The touch panel 7071 may include two parts: a touch detection device and a touch controller. Among them, the touch detection device detects the user's touch direction and detects the signal caused by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device and converts it into touch point coordinates, which are then sent to the processor 710, which receives and executes the command sent by the processor 710. In addition, the touch panel 7071 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch panel 7071, the user input unit 707 may also include other input devices 7072. Specifically, other input devices 7072 may include but are not limited to a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, and a joystick, which are not described in detail here.

[0158] Furthermore, the touch panel 7071 may be overlaid on the display panel 7061. When the touch panel 7071 detects a touch operation on or near it, it transmits the information to the processor 710 to determine the type of touch event. Subsequently, the processor 710 provides corresponding visual output on the display panel 7061 according to the type of touch event. Figure 7 In the figure, the touch panel 7071 and the display panel 7061 are two independent components to realize the input and output functions of the electronic device. However, in some embodiments, the touch panel 7071 and the display panel 7061 can be integrated to realize the input and output functions of the electronic device, which is not limited here.

[0159] The interface unit 708 is an interface for connecting external devices to the electronic device 700. For example, the external devices may include a wired or wireless headset port, an external power supply (or battery charger) port, a wired or wireless data port, a memory card port, a port for connecting a device with an identification module, an audio input / output (I / O) port, a video I / O port, a headphone port, etc. The interface unit 708 may be used to receive input (e.g., data information, power, etc.) from the external device and transmit the received input to one or more elements within the electronic device 700, or may be used to transmit data between the electronic device 700 and the external device.

[0160] Memory 709 can be used to store software programs and various data. Memory 709 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function or an image playback function); the data storage area may store data generated based on the use of the mobile phone (such as audio data, a phone book, etc.). Furthermore, memory 709 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0161] The processor 710 is the control center of the electronic device. It connects the various components of the electronic device using various interfaces and circuits. By running or executing software programs and / or modules stored in the memory 709 and accessing data stored in the memory 709, it performs various functions of the electronic device and processes data, thereby monitoring the electronic device as a whole. The processor 710 may include one or more processing units; preferably, the processor 710 may integrate an application processor and a modem processor, wherein the application processor primarily processes the operating system, user interface, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into the processor 710.

[0162] The electronic device 700 may also include a power supply 711 (such as a battery) to supply power to each component. Preferably, the power supply 711 may be logically connected to the processor 710 through a power management system, thereby managing functions such as charging, discharging, and power consumption through the power management system.

[0163] In addition, the electronic device 700 includes some functional modules not shown, which will not be described here.

[0164] Preferably, an embodiment of the present application also provides an electronic device, including a processor 710, a memory 709, and a computer program stored in the memory 709 and runnable on the processor 710. When the computer program is executed by the processor 710, it implements the above-mentioned voiceprint recognition model training method or each process of the above-mentioned voiceprint recognition method, and can achieve the same technical effect, which will not be repeated here.

[0165] The present application also provides a computer-readable storage medium having a computer program stored thereon. When executed by the processor 710, the computer program implements the above-described voiceprint recognition model training method or each process of the above-described voiceprint recognition method embodiment, and can achieve the same technical effect. To avoid repetition, the details are not described here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0166] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0167] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0168] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A voiceprint recognition model training method, characterized in that: The method comprises: Inputting the first voiceprint data into the teacher network and the student network included in the to-be-trained model, respectively, for the Nth iteration of training; wherein N is a positive integer, the teacher network has converged based on the training of the second voiceprint data, and the duration of the second voiceprint data is greater than the duration of the first voiceprint data; Calculating a first mean vector of a first eigenvector output by the teacher network and a second eigenvector output by the student network; When the output value corresponding to the first mean vector meets the first preset condition, the model to be trained after the Nth iterative training is determined as the voiceprint recognition model, and the voiceprint recognition model includes the teacher network and the student network after the Nth iterative training.

2. The method according to claim 1, characterized in that The first preset condition includes: the difference between the output value corresponding to the mean vector and the actual value of the voiceprint data input into the to-be-trained model during this training is less than a first preset value.

