Talk skill identification method and device, equipment and storage medium
Through the training of large language models to identify insurance fraud speech, the problem of inefficient recognition of insurance platforms is solved, the efficient identification and stability of insurance fraud speech is achieved, and manual intervention is reduced.
Patent Information
- Application Number
- CN202411284335.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-12
- Publication Date
- 2024-12-31
AI Technical Summary
Existing insurance platforms are inefficient in identifying insurance fraud speeches and are prone to inconsistent identification problems, resulting in escaping monitoring of insurance fraud.
Using the large language model training method, by obtaining the fraudulent and non-fraud samples in the preset recording, the model is trained to identify the fraudulent speech, and the model is optimized using the predefined loss value and convergence conditions to automatically identify the words in the insurance reporting phone number. Technique.
It improves the recognition efficiency and stability of insurance fraud tactics, reduces the need for manual identification, and reduces the identification time and economic losses.
Smart Images
Figure CN119228552A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of artificial intelligence and financial technology, and in particular to a speech recognition method, device, equipment and storage medium. Background Art
[0002] Insurance platforms play a vital role in modern society, providing individuals and businesses with indispensable services for financial protection and risk management. With the continuous expansion of insurance business, insurance fraud has become one of the serious challenges faced by insurance platforms. In order to illegally obtain insurance money, insurance fraudsters often use various insurance fraud methods. They are well versed in the insurance reporting process and are proficient in how to deal with claims adjusters to cleverly obtain compensation. The insurance reporting process of these insurance fraudsters is usually accompanied by carefully designed insurance fraud rhetoric.
[0003] However, the identification process of insurance fraud claims on existing insurance platforms is cumbersome, which is not conducive to improving the efficiency of identifying insurance fraud claims. The reason is that existing insurance platforms mainly use manual identification to identify insurance fraud claims on insurance platforms, and manual identification is inefficient and prone to identification inconsistencies. Because the same fraud claims may be handled by different customer service staff or claims personnel multiple times, which increases the difficulty of identification and also causes more insurance fraud behaviors to escape monitoring. Therefore, it is not conducive to improving the efficiency of identifying insurance fraud claims. Summary of the invention
[0004] The present invention provides a speech recognition method, device, computer equipment and storage medium to solve the technical problem that the recognition process of insurance fraud speech on the existing insurance platform is cumbersome and is not conducive to improving the recognition efficiency of insurance fraud speech.
[0005] In a first aspect, a method for speech recognition is provided, comprising:
[0006] Obtain preset recordings of the insurance platform, obtain insurance fraud recordings and non-insurance fraud recordings from the preset recordings, combine the insurance fraud recordings and insurance fraud speech labels to form insurance fraud samples, and combine the non-insurance fraud recordings and non-insurance fraud speech labels to form non-insurance fraud samples;
[0007] Selecting the insurance fraud sample or the non-insurance fraud sample as sample data, training a large language model based on the sample data, and obtaining a predicted speech label output by the large language model based on the sample data;
[0008] Obtaining a preset speech tag in the sample data, and obtaining a first loss value and a second loss value between the predicted speech tag and the preset speech tag through a predefined acquisition method;
[0009] Determine a total loss value of the predicted speech tag according to the first loss value and the second loss value;
[0010] When the total loss value satisfies a predefined convergence condition, stopping training the large language model and saving the trained large language model;
[0011] Obtain the insurance reporting phone number of the insurance platform, obtain the current recording of the insurance reporting phone, obtain the current speech label output by the trained large language model based on the current recording, and when the current speech label is the insurance fraud speech label, identify the current speech of the insurance reporting phone as insurance fraud speech.
[0012] Furthermore, the selecting the insurance fraud sample or the non-insurance fraud sample as sample data, training a large language model based on the sample data, and obtaining a predicted speech tag output by the large language model based on the sample data includes:
[0013] A plurality of different insurance fraud samples and a plurality of non-insurance fraud samples are combined into a data set, and the insurance fraud samples or the non-insurance fraud samples in the data set are randomly selected as sample data;
[0014] A large language model is trained based on the sample data, and a predicted speech label output by the large language model based on the sample data is obtained.
[0015] Furthermore, the step of obtaining a preset speech tag in the sample data and obtaining a first loss value and a second loss value between the predicted speech tag and the preset speech tag by a predefined acquisition method includes:
[0016] When the sample data is the insurance fraud sample, the insurance fraud speech tag is set to the preset speech tag in the sample data; when the sample data is the non-insurance fraud sample, the non-insurance fraud speech tag is set to the preset speech tag in the sample data;
[0017] Through the preset cosine similarity loss function, a first loss value between the predicted speech label and the preset speech label is obtained, and through the preset regularization loss function, a second loss value between the predicted speech label and the preset speech label is obtained.
