Speech data anomaly classification method and device based on large model

Abstract

The application provides a large model-based voice data anomaly classification method and device, the method comprising: obtaining at least one to-be-classified text data segment feature of target service corresponding to-be-classified voice data; determining a target text classification category of each to-be-classified text data segment feature in a text classification category corresponding to the target service; determining the target text classification category of the target text classification category in the text classification category corresponding to the target service as a target text data segment feature; the text classification category corresponding to the target service is contained in the target text classification category; and classifying the voice data according to the target text classification category corresponding to all target text data segment features. The application classifies the voice data according to the target text classification category of the to-be-classified text data segment feature of the to-be-classified voice data, can not rely on a mobile phone number library, a bank card transaction library and an abnormal voice template library, and improves the anomaly classification accuracy.

Description

Technical Field

[0001] This invention relates to the field of speech classification technology, and more specifically, to a method and apparatus for anomaly classification of speech data based on large models. Background Technology

[0002] In recent years, with the continuous progress of the social economy and the rapid development of the Internet, the widespread use of smartphones and the rapid growth of the number of Internet users, voice data in telephone call records often contains anomalies, causing huge losses to the general public.

[0003] Currently, methods for anomaly classification of voice data are mainly divided into three categories based on their technical principles: manually constructing an abnormal mobile phone number database, building a classifier based on call detail records and bank card transaction data using machine learning, and classifying voice content as abnormal based on voice template matching methods. These methods have reduced the occurrence of losses to some extent.

[0004] However, the above methods heavily rely on existing databases, including mobile phone number databases, bank card transaction databases, and abnormal voice template databases. This makes it difficult to quickly and efficiently respond to the variability of abnormal voice messages when mobile phone numbers, bank card numbers, and voice scripts change. Therefore, the classification accuracy of the above methods is relatively low. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a method and apparatus for anomaly classification of speech data based on a large model. This method can classify speech data into anomalies by dividing the target text into categories according to the features of the text data segments to be classified. It does not rely on mobile phone number databases, bank card transaction databases, or anomaly speech template databases, thus improving the accuracy of anomaly classification.

[0006] In a first aspect, embodiments of this application provide a method for classifying speech data anomalies based on a large model, the method comprising:

[0007] Obtain at least one feature of the text data segment to be classified from the speech data to be classified corresponding to the target business obtained based on the large model;

[0008] Determine the target text classification category for each text data segment to be classified within the text classification category corresponding to the target business; each target business corresponds to multiple text classification categories.

[0009] The text segment features to be classified within the text classification category corresponding to the target business are determined as the target text segment features; the text classification category corresponding to the target business is included in the text segmentation category corresponding to the target business.

[0010] Anomaly classification is performed on the speech data based on the target text categories corresponding to the features of all target text data segments.

[0011] In one possible implementation, the text segmentation category corresponding to the target service is determined through the following steps:

[0012] Obtain features of at least one abnormal text data segment corresponding to the abnormal voice sample data of the target service;

[0013] Dimensionality reduction is performed on the features of each abnormal text data segment to obtain the two-dimensional features of each abnormal text data segment.

[0014] The number of clusters corresponding to the target business is determined based on the distance between the two-dimensional features of each abnormal text data segment.

[0015] Determine whether the number of clusters is greater than or equal to a preset value;

[0016] If the number of clusters is greater than or equal to a preset value, then all abnormal text data segments are classified according to the number of clusters to obtain the text classification category corresponding to the target business.

[0017] In one possible implementation, the method further includes:

[0018] If the number of clusters is less than the preset value, abnormal speech data other than abnormal speech sample data will be obtained from the abnormal speech database and added to the abnormal speech sample data.

[0019] Jump to obtain at least one abnormal text data segment feature corresponding to the abnormal voice sample data of the target service to continue execution.

[0020] In one possible implementation, obtaining at least one feature of a text data segment to be classified from the voice data to be classified corresponding to the target service includes:

[0021] Obtain the speech data to be classified;

[0022] Based on the discrimination criteria of the target business, the large model is used to determine whether the voice data to be classified is voice data related to the target business.

[0023] If the voice data to be classified is voice data related to the target business, then at least one feature of the text data segment to be classified in the voice data to be classified is obtained.

[0024] In one possible implementation, obtaining at least one feature of a text data segment from the speech data to be classified includes:

[0025] Convert the speech data to be classified into text data to be classified;

[0026] The text data to be classified is divided into short sentences to obtain at least one text data segment to be classified.