3. The method according to claim 1, characterized in that The teacher network is trained and converged according to the second voiceprint data in the following ways: Inputting the second voiceprint data into the teacher network and performing an M-th iterative training, where M is a positive integer less than N; When the third eigenvector output by the teacher network satisfies the second preset condition, it is determined that the teacher network converges after the M-th iterative training.

4. The method according to claim 3, characterized in that The second preset condition includes: the difference between the output value corresponding to the feature vector and the actual value input into the teacher network during this training is less than a second preset value.

5. The method according to claim 3, characterized in that The second voiceprint data includes at least one of a speech feature, a first-order difference feature, and a timbre information feature.

6. The method according to claim 1, characterized in that The number of first voiceprint data input into the student network is L, where L is a positive integer greater than 1; The calculating of a first mean vector of a first eigenvector output by the teacher network and a second eigenvector output by the student network comprises: Obtain L second eigenvectors output by the student network, wherein the second eigenvectors correspond one-to-one to the first voiceprint data input into the student network; Calculate the second mean vector of the L second eigenvectors; Obtain a first eigenvector output by the teacher network, and calculate a mean vector of the first eigenvector and the second mean vector to obtain the first mean vector, wherein the first eigenvector corresponds to the first voiceprint data input into the teacher network.

7. The method according to claim 1, characterized in that The teacher network and the student network are both ResNet-FPN network models, and the number of network layers of the student network is smaller than the number of network layers of the teacher network.

8. A voiceprint recognition method, characterized in that: A student network included in a voiceprint recognition model is applied, the student network is connected to a target classifier, and the voiceprint recognition model is trained using the voiceprint recognition model training method according to any one of claims 1 to 7, the method comprising: Obtaining first voiceprint data of the user to be identified; Inputting the first voiceprint data into the student network, and outputting a first feature vector corresponding to the first voiceprint data; Input the first feature vector and the pre-stored second feature vector into the target classifier, and output a likelihood distribution value; wherein the second feature vector is output by inputting the second voiceprint data of the target user into the student network; When the likelihood distribution value is greater than a preset threshold, it is determined that the user to be identified and the target user are the same user.

9. A voiceprint recognition model training device, characterized in that: include: A first training module is configured to input the first voiceprint data into a teacher network and a student network included in a to-be-trained model, respectively, for Nth iteration of training; wherein N is a positive integer, the teacher network has converged based on the training of the second voiceprint data, and the duration of the second voiceprint data is greater than the duration of the first voiceprint data; A calculation module, configured to calculate a first mean vector of a first eigenvector output by the teacher network and a second eigenvector output by the student network; The first determination module is used to determine the model to be trained after the Nth iterative training as the voiceprint recognition model when the output value corresponding to the first mean vector meets the first preset condition, and the voiceprint recognition model includes the teacher network and the student network after the Nth iterative training.

10. A voiceprint recognition device, characterized in that: A student network included in a voiceprint recognition model, the student network being connected to a target classifier, the voiceprint recognition model being trained using the voiceprint recognition model training method according to any one of claims 1 to 7, and the voiceprint recognition device comprising: An acquisition module, configured to acquire first voiceprint data of a user to be identified; a first output module, configured to input the first voiceprint data into the student network and output a first feature vector corresponding to the first voiceprint data; A second output module is configured to input the first feature vector and a pre-stored second feature vector into the target classifier and output a likelihood distribution value; wherein the second feature vector is output by inputting the second voiceprint data of the target user into the student network; The second determining module is configured to determine that the user to be identified and the target user are the same user if the likelihood distribution value is greater than a preset threshold.

11. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the voiceprint recognition model training method as described in any one of claims 1 to 7 are implemented, or when the processor executes the computer program, the steps of the voiceprint recognition method as described in claim 8 are implemented.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the voiceprint recognition model training method according to any one of claims 1 to 7, or, when executed by a processor, implements the steps of the voiceprint recognition method according to claim 8.

Images