[0018] Further, determining the total loss value of the predicted speech tag according to the first loss value and the second loss value includes:
[0019] Obtain a preset file, and import a preset total loss value generation model through the preset file;
[0020] A model is generated based on the first loss value, the second loss value and a preset total loss value to generate the total loss value of the predicted speech label.
[0021] Furthermore, when the total loss value satisfies a predefined convergence condition, stopping training the large language model and saving the trained large language model comprises:
[0022] Performing multiple rounds of training on the large language model, and obtaining a change in the total loss value during the multiple rounds of training;
[0023] A preset value corresponding to a convergence condition is obtained. When the change amount is less than the preset value, it is determined that the total loss value satisfies the convergence condition, training of the large language model is stopped, and the trained large language model is saved.
[0024] Further, the obtaining of the insurance reporting phone number of the insurance platform, obtaining the current recording of the insurance reporting phone number, obtaining the current speech tag output by the trained large language model based on the current recording, and when the current speech tag is the insurance fraud speech tag, identifying the current speech of the insurance reporting phone number as insurance fraud speech, includes:
[0025] Obtaining an insurance reporting phone number of the insurance platform, obtaining a current recording of the insurance reporting phone number, extracting a feature vector of the current recording, and inputting the feature vector into the trained large language model;
[0026] The current speech label output by the trained large language model based on the feature vector is obtained, and when the current speech label is the insurance fraud speech label, the current speech of the insurance reporting phone call is identified as insurance fraud speech.
[0027] Further, in the step of obtaining the insurance reporting phone number of the insurance platform, obtaining the current recording of the insurance reporting phone number, and obtaining the current speech tag output by the trained large language model based on the current recording, when the current speech tag is the insurance fraud speech tag, after identifying that the current speech of the insurance reporting phone number is the insurance fraud speech, the speech recognition method includes:
[0028] A display window corresponding to the insurance reporting telephone number is obtained, and an icon corresponding to the insurance fraud speech is displayed in the display window.
[0029] In a second aspect, a speech recognition device is provided, comprising:
[0030] The first acquisition module is used to acquire preset recordings of the insurance platform, acquire insurance fraud recordings and non-insurance fraud recordings in the preset recordings, combine the insurance fraud recordings and insurance fraud speech labels to form insurance fraud samples, and combine the non-insurance fraud recordings and non-insurance fraud speech labels to form non-insurance fraud samples;
[0031] A second acquisition module is used to select the insurance fraud sample or the non-insurance fraud sample as sample data, train a large language model based on the sample data, and obtain a predicted speech tag output by the large language model based on the sample data;
[0032] A third acquisition module is used to acquire a preset speech tag in the sample data, and acquire a first loss value and a second loss value between the predicted speech tag and the preset speech tag through a predefined acquisition method;
[0033] A determination module, configured to determine a total loss value of the predicted speech tag according to the first loss value and the second loss value;
[0034] A saving module, used for stopping the training of the large language model and saving the trained large language model when the total loss value satisfies a predefined convergence condition;
[0035] An identification module is used to obtain the insurance reporting phone number of the insurance platform, obtain the current recording of the insurance reporting phone number, obtain the current speech label output by the trained large language model based on the current recording, and when the current speech label is the insurance fraud speech label, identify the current speech of the insurance reporting phone number as insurance fraud speech.
[0036] In a third aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned speech recognition method when executing the computer program.
[0037] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned speech recognition method are implemented.
[0038] The present application provides a speech recognition method, apparatus, computer equipment and storage medium, which obtain preset recordings of an insurance platform, obtain insurance fraud recordings and non-insurance fraud recordings in the preset recordings, combine the insurance fraud recordings and insurance fraud speech labels into insurance fraud samples, and combine the non-insurance fraud recordings and non-insurance fraud speech labels into non-insurance fraud samples; select the insurance fraud samples or the non-insurance fraud samples as sample data, train a large language model based on the sample data, and obtain predicted speech labels output by the large language model based on the sample data; obtain the preset speech labels in the sample data, and obtain a first loss value and a second loss value between the predicted speech label and the preset speech label through a predefined acquisition method; determine the total loss value of the predicted speech label based on the first loss value and the second loss value; when the total loss value meets a predefined convergence condition, stop training the large language model and save the trained large language model ; Obtain the insurance reporting phone number of the insurance platform, obtain the current recording of the insurance reporting phone, obtain the current speech label output by the trained large language model based on the current recording, and when the current speech label is the insurance fraud speech label, identify the current speech of the insurance reporting phone as insurance fraud speech. The beneficial effects are in two aspects. On the one hand, obtain the insurance reporting phone number of the insurance platform, obtain the current recording of the insurance reporting phone, obtain the current speech label output by the trained large language model based on the current recording, and when the current speech label is the insurance fraud speech label, identify the current speech of the insurance reporting phone as insurance fraud speech. Since no manual recognition is required, the recognition time of the insurance platform's insurance fraud speech is reduced, which is conducive to improving the recognition efficiency of the insurance platform's insurance fraud speech. On the other hand, since the trained large language model is not prone to recognition inconsistencies, it is conducive to improving the stability of identifying insurance fraud speech. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.