[0027] Extract sentence vectors from the text data segments to be classified to obtain the features of the text data segments to be classified.

[0028] In one possible implementation, anomaly classification of speech data is performed based on the target text classification categories corresponding to the features of all target text data segments, including:

[0029] Calculate the anomaly classification score of the speech data based on the preset anomaly score corresponding to the target text classification category based on the features of all target text data segments;

[0030] If the anomaly classification score is greater than or equal to the preset anomaly classification score, then the voice data is abnormal.

[0031] Secondly, embodiments of this application also provide a speech data anomaly classification device based on a large model, the speech data anomaly classification device based on a large model includes:

[0032] The acquisition module is used to acquire at least one feature of a text data segment to be classified from the speech data to be classified corresponding to the target business obtained based on the large model.

[0033] The determination module is used to determine the target text classification category of each text data segment to be classified in the text classification category corresponding to the target business; each target business corresponds to multiple text classification categories;

[0034] The determination module is also used to determine the features of the text data segments to be classified in the text classification category corresponding to the target business as the features of the target text data segments; the text classification category corresponding to the target business is included in the text classification category corresponding to the target business;

[0035] The discrimination module is used to classify speech data as anomalies based on the target text categories corresponding to the features of all target text data segments.

[0036] In one possible implementation, the determining module is further configured to:

[0037] Obtain features of at least one abnormal text data segment corresponding to the abnormal voice sample data of the target service;

[0038] Dimensionality reduction is performed on the features of each abnormal text data segment to obtain the two-dimensional features of each abnormal text data segment.

[0039] The number of clusters corresponding to the target business is determined based on the distance between the two-dimensional features of each abnormal text data segment.

[0040] Determine whether the number of clusters is greater than or equal to a preset value;

[0041] If the number of clusters is greater than or equal to a preset value, then all abnormal text data segments are classified according to the number of clusters to obtain the text classification category corresponding to the target business.

[0042] In one possible implementation, the device further includes: a jump module;

[0043] The acquisition module is also used to acquire abnormal speech data other than abnormal speech sample data from the abnormal speech database and add it to the abnormal speech sample data if the number of clusters is less than a preset value.

[0044] The jump module is used to jump to obtain at least one abnormal text data segment feature corresponding to the abnormal voice sample data of the target service in order to continue execution.

[0045] In one possible implementation, the acquisition module is specifically used to acquire the voice data to be classified; based on the discrimination conditions of the target service, determine whether the voice data to be classified is voice data related to the target service based on the large model; if the voice data to be classified is voice data related to the target service, then acquire at least one feature of the text data segment to be classified from the voice data to be classified.

[0046] In one possible implementation, the acquisition module is specifically used to convert the speech data to be classified into text data to be classified; to divide the text data to be classified into short sentences to obtain at least one text data segment to be classified; and to extract the sentence vector of the text data segment to be classified to obtain the features of the text data segment to be classified.

[0047] In one possible implementation, the discrimination module is specifically used to calculate the abnormal classification score of the speech data based on the preset abnormal score corresponding to the target text classification category of all target text data segment features; if the abnormal classification score is greater than or equal to the preset abnormal classification score, then the speech data is abnormal.

[0048] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a storage medium, and a bus. The storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the large-model-based speech data anomaly classification method as described in any of the first aspects.

[0049] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of any of the large-model-based speech data anomaly classification methods in the first aspect.

[0050] This application provides a method and apparatus for anomaly classification of voice data based on a large model. The method includes: acquiring at least one text data segment feature of the voice data to be classified corresponding to a target service; determining the target text classification category of each text data segment feature in the text classification category corresponding to the target service; each target service corresponds to multiple text classification categories; determining the text data segment feature of the target text classification category in the text classification category corresponding to the target service as the target text data segment feature; the text classification category corresponding to the target service is included in the text classification category corresponding to the target service; and classifying the voice data for anomalies based on the target text classification categories corresponding to all target text data segment features. This application classifies voice data for anomalies by using the target text classification categories of the text data segment features of the voice data to be classified, which can improve the accuracy of anomaly classification without relying on mobile phone number databases, bank card transaction databases, and abnormal voice template databases. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 A flowchart of a speech data anomaly classification method based on a large model provided in an embodiment of this application is shown;

[0053] Figure 2 A schematic diagram of the ICIO framework provided in an embodiment of this application is shown;

[0054] Figure 3 A flowchart of another speech data anomaly classification method based on a large model provided in an embodiment of this application is shown;

[0055] Figure 4 This illustration shows a structural schematic diagram of a speech data anomaly classification device based on a large model provided in an embodiment of this application;

[0056] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0058] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0059] To enable those skilled in the art to utilize the content of this application, and in conjunction with the specific application scenario of "speech classification technology," the following implementation methods are provided. For those skilled in the art, the general principles defined herein can be applied to other embodiments and application scenarios without departing from the spirit and scope of this application. Although this application is primarily described within the "speech classification technology field," it should be understood that this is merely an exemplary embodiment.

[0060] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0061] The following is a detailed description of a speech data anomaly classification method based on a large model provided in the embodiments of this application.

[0062] Reference Figure 1 The diagram shown is a flowchart illustrating a speech data anomaly classification method based on a large model provided in this application embodiment. The specific execution process of this speech data anomaly classification method based on a large model is as follows:

[0063] S101. Obtain at least one feature of the text data segment to be classified in the target business corresponding to the speech data to be classified based on the large model.

[0064] S102. Determine the target text classification category of each text data segment to be classified in the text classification category corresponding to the target business; each target business corresponds to multiple text classification categories.

[0065] S103. The text data segment features to be classified in the text classification category corresponding to the target business are determined as the target text data segment features; the text classification category corresponding to the target business is included in the text classification category corresponding to the target business.

[0066] S104. Classify speech data for anomalies based on the target text categories corresponding to the features of all target text data segments.

[0067] This application provides a method for anomaly classification of voice data based on a large model. The method includes: acquiring at least one text data segment feature of the voice data to be classified corresponding to a target service; determining the target text classification category of each text data segment feature within the text classification category corresponding to the target service; each target service corresponds to multiple text classification categories; determining the text data segment feature of the target text classification category within the text classification category corresponding to the target service as the target text data segment feature; the text classification category corresponding to the target service is included in the text classification category corresponding to the target service; and classifying the voice data for anomalies based on the target text classification categories corresponding to all target text data segment features. This application classifies voice data for anomalies by using the target text classification categories of the text data segment features of the voice data to be classified, which can improve the accuracy of anomaly classification without relying on mobile phone number databases, bank card transaction databases, and abnormal voice template databases.

[0068] The exemplary steps of the embodiments of this application are described below:

[0069] S101. Obtain at least one feature of the text data segment to be classified in the target business corresponding to the speech data to be classified based on the large model.

[0070] In this embodiment of the application, voice data to be classified is obtained; based on the discrimination criteria of the target service, it is determined whether the voice data to be classified is voice data related to the target service; if the voice data to be classified is voice data related to the target service, at least one feature of the text data segment to be classified of the voice data to be classified is obtained.

[0071] Here, the target business refers to the business where users need to classify voice anomalies, such as loan services. When the voice data to be classified is related to the target business, at least one feature of the text data segment to be classified is obtained from the voice data to be classified.

[0072] Specifically, based on the discrimination criteria of the target service, the large model is used to determine whether the voice data to be classified is related to the target service. This includes: inputting the target service for which voice anomaly classification needs to be performed, the discrimination criteria of the target service, the text content summary corresponding to the voice data to be classified, and the discrimination result output indication into the large model that incorporates the ICIO framework, to obtain the discrimination result of whether the voice data to be classified is related to the target service.

[0073] Here, to optimize the prompts and improve processing performance, we use the ICIO framework to refine the model input. For example... Figure 2 The figure shows a schematic diagram of the ICIO framework provided in the embodiment of this application. As shown in the figure, ICIO is a simple prompt word framework that decomposes the prompt word into four parts, which describe the task from four dimensions: instruction (i.e., the target service that needs to be classified for speech anomalies), background (i.e. the discrimination conditions of the target service), input data (i.e., the text content summary corresponding to the speech data to be classified), and output guidance (i.e. the discrimination result output indication).

[0074] Taking "target business is loans" as an example, according to the requirements of the ICIO framework, the four elements of instruction, background, input data, and output guidance are refined as follows: Instruction: Please help me determine whether the topic of a document is loans. Background information: 1. The document content is about the conditions, processes, terms, and operating steps in applying for a loan; 2. The document content is an on-site communication dialogue in the loan application process, usually occurring between customer service and customers; 3. Payments, withdrawals, transfers, app downloads, and mobile banking operations involved in loan application are all loan application operations; 4. The main content of the document is the terms or details of loan application. If a large amount of other content appears, the topic is not loans. Input data: The document is given in the form of key content points. Output guidance: The output format is JSON, including three fields: Loan content list, reason, and whether it meets the requirements; The "Loan content list" field lists the loan-related content that appears in the document, including application conditions, processes, terms, and operating steps, etc. If there is no such content, an empty list is output; The "reason" field is the reason for the judgment; "whether it meets the requirements" gives the final conclusion.