[0040] Figure 1 This is a schematic diagram of an application environment of a speech recognition method in one embodiment of the present invention;
[0041] Figure 2 A flowchart of a method for speech recognition provided by an embodiment of the present invention;
[0042] Figure 3 yes Figure 1 A schematic flow chart of a specific implementation of step S23;
[0043] Figure 4 yes Figure 1 A schematic flow chart of a specific implementation of step S25;
[0044] Figure 5 yes Figure 1 A schematic flow chart of a specific implementation of step S26;
[0045] Figure 6 is a structural schematic diagram of a speech recognition device in one embodiment of the present invention;
[0046] Figure 7 is a schematic diagram of a structure of a computer device in one embodiment of the present invention;
[0047] Figure 8 It is another structural schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION
[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of 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.
[0049] See also Figure 1 , Figure 1 is a schematic diagram of an application environment of a speech recognition method in an embodiment of the present invention. The speech recognition method provided by the embodiment of the present invention can be applied in Figure 1 In an application environment, a client communicates with a server through a network.
[0050] The server obtains preset recordings of the insurance platform through the client, obtains insurance fraud recordings and non-insurance fraud recordings from the preset recordings, combines the insurance fraud recordings and insurance fraud speech labels into insurance fraud samples, and combines the non-insurance fraud recordings and non-insurance fraud speech labels into non-insurance fraud samples;
[0051] Selecting the insurance fraud sample or the non-insurance fraud sample as sample data, training a large language model based on the sample data, and obtaining a predicted speech label output by the large language model based on the sample data;
[0052] Obtaining a preset speech tag in the sample data, and obtaining a first loss value and a second loss value between the predicted speech tag and the preset speech tag through a predefined acquisition method;
[0053] Determine a total loss value of the predicted speech tag according to the first loss value and the second loss value;
[0054] When the total loss value satisfies a predefined convergence condition, stopping training the large language model and saving the trained large language model;
[0055] Obtain the insurance reporting phone number of the insurance platform, obtain the current recording of the insurance reporting phone, obtain the current speech label output by the trained large language model based on the current recording, and when the current speech label is the insurance fraud speech label, identify the current speech of the insurance reporting phone as insurance fraud speech.
[0056] In the embodiment of the present invention, compared with the time-consuming and labor-intensive traditional manual recognition method, when the current speech tag is the insurance fraud speech tag, this speech recognition method can automatically identify the current speech of the insurance call as insurance fraud speech, which greatly reduces the workload of customer service personnel or claims personnel, allowing customer service personnel or claims personnel to focus more on other core tasks, improves recognition efficiency, and achieves cost reduction and efficiency improvement. By timely discovering and preventing insurance fraud, insurance companies can avoid paying illegal claims, thereby reducing economic losses. This technology is not only applicable to auto insurance fraud, but can also be used for other types of insurance, such as medical insurance and property insurance. This means that the speech recognition method has broad application prospects and can improve the ability to identify and prevent insurance fraud.
[0057] The client may include but is not limited to various personal computers, laptops, smart phones, tablet computers, and portable wearable devices.
[0058] The server can be implemented with an independent task database or a task database cluster composed of multiple task databases. The present invention is described in detail below through specific embodiments.
[0059] See also Figure 2 , Figure 2 A flowchart of a speech recognition method provided by an embodiment of the present invention includes the following steps:
[0060] S21, obtaining preset recordings of the insurance platform, obtaining insurance fraud recordings and non-insurance fraud recordings from the preset recordings, combining the insurance fraud recordings and insurance fraud speech labels to form insurance fraud samples, and combining the non-insurance fraud recordings and non-insurance fraud speech labels to form non-insurance fraud samples;
[0061] Among them, the insurance platform is an online platform that provides the purchase and management functions of insurance services. This platform is usually provided by insurance companies or brokerage companies, and provides consumers with a convenient and efficient insurance service channel through digital technology.