[0075] Connect the four elements from the above example into a coherent sentence in a logical order to obtain the final optimized version of the ICIO framework. The input is as follows: Please help me determine whether the topic of a document is a loan. The document is a phone record; only the key points are provided. Judge based on these points. The judgment rules are: 1. The document content is about the conditions, procedures, terms, and steps involved in applying for a loan; 2. The document content is an on-site communication dialogue during the loan application process, usually between customer service and the customer; 3. Payments, withdrawals, transfers, app downloads, and mobile banking operations involved in the loan application process are all loan application operations; 4. The main content of the document is the terms or details of the loan application. If a large amount of other content appears, the topic is not a loan. The output format is JSON, including three fields: Loan Content List, Reason, and Whether it meets the requirements. The "Loan Content List" field lists the loan-related content appearing in the document, including application conditions, procedures, terms, and steps. If none are found, an empty list is output. The "Reason" field is the reason for the judgment; "Whether it meets the requirements" gives the final conclusion. Below are the key points of the document content: ...

[0076] Specifically, obtaining features of at least one text data segment from the speech data to be classified includes: converting the speech data to be classified into text data to be classified; dividing the text data to be classified into short sentences to obtain at least one text data segment; and extracting sentence vectors from the text data segment to obtain the features of the text data segment to be classified.

[0077] S102. Determine the target text classification category of each text data segment to be classified in the text classification category corresponding to the target business; each target business corresponds to multiple text classification categories.

[0078] In this embodiment, the K-means clustering algorithm is used to determine the category of each text data segment to be classified within the text segmentation categories corresponding to the target business, thus obtaining the target text segmentation category corresponding to the features of each text data segment to be classified. Each text data segment to be classified corresponds to only one category in the text segmentation categories.

[0079] For example, if the text classification categories corresponding to the target business include a, b, and c, then the category to which the text data segment feature X belongs among a, b, and c is determined, thus obtaining the target text classification category of the text data segment feature X. The target text classification category is one of a, b, and c.

[0080] S103. The text data segment features to be classified in the text classification category corresponding to the target business are determined as the target text data segment features; the text classification category corresponding to the target business is included in the text classification category corresponding to the target business.

[0081] In this application's implementation, the text classification category is a category used to determine whether the voice data to be classified for a target service is abnormal, and it is related to the target service. Each target service corresponds to its own text classification category.

[0082] For example, the text segmentation categories corresponding to the target business include a, b, c, d, e, and f, and the text classification categories include a, c, and d. The target text segmentation category of the text data segment feature A to be classified is a, the target text segmentation category of the text data segment feature B to be classified is c, the target text segmentation category of the text data segment feature C to be classified is a, and the target text segmentation category of the text data segment feature D to be classified is f. Then, the text data segment features A, B, and C to be classified are determined as the target text data segment features.

[0083] Further, the text segmentation category corresponding to the target service is determined through the following steps: At least one abnormal text data segment feature corresponding to the abnormal voice sample data of the target service is obtained; the dimensionality of each abnormal text data segment feature is reduced to obtain a two-dimensional feature corresponding to each abnormal text data segment; the number of clusters corresponding to the target service is determined based on the distance between the two-dimensional features of each abnormal text data segment; it is determined whether the value of the number of clusters is greater than or equal to a preset value; if the value of the number of clusters is greater than or equal to the preset value, all abnormal text data segment features are categorized according to the number of clusters to obtain the text segmentation category corresponding to the target service; if the value of the number of clusters is less than the preset value, abnormal voice data other than the abnormal voice sample data is obtained from the abnormal voice database and added to the abnormal voice sample data; the process then jumps to obtaining at least one abnormal text data segment feature corresponding to the abnormal voice sample data of the target service to continue execution.

[0084] S104. Classify speech data for anomalies based on the target text categories corresponding to the features of all target text data segments.

[0085] In this embodiment, each target service's corresponding text classification category has its own preset score, which represents the weight of that text classification category in identifying speech anomalies. The target text segment features corresponding to the target text segment features are included in the text classification categories. For example, the anomaly identified by this method could be telecommunications fraud, etc., which is only one application scenario.

[0086] Specifically, the speech data is classified as anomaly based on the target text categories corresponding to the features of all target text data segments through the following steps:

[0087] Step 1: Calculate the anomaly classification score of the speech data based on the preset anomaly scores corresponding to the target text categories based on the features of all target text data segments.

[0088] In this embodiment of the application, the sum of the preset anomaly scores corresponding to the target text classification categories of all target text data segment features is determined as the anomaly classification score of the speech data.

[0089] For example, the text classification categories include a, c, and d. The preset score for text classification category a is 20, the preset score for text classification category c is 20, and the preset score for text classification category d is 10. The target text classification category for target text data segment feature A is a, the target text classification category for target text data segment feature B is d, and the target text classification category for target text data segment feature C is a. Then, the sum of the scores for target text classification category a of target text data segment feature A, the score for target text classification category d of target text data segment feature B, and the score for target text classification category a of target text data segment feature C is determined as the anomaly classification score of the speech data, i.e., 20 + 10 + 20 = 50.

[0090] Step 2: If the anomaly classification score is greater than or equal to the preset anomaly classification score, then the voice data is abnormal.

[0091] In this embodiment, the higher the anomaly classification score, the greater the likelihood of abnormal voice data. If the anomaly classification score is greater than or equal to a preset anomaly classification score, the voice data is abnormal; if the anomaly classification score is less than the preset anomaly classification score, the voice data is not abnormal.

[0092] Reference Figure 3 The diagram shown illustrates another speech data anomaly classification method based on a large model provided in this application. The exemplary steps of this application embodiment are described below:

[0093] S301. Obtain at least one abnormal text data segment feature corresponding to the abnormal voice sample data of the target service.

[0094] In this embodiment, a preset number of abnormal voice data are obtained from an abnormal voice database to obtain abnormal voice sample data; the abnormal voice database stores abnormal voice data corresponding to the target service; the abnormal voice sample data is converted into abnormal text data; the abnormal text data is divided into short sentences to obtain at least one initial abnormal text data segment; the initial abnormal text data segments whose length is outside the preset normal length range are removed to obtain the final abnormal text data segment; the sentence vector of the final abnormal text data segment is extracted to obtain the abnormal text data segment features.

[0095] The preset quantity is generally between one hundred and two hundred.

[0096] S302. Dimensionally reduce the features of each abnormal text data segment to obtain the two-dimensional features of each abnormal text data segment.

[0097] In this embodiment of the application, the t-SNE algorithm is used to reduce the dimensionality of the features of each abnormal text data segment to obtain the two-dimensional features of the abnormal text data segment corresponding to each abnormal text data segment.

[0098] Here, t-SNE stands for t-distributed stochastic neighbor embedding. Essentially, t-SNE is an embedding model that maps data from a high-dimensional space to a low-dimensional space while preserving the local characteristics of the dataset. t-SNE can be considered one of the most effective data dimensionality reduction and visualization methods currently available.

[0099] S303. Determine the number of clusters corresponding to the target business based on the distance between the two-dimensional features of each abnormal text data segment.

[0100] In this embodiment, it is determined whether the two-dimensional features of the current abnormal text data segment are assigned to any category. If not, the current abnormal text data segment and other abnormal text data segments whose distance from the current abnormal text data segment is within a preset range are assigned to the same category. The process continues to determine the two-dimensional features of the next abnormal text data segment. Otherwise, the process continues to determine the two-dimensional features of the next abnormal text data segment. The number of categories obtained after determining all abnormal text data segments is determined as the number of clusters corresponding to the target business. This number of clusters is the number of text classification categories corresponding to the target business.

[0101] In addition to determining the number of clusters corresponding to the target business based on the distance between the two-dimensional features of each abnormal text data segment, the distribution map of the two-dimensional features of the abnormal text data segments generated by the t-SNE algorithm can also be sent to the user so that the user can determine the number of clusters.

[0102] S304. Determine whether the number of clusters is greater than or equal to a preset value.

[0103] S305. If the number of clusters is greater than or equal to a preset value, then classify all abnormal text data segments according to the number of clusters to obtain the text classification category corresponding to the target business.

[0104] In this embodiment of the application, if the number of clusters is greater than or equal to a preset value, it indicates that the category of the abnormal text data segment features of the currently acquired abnormal voice sample data can comprehensively characterize the feature category of the abnormal voice data of the target service; then, based on the number of clusters, the K-means clustering algorithm is used to cluster the abnormal text data segment features to obtain the text classification category corresponding to the target service.