[0062] For example, an insurance fraud recording is a recording of defrauding insurance money.
[0063] Exemplarily, a non-insurance fraud recording is a recording that is not intended to defraud insurance money.
[0064] Exemplarily, the insurance fraud speech label is a label for speech used to defraud insurance money.
[0065] Exemplarily, the non-insurance fraud speech label is a label for speech that is not used to defraud insurance money.
[0066] For example, insurance fraud rhetoric is rhetoric used to defraud insurance money.
[0067] For example, non-insurance fraud sales talk is sales talk for obtaining insurance money normally.
[0068] Among them, insurance fraud tactics usually have the following characteristics:
[0069] Fabrication of facts: The policyholder, the insured or the beneficiary fabricates insurance accidents, exaggerates the extent of losses or forges supporting documents in order to defraud insurance money.
[0070] Concealing the truth: intentionally concealing important facts related to an insurance accident or providing false information in order to evade investigation or audit by the insurance company.
[0071] Exploiting loopholes: Taking advantage of loopholes in insurance terms or management loopholes in the insurance company to obtain insurance money through improper means.
[0072] Exemplarily, obtaining preset recordings of an insurance platform, obtaining insurance fraud recordings and non-insurance fraud recordings in the preset recordings, combining the insurance fraud recordings and insurance fraud speech labels to form insurance fraud samples, and combining the non-insurance fraud recordings and non-insurance fraud speech labels to form non-insurance fraud samples, including:
[0073] Obtain a database of the insurance platform, obtain preset recordings in the database through an access interface of the database, obtain insurance fraud recordings and non-insurance fraud recordings in the preset recordings, combine the insurance fraud recordings and insurance fraud speech labels into insurance fraud samples, and combine the non-insurance fraud recordings and non-insurance fraud speech labels into non-insurance fraud samples.
[0074] S22, selecting the insurance fraud sample or the non-insurance fraud sample as sample data, training a large language model based on the sample data, and obtaining a predicted speech tag output by the large language model based on the sample data;
[0075] The selecting the insurance fraud sample or the non-insurance fraud sample as sample data, training a large language model based on the sample data, and obtaining a predicted speech label output by the large language model based on the sample data includes:
[0076] A plurality of different insurance fraud samples and a plurality of non-insurance fraud samples are combined into a data set, and the insurance fraud samples or the non-insurance fraud samples in the data set are randomly selected as sample data;
[0077] A large language model is trained based on the sample data, and a predicted speech label output by the large language model based on the sample data is obtained.
[0078] S23, obtaining a preset speech tag in the sample data, and obtaining a first loss value and a second loss value between the predicted speech tag and the preset speech tag through a predefined acquisition method;
[0079] S24, determining a total loss value of the predicted speech tag according to the first loss value and the second loss value;
[0080] The step of determining the total loss value of the predicted speech tag according to the first loss value and the second loss value includes:
[0081] Obtain a preset file, and import a preset total loss value generation model through the preset file;
[0082] A model is generated based on the first loss value, the second loss value and a preset total loss value to generate the total loss value of the predicted speech label.
[0083] Exemplarily, a total loss value of the predicted speech tag is generated according to the first loss value, the second loss value and a preset total loss value generation model;
[0084] Among them, the total loss value generation model is:
[0085] Loss = α*Loss1+β*Loss2
[0086] Among them, Loss1 represents the first loss value, Loss2 represents the second loss value, and α and β are hyperparameters.
[0087] Among them, hyperparameters are tuning parameters in machine learning algorithms, which are used to control the learning and training processes of the algorithm to improve algorithm performance and model accuracy. Hyperparameters are set before model training and remain unchanged during the training process.
[0088] S25, when the total loss value satisfies a predefined convergence condition, stop training the large language model, and save the trained large language model;
[0089] S26, obtaining the insurance reporting phone number of the insurance platform, obtaining the current recording of the insurance reporting phone number, obtaining the current speech tag output by the trained large language model based on the current recording, and when the current speech tag is the insurance fraud speech tag, identifying the current speech of the insurance reporting phone number as insurance fraud speech.
[0090] Among them, the insurance reporting phone number is an emergency contact method provided by the insurance company to customers. When customers encounter an insurance accident or need to report an accident, they can call this insurance reporting phone number to report the accident to the insurance company and start the claim process. The phone number of the insurance reporting phone number is usually clearly marked in the insurance contract, insurance policy or the official website and promotional materials of the insurance company.