[0105] Taking loan as the target business as an example, the clustered text classification includes three main categories: the first category is loan business-related categories, the second category is common semantic categories in ordinary dialogues, and the third category is categories with messy content and no obvious theme.

[0106] Text related to loans is categorized as follows (i.e., the text category corresponding to the target business) (the numbers in parentheses are category IDs): Repayment (0), Cash flow (3), Income / Interest / Amount of Money (4), Error (9), Screenshot / Electronic Receipt (10), Contract / Order (11), Retry / Refresh (15), Review / Approval (18), ID Card (19), Credit Card / Bank Card / Mobile Banking (20), Received / Received / Disbursed (22), Data / Personal Information (23), Freeze / Unfreeze / Loan Fraud (24), Name / Mobile Number / Job (25), Withdrawal (27). Common semantic categories in ordinary conversations include: Emphasis (2), Reply / Contact (5), Inquiry / Confirmation (6), Urge / Hurry / As soon as possible (7), Wait a moment (8), Deny (12), Processing (13), Assumption / Rhetorical Question (21), How long later (28), Greeting (29). Categories without a theme: Classes 1, 14, 16, 17, and 26 (no special meaning, therefore no specific explanation).

[0107] Below are examples of abnormal text data segments for three categories:

[0108] Repayment (0): Remember to repay on time next month. The installment period can be up to three years. You can choose to repay in advance if you have personal funds. If you repay in advance, you can repay in advance. If you repay in advance, you don't have to pay interest later. You can repay on time. As long as you repay on time, you can repay in up to thirty-six installments. The monthly interest plus principal repayment, etc.

[0109] Reply / Contact (5): Sometimes I can't reply to you in time. Please call me. Please make a voice call. Please keep your phone accessible. You hung up the phone. Please contact me again. Please give me a reply. Or do you say you can't contact me? Didn't you say you would call me back after 12 o'clock? Please call me back again.

[0110] Category 1: After this matter is resolved, no matter what problems arise in the future, my name is Zhang San, employee number 1085. If it is convenient, it means that whatever industry you are currently engaged in, no matter how many phases you are divided into, as long as three months later, if you had known this situation earlier, you can ensure that all four corners can be photographed, etc.

[0111] S306. If the number of clusters is less than the preset value, then obtain abnormal voice data other than abnormal voice sample data from the abnormal voice database and add it to the abnormal voice sample data; jump to obtain at least one abnormal text data segment feature corresponding to the abnormal voice sample data of the target service to continue execution.

[0112] In this application embodiment, if the number of clusters is less than a preset value, it indicates that the category of the abnormal text data segment features of the currently acquired abnormal voice sample data cannot characterize the feature category of the abnormal voice data of the target service; then it is necessary to acquire a preset number of abnormal voice data from the abnormal voice database as abnormal voice sample data and continue to determine the text classification category corresponding to the target service.

[0113] Specifically, a preset number of abnormal voice data, excluding abnormal voice sample data, are obtained from the abnormal voice database and added to the abnormal voice sample data; then, the process jumps to obtain at least one abnormal text data segment feature corresponding to the abnormal voice sample data of the target service to continue execution.

[0114] This application provides another method for anomaly classification of voice data. This method can determine the text classification category corresponding to the target business. Compared with the model training method in the prior art, this method does not require a large number of samples for training, thus reducing the workload. In addition, this method determines the text classification category of the target business data by determining the categories of abnormal text sample data in abnormal voice sample data, which can be independent of mobile phone number databases, bank card transaction databases, and abnormal voice template databases, thereby improving the accuracy of anomaly classification.

[0115] Based on the same inventive concept, this application also provides a large-model-based speech data anomaly classification device corresponding to the large-model-based speech data anomaly classification method. Since the principle of the device in this application is similar to the large-model-based speech data anomaly classification method described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0116] Reference Figure 4 The diagram shown is a schematic of a speech data anomaly classification device based on a large model provided in an embodiment of this application. The speech data anomaly classification device based on a large model includes:

[0117] The acquisition module 401 is used to acquire at least one feature of a text data segment to be classified from the speech data to be classified corresponding to the target business obtained based on the large model.

[0118] The determination module 402 is used to determine the target text classification category of the features of each text data segment to be classified in the text classification category corresponding to the target business; each target business corresponds to multiple text classification categories;

[0119] The determination module 402 is further used to determine the unclassified text data segment features in the text classification category corresponding to the target business as the target text data segment features; the text classification category corresponding to the target business is included in the text classification category corresponding to the target business.