[0091] Among them, in the step of obtaining the insurance reporting phone number of the insurance platform, obtaining the current recording of the insurance reporting phone number, obtaining the current speech tag output by the trained large language model based on the current recording, when the current speech tag is the insurance fraud speech tag, after identifying that the current speech of the insurance reporting phone number is the insurance fraud speech, the speech recognition method includes:
[0092] A display window corresponding to the insurance reporting telephone number is obtained, and an icon corresponding to the insurance fraud speech is displayed in the display window.
[0093] Among them, icons can be used to remind customer service personnel or claims personnel, which is helpful for customer service personnel or claims personnel to promptly discover and prevent insurance fraud.
[0094] In the embodiment of the present invention, the beneficial effects are in two aspects. On the one hand, the insurance reporting phone number of the insurance platform is obtained, the current recording of the insurance reporting phone number is obtained, and the current speech label output by the trained large language model based on the current recording is obtained. When the current speech label is the insurance fraud speech label, the current speech of the insurance reporting phone number is identified as insurance fraud speech. Since manual identification is not required, the recognition time of the insurance fraud speech of the insurance platform is reduced, which is conducive to improving the recognition efficiency of the insurance fraud speech of the insurance platform. On the other hand, since the trained large language model is not prone to recognition inconsistencies, it is conducive to improving the stability of identifying insurance fraud speech.
[0095] See also Figure 3 , Figure 3 yes Figure 1 A specific implementation flow diagram of step S23 is described in detail as follows:
[0096] S31, when the sample data is the insurance fraud sample, the insurance fraud speech tag is set as a preset speech tag in the sample data; when the sample data is the non-insurance fraud sample, the non-insurance fraud speech tag is set as a preset speech tag in the sample data;
[0097] S32, obtaining a first loss value between the predicted speech label and the preset speech label through a preset cosine similarity loss function, and obtaining a second loss value between the predicted speech label and the preset speech label through a preset regularization loss function.
[0098] Among them, the cosine similarity loss function (Cosine Similarity Loss) is mainly used to measure the similarity between the vector of the predicted speech label and the vector of the preset speech label, and calculate the first loss value based on this.
[0099] Among them, when the vectors of the two predicted speech labels are exactly the same as the vector of the preset speech label, that is, the angle is 0 degrees, the cosine similarity is 1, and the first loss value is 0, indicating that there is no error. When the vector of the predicted speech label is completely dissimilar to the vector of the preset speech label, that is, the angle is 180 degrees, the cosine similarity is -1, and the first loss value is 2, indicating a maximum error. Within the range of 0 to 180 degrees, the larger the angle, the lower the cosine similarity, the higher the first loss value, and the greater the degree of difference between the vectors of the two predicted speech labels and the vector of the preset speech label.
[0100] Among them, the regularized loss function usually consists of two parts: the original loss function and the regularization term. The original loss function measures the degree of fit of the large language model to the training data. The original loss function includes the mean square error loss function and the cross entropy loss function. The regularization term is a measure of the complexity of the large language model and is used to improve the generalization ability of the large language model.
[0101] In the embodiment of the present invention, the performance of the large language model is improved by using the first loss value and the second loss value.
[0102] See also Figure 4 , Figure 4 yes Figure 1 A specific implementation flow diagram of step S25 is described in detail as follows:
[0103] S41, performing multiple rounds of training on the large language model, and obtaining a change in the total loss value during the multiple rounds of training;
[0104] S42, obtaining a preset value corresponding to a convergence condition, and when the change amount is less than the preset value, determining that the total loss value satisfies the convergence condition, stopping training of the large language model, and saving the trained large language model.
[0105] Among them, when the change amount is less than a preset value, it is determined that the total loss value meets the convergence condition, that is, the total loss value has converged. At this time, the training of the large language model is stopped and the trained large language model is saved.
[0106] For the sake of illustration, an example is given below:
[0107] For example, the large language model is trained for three rounds, and during the three rounds of training, the changes in the total loss value are 0.15, 0.15, and 0.13 respectively;
[0108] When the preset value is 0.2, the change amount is less than the preset value, it is determined that the total loss value meets the convergence condition, the training of the large language model is stopped, and the trained large language model is saved.
[0109] In the embodiment of the present invention, training a large language model requires huge computing resources and time. Saving the trained large language model can avoid repeated training, saving training time and training cost.
[0110] See also Figure 5 , Figure 5 yes Figure 1 A specific implementation flow diagram of step S26 is described in detail as follows:
[0111] S51, obtaining the insurance reporting phone number of the insurance platform, obtaining the current recording of the insurance reporting phone number, extracting the feature vector of the current recording, and inputting the feature vector into the trained large language model;
[0112] S52, obtaining the current speech label output by the trained large language model based on the feature vector, and when the current speech label is the insurance fraud speech label, identifying the current speech of the insurance reporting call as insurance fraud speech.