[0120] The discrimination module 403 is used to classify speech data as anomalies based on the target text classification corresponding to the features of all target text data segments.

[0121] In one possible implementation, the determining module 402 is further configured to:

[0122] Obtain features of at least one abnormal text data segment corresponding to the abnormal voice sample data of the target service;

[0123] Dimensionality reduction is performed on the features of each abnormal text data segment to obtain the two-dimensional features of each abnormal text data segment.

[0124] The number of clusters corresponding to the target business is determined based on the distance between the two-dimensional features of each abnormal text data segment.

[0125] Determine whether the number of clusters is greater than or equal to a preset value;

[0126] If the number of clusters is greater than or equal to a preset value, then all abnormal text data segments are classified according to the number of clusters to obtain the text classification category corresponding to the target business.

[0127] In one possible implementation, the device further includes: a jump module 404;

[0128] The acquisition module 401 is also used to acquire abnormal speech data other than abnormal speech sample data from the abnormal speech database and add it to the abnormal speech sample data if the number of clusters is less than a preset value.

[0129] Jump module 404 is used to jump to obtain at least one abnormal text data segment feature corresponding to the abnormal voice sample data of the target service in order to continue execution.

[0130] In one possible implementation, the acquisition module 401 is specifically used to acquire the voice data to be classified; based on the discrimination conditions of the target service, determine whether the voice data to be classified is voice data related to the target service based on the large model; if the voice data to be classified is voice data related to the target service, then acquire at least one feature of the text data segment to be classified of the voice data to be classified.

[0131] In one possible implementation, the acquisition module 401 is specifically used to convert the speech data to be classified into text data to be classified; to divide the text data to be classified into short sentences to obtain at least one text data segment to be classified; and to extract the sentence vector of the text data segment to be classified to obtain the features of the text data segment to be classified.

[0132] In one possible implementation, the discrimination module 403 is specifically used to calculate the abnormal classification score of the speech data based on the preset abnormal score corresponding to the target text classification category of all target text data segment features; if the abnormal classification score is greater than or equal to the preset abnormal classification score, then the speech data is abnormal.

[0133] This application provides an anomaly classification device for voice data. The device includes: an acquisition module 401, used to acquire at least one text data segment feature of the voice data to be classified corresponding to a target service, obtained based on a large model; a determination module 402, used to determine the target text classification category of each text data segment feature in the text classification category corresponding to the target service; each target service corresponds to multiple text classification categories; the determination module 402 is further used to determine the text data segment feature of the target text classification category in the text classification category corresponding to the target service as the target text data segment feature; the text classification category corresponding to the target service is included in the text classification category corresponding to the target service; and a discrimination module 403, used to perform anomaly classification on the voice data based on the target text classification categories corresponding to all target text data segment features. This application performs anomaly classification on voice data by using the target text classification category of the text data segment features of the voice data to be classified, which can improve the accuracy of anomaly classification without relying on mobile phone number databases, bank card transaction databases, and anomaly voice template databases.

[0134] like Figure 5 As shown in the embodiment of this application, an electronic device 500 includes a processor 501, a memory 502, and a bus. The memory 502 stores machine-readable instructions that can be executed by the processor 501. When the electronic device is running, the processor 501 communicates with the memory 502 through the bus. The processor 501 executes the machine-readable instructions to perform the steps of the above-described large-model-based voice data anomaly classification method.

[0135] Specifically, the memory 502 and processor 501 mentioned above can be general-purpose memory and processor, without any specific limitations. When the processor 501 runs the computer program stored in the memory 502, it can execute the above-mentioned speech data anomaly classification method based on a large model.

[0136] Corresponding to the above-described large-model-based speech data anomaly classification method, this application embodiment also provides a computer-readable storage medium storing a computer program, which, when run by a processor, executes the steps of the above-described large-model-based speech data anomaly classification method.