[0113] In the embodiment of the present invention, since manual identification is not required, the time for identifying insurance fraud speech on the insurance platform is reduced, which is conducive to improving the efficiency of identifying insurance fraud speech on the insurance platform.
[0114] See also Figure 6 , Figure 6 is a schematic diagram of a structure of a speech recognition device in one embodiment of the present invention. Figure 6 As shown, the speech recognition device includes a first acquisition module 101, a second acquisition module 102, a third acquisition module 103, a determination module 104, a storage module 105, and an identification module 106. The functional modules are described in detail as follows:
[0115] The first acquisition module 101 is used to acquire preset recordings of the insurance platform, acquire insurance fraud recordings and non-insurance fraud recordings in the preset recordings, combine the insurance fraud recordings and insurance fraud speech labels to form insurance fraud samples, and combine the non-insurance fraud recordings and non-insurance fraud speech labels to form non-insurance fraud samples;
[0116] A second acquisition module 102 is used to select the insurance fraud sample or the non-insurance fraud sample as sample data, train a large language model based on the sample data, and obtain a predicted speech tag output by the large language model based on the sample data;
[0117] The third acquisition module 103 is used to acquire the preset speech tag in the sample data, and acquire the first loss value and the second loss value between the predicted speech tag and the preset speech tag through a predefined acquisition method;
[0118] A determination module 104, configured to determine a total loss value of the predicted speech tag according to the first loss value and the second loss value;
[0119] A saving module 105 is used to stop training the large language model and save the trained large language model when the total loss value meets a predefined convergence condition;
[0120] The identification module 106 is used to obtain the insurance reporting phone number of the insurance platform, obtain the current recording of the insurance reporting phone number, obtain the current speech label output by the trained large language model based on the current recording, and when the current speech label is the insurance fraud speech label, identify the current speech of the insurance reporting phone number as insurance fraud speech.
[0121] In one embodiment, the second acquisition module 102 includes:
[0122] An extraction subunit, used for forming a data set from a plurality of different insurance fraud samples and a plurality of non-insurance fraud samples, and randomly extracting the insurance fraud samples or the non-insurance fraud samples from the data set as sample data;
[0123] An acquisition subunit is used to train a large language model based on the sample data and acquire predicted speech tags output by the large language model based on the sample data.
[0124] In one embodiment, the third acquisition module 103 includes:
[0125] A setting subunit is used to set the insurance fraud speech tag to a preset speech tag in the sample data when the sample data is the insurance fraud sample, and to set the non-insurance fraud speech tag to a preset speech tag in the sample data when the sample data is the non-insurance fraud sample;
[0126] An acquisition subunit is used to obtain a first loss value between the predicted speech label and the preset speech label through a preset cosine similarity loss function, and to obtain a second loss value between the predicted speech label and the preset speech label through a preset regularization loss function.
[0127] In one embodiment, the determining module 104 includes:
[0128] An import subunit is used to obtain a preset file and import a preset total loss value generation model through the preset file;
[0129] A generating subunit is used to generate a model according to the first loss value, the second loss value and a preset total loss value to generate the total loss value of the predicted speech label.
[0130] In one embodiment, the storage module 105 includes:
[0131] A training subunit, used to perform multiple rounds of training on the large language model, and obtain a change in the total loss value during the multiple rounds of training;
[0132] The saving subunit is used to obtain a preset value corresponding to a convergence condition. When the change amount is less than the preset value, it is determined that the total loss value meets the convergence condition, the training of the large language model is stopped, and the trained large language model is saved.
[0133] In one embodiment, the identification module 106 includes:
[0134] An input subunit, used to obtain the insurance reporting phone number of the insurance platform, obtain the current recording of the insurance reporting phone number, extract the feature vector of the current recording, and input the feature vector into the trained large language model;
[0135] The identification subunit is used to obtain the current speech label output by the trained large language model based on the feature vector, and when the current speech label is the insurance fraud speech label, identify the current speech of the insurance reporting phone call as insurance fraud speech.
[0136] In one embodiment, the speech recognition device further includes:
[0137] The display module is used to obtain a display window corresponding to the insurance reporting telephone number, and display an icon corresponding to the insurance fraud speech through the display window.