[0137] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0138] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0139] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0140] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the information processing methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0141] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for anomaly classification of speech data based on a large model, characterized in that, The speech data anomaly classification method based on large models includes: Obtain at least one feature of the text data segment to be classified from the speech data to be classified corresponding to the target business obtained based on the large model; Determine the target text classification category for each text data segment to be classified within the text classification category corresponding to the target service; each target service corresponds to multiple text classification categories. The text data segment features to be classified in the text classification category corresponding to the target business are determined as the target text data segment features; the text classification category corresponding to the target business is included in the text classification category corresponding to the target business; The speech data is classified into anomalies based on the target text categories corresponding to the features of all target text data segments. The text segmentation category corresponding to the target service is determined through the following steps: acquiring at least one abnormal text data segment feature corresponding to the abnormal voice sample data of the target service; performing dimensionality reduction on each abnormal text data segment feature to obtain a two-dimensional feature of the abnormal text data segment corresponding to each abnormal text data segment; determining the number of clusters corresponding to the target service based on the distance between the two-dimensional features of each abnormal text data segment; determining whether the value of the number of clusters is greater than or equal to a preset value; if the value of the number of clusters is greater than or equal to the preset value, then classifying all abnormal text data segment features according to the number of clusters to obtain the text segmentation category corresponding to the target service. The step of obtaining at least one text data segment feature of the speech data to be classified corresponding to the target business obtained based on the large model includes: obtaining the speech data to be classified; determining whether the speech data to be classified is speech data related to the target business based on the large model according to the discrimination conditions of the target business; if the speech data to be classified is speech data related to the target business, then obtaining at least one text data segment feature of the speech data to be classified. The step of obtaining at least one text data segment feature of the speech data to be classified includes: converting the speech data to be classified into text data to be classified; dividing the text data to be classified into short sentences to obtain at least one text data segment to be classified; and extracting the sentence vector of the text data segment to obtain the text data segment feature. The step of classifying the speech data for anomalies based on the target text classification categories corresponding to the features of all target text data segments includes: calculating the anomaly classification score of the speech data based on the preset anomaly scores corresponding to the target text classification categories corresponding to the features of all target text data segments; if the anomaly classification score is greater than or equal to the preset anomaly classification score, then the speech data is anomaly.

2. The speech data anomaly classification method based on a large model according to claim 1, characterized in that, The method further includes: If the number of clusters is less than the preset value, then abnormal speech data other than the abnormal speech sample data is obtained from the abnormal speech database and added to the abnormal speech sample data. Jump to the step of obtaining at least one abnormal text data segment feature corresponding to the abnormal voice sample data of the target service to continue execution.

3. A speech data anomaly classification device based on a large model, characterized in that, The speech data anomaly classification device based on a large model includes: The acquisition module is used to acquire at least one feature of a text data segment to be classified from the speech data to be classified corresponding to the target business obtained based on the large model. The determination module is used to determine the target text classification category of each text data segment to be classified in the text classification category corresponding to the target business; each target business corresponds to multiple text classification categories; The determining module is further configured to determine the unclassified text data segment features in the text classification category corresponding to the target service as the target text data segment features; the text classification category corresponding to the target service is included in the text classification category corresponding to the target service; The discrimination module is used to classify the speech data as anomalies based on the target text classification corresponding to the features of all target text data segments. The determining module is further configured to: acquire at least one abnormal text data segment feature corresponding to the abnormal voice sample data of the target service; reduce the dimensionality of each abnormal text data segment feature to obtain a two-dimensional feature of the abnormal text data segment corresponding to each abnormal text data segment; determine the number of clusters corresponding to the target service based on the distance between the two-dimensional features of each abnormal text data segment; determine whether the value of the number of clusters is greater than or equal to a preset value; if the value of the number of clusters is greater than or equal to the preset value, classify all abnormal text data segment features according to the number of clusters to obtain the text classification category corresponding to the target service; The acquisition module is specifically used to acquire the speech data to be classified; based on the discrimination conditions of the target service, determine whether the speech data to be classified is speech data related to the target service based on the large model; if the speech data to be classified is speech data related to the target service, then acquire at least one feature of the text data segment to be classified of the speech data to be classified. The acquisition module is specifically used to convert the speech data to be classified into text data to be classified; to divide the text data to be classified into short sentences to obtain at least one text data segment to be classified; and to extract the sentence vector of the text data segment to obtain the features of the text data segment to be classified. The discrimination module is specifically used to calculate the abnormal classification score of the speech data based on the preset abnormal score corresponding to the target text classification category of all target text data segment features; if the abnormal classification score is greater than or equal to the preset abnormal classification score, then the speech data is abnormal.

4. An electronic device, characterized in that, include: The device includes a processor, a storage medium, and a bus. The storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the large-model-based speech data anomaly classification method as described in claim 1 or 2.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the large-model-based speech data anomaly classification method as described in claim 1 or 2.

Images