[0138] In the embodiment of the present invention, the beneficial effects are in two aspects. On the one hand, the insurance reporting phone number of the insurance platform is obtained, the current recording of the insurance reporting phone number is obtained, and the current speech label output by the trained large language model based on the current recording is obtained. When the current speech label is the insurance fraud speech label, the current speech of the insurance reporting phone number is identified as insurance fraud speech. Since manual identification is not required, the recognition time of the insurance fraud speech of the insurance platform is reduced, which is conducive to improving the recognition efficiency of the insurance fraud speech of the insurance platform. On the other hand, since the trained large language model is not prone to recognition inconsistencies, it is conducive to improving the stability of identifying insurance fraud speech.
[0139] For the specific limitations of the speech recognition device, please refer to the limitations of the speech recognition method above, which will not be repeated here.
[0140] Each module in the above speech recognition device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.
[0141] See also Figure 7 , Figure 7 is a schematic diagram of a computer device in one embodiment of the present invention. In one embodiment, a computer device is provided. The computer device may be a server. The internal structure diagram thereof may be as shown in FIG. Figure 7 The computer device includes a processor, a memory, a network interface and a database connected through a system bus.
[0142] The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client through a network connection. When the computer program is executed by the processor, the functions or steps of a speech recognition method can be implemented on the server side.
[0143] See also Figure 8 , Figure 8 is another structural diagram of a computer device in one embodiment of the present invention. In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as shown in FIG. Figure 8 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external task database via a network connection. When the computer program is executed by the processor, the functions or steps of a speech recognition method can be implemented on the client.
[0144] In one embodiment, a computer device is provided, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor.
[0145] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions of the server and the client in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.
[0146] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a graphics processing unit (GPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.
[0147] The above description and accompanying drawings fully illustrate the embodiments of the present disclosure so that those skilled in the art can practice them. Other embodiments may include structural, logical, electrical, process and other changes. The embodiments represent possible changes only. Unless explicitly required, separate components and functions are optional, and the order of operation may vary. Parts and subsamples of some embodiments may be included in or replace parts and subsamples of other embodiments. Moreover, the words used in this application are only used to describe the embodiments and are not used to limit the claims. As used in the description of the embodiments and claims, unless the context clearly indicates, the singular forms of "a", "an" and "the" are intended to include plural forms as well. Similarly, the term "and / or" as used in this application refers to any and all possible combinations of listings containing one or more associated ones. In addition, when used in the present application, the term "comprise" and its variants "comprises" and / or comprising refer to the presence of a stated sub-sample, whole, step, operation, element, and / or component, but do not exclude the presence or addition of one or more other sub-samples, wholes, steps, operations, elements, components and / or groups of these. In the absence of further restrictions, the elements defined by the sentence "comprising a ..." do not exclude the presence of other identical elements in the process, method or device comprising the elements. In this article, each embodiment may focus on the differences from other embodiments, and the same and similar parts between the various embodiments may refer to each other. For the methods, products, etc. disclosed in the embodiments, if they correspond to the method part disclosed in the embodiments, then the relevant parts can refer to the description of the method part.
[0148] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software may depend on the specific application and design constraints of the technical solution. Technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present disclosure. Technicians can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.
[0149] In the embodiments disclosed herein, the disclosed methods and products (including but not limited to devices, equipment, etc.) can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units can be only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some sub-samples can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to implement this embodiment. In addition, each functional unit in the embodiment of the present disclosure may be integrated in a processing unit, or each unit may exist physically alone, or two or more units may be integrated in one unit.
[0150] The flowchart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to the embodiment of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the 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. 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 consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. In the description corresponding to the flowchart and the block diagram in the accompanying drawings, the operations or steps corresponding to different boxes can also occur in a different order from the order disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. Each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified functions or actions, or may be implemented by a combination of dedicated hardware and computer instructions.
Claims
1. A speech recognition method, characterized in that: include: Obtain preset recordings of the insurance platform, obtain insurance fraud recordings and non-insurance fraud recordings from the preset recordings, combine the insurance fraud recordings and insurance fraud speech labels to form insurance fraud samples, and combine the non-insurance fraud recordings and non-insurance fraud speech labels to form non-insurance fraud samples; Selecting the insurance fraud sample or the non-insurance fraud sample as sample data, training a large language model based on the sample data, and obtaining a predicted speech label output by the large language model based on the sample data; Obtaining a preset speech tag in the sample data, and obtaining a first loss value and a second loss value between the predicted speech tag and the preset speech tag through a predefined acquisition method; Determine a total loss value of the predicted speech tag according to the first loss value and the second loss value; When the total loss value satisfies a predefined convergence condition, stopping training the large language model and saving the trained large language model; Obtain the insurance reporting phone number of the insurance platform, obtain the current recording of the insurance reporting phone, obtain the current speech label output by the trained large language model based on the current recording, and when the current speech label is the insurance fraud speech label, identify the current speech of the insurance reporting phone as insurance fraud speech.
2. The speech recognition method according to claim 1, characterized in that: The selecting the insurance fraud sample or the non-insurance fraud sample as sample data, training a large language model based on the sample data, and obtaining a predicted speech label output by the large language model based on the sample data includes: A plurality of different insurance fraud samples and a plurality of non-insurance fraud samples are combined into a data set, and the insurance fraud samples or the non-insurance fraud samples in the data set are randomly selected as sample data; A large language model is trained based on the sample data, and a predicted speech label output by the large language model based on the sample data is obtained.
3. The speech recognition method according to claim 1, characterized in that: The obtaining of the preset speech tag in the sample data, and obtaining the first loss value and the second loss value between the predicted speech tag and the preset speech tag by a predefined obtaining method, includes: When the sample data is the insurance fraud sample, the insurance fraud speech tag is set to the preset speech tag in the sample data; when the sample data is the non-insurance fraud sample, the non-insurance fraud speech tag is set to the preset speech tag in the sample data; Through the preset cosine similarity loss function, a first loss value between the predicted speech label and the preset speech label is obtained, and through the preset regularization loss function, a second loss value between the predicted speech label and the preset speech label is obtained.
4. The speech recognition method according to claim 1, characterized in that: The determining, according to the first loss value and the second loss value, a total loss value of the predicted speech tag includes: Obtain a preset file, and import a preset total loss value generation model through the preset file; A model is generated based on the first loss value, the second loss value and a preset total loss value to generate the total loss value of the predicted speech label.
5. The speech recognition method according to claim 1, characterized in that: When the total loss value satisfies a predefined convergence condition, stopping training the large language model and saving the trained large language model comprises: Performing multiple rounds of training on the large language model, and obtaining a change in the total loss value during the multiple rounds of training; A preset value corresponding to a convergence condition is obtained. When the change amount is less than the preset value, it is determined that the total loss value satisfies the convergence condition, training of the large language model is stopped, and the trained large language model is saved.
6. The speech recognition method according to claim 1, characterized in that: The step of obtaining the insurance reporting phone number of the insurance platform, obtaining the current recording of the insurance reporting phone number, obtaining the current speech tag output by the trained large language model based on the current recording, and identifying the current speech of the insurance reporting phone number as insurance fraud speech when the current speech tag is the insurance fraud speech tag, includes: Obtaining an insurance reporting phone number of the insurance platform, obtaining a current recording of the insurance reporting phone number, extracting a feature vector of the current recording, and inputting the feature vector into the trained large language model; The current speech label output by the trained large language model based on the feature vector is obtained, and when the current speech label is the insurance fraud speech label, the current speech of the insurance reporting phone call is identified as insurance fraud speech.
7. The speech recognition method according to claim 1, characterized in that: In the step of obtaining the insurance reporting phone number of the insurance platform, obtaining the current recording of the insurance reporting phone number, and obtaining the current speech tag output by the trained large language model based on the current recording, when the current speech tag is the insurance fraud speech tag, after identifying the current speech of the insurance reporting phone number as insurance fraud speech, the speech recognition method includes: A display window corresponding to the insurance reporting telephone number is obtained, and an icon corresponding to the insurance fraud speech is displayed in the display window.
8. A speech recognition device, characterized in that: include: The first acquisition module is used to acquire preset recordings of the insurance platform, acquire insurance fraud recordings and non-insurance fraud recordings in the preset recordings, combine the insurance fraud recordings and insurance fraud speech labels to form insurance fraud samples, and combine the non-insurance fraud recordings and non-insurance fraud speech labels to form non-insurance fraud samples; A second acquisition module is used to select the insurance fraud sample or the non-insurance fraud sample as sample data, train a large language model based on the sample data, and obtain a predicted speech tag output by the large language model based on the sample data; A third acquisition module is used to acquire a preset speech tag in the sample data, and acquire a first loss value and a second loss value between the predicted speech tag and the preset speech tag through a predefined acquisition method; A determination module, configured to determine a total loss value of the predicted speech tag according to the first loss value and the second loss value; A saving module, used for stopping the training of the large language model and saving the trained large language model when the total loss value meets a predefined convergence condition; An identification module is used to obtain the insurance reporting phone number of the insurance platform, obtain the current recording of the insurance reporting phone number, obtain the current speech label output by the trained large language model based on the current recording, and when the current speech label is the insurance fraud speech label, identify the current speech of the insurance reporting phone number as insurance fraud speech.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the speech recognition method as described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the speech recognition method as described in any one of claims 1 to 7 are implemented.