Call recording violation detection method and device, computer equipment and storage medium
Automatically detect call recordings through voice conversion and violation detection models, solving the problems of low efficiency and accuracy in the existing technology, and achieving efficient and accurate violation detection.
Patent Information
- Application Number
- CN202510664344.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-21
- Publication Date
- 2025-08-22
AI Technical Summary
The existing violation detection methods for call recording are relatively low in efficiency and accuracy, and manual inspection is not objective and comprehensive enough and the workload is large.
The recorded data is converted into text data through voice conversion technology, keyword matching is performed and pre-trained violation detection model is used to detect, and the violation level is judged.
It improves the efficiency and accuracy of call recording violation detection, and realizes unattended automated detection.
Smart Images
Figure CN120529016A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, applicable to medical and financial scenarios, and in particular to a call recording violation detection method, device, computer equipment and storage medium. Background Art
[0002] Customer service calls often involve sensitive information and important decisions, making it crucial to ensure that customer service personnel adhere to regulatory requirements when communicating with customers. To strengthen customer service personnel's communication standards, call recordings must be screened for violations. Currently, this is primarily done manually, but manual inspections are limited in objectivity and comprehensiveness, resulting in high error rates and a heavy workload. This results in low quality inspection efficiency and makes it difficult to accurately assess the severity of violations in call recordings. Summary of the Invention
[0003] The present invention provides a call recording violation detection method, device, computer equipment and storage medium to solve the technical problems of low efficiency and accuracy of existing call recording violation detection methods.
[0004] In a first aspect, a method for detecting a call recording violation is provided, the method comprising:
[0005] Get the call recording data to be tested;
[0006] Performing voice conversion processing on the call recording data to obtain text data of the call recording data;
[0007] Keyword matching is performed on the text data to determine whether the text data includes preset keywords. If the preset keywords are included, the violation level of the text data is determined to be a serious violation. If the preset keywords are not included, the text data is detected using a pre-trained violation detection model to obtain the violation level of the text data, which includes no violation, minor violation, and serious violation.
[0008] In a second aspect, a call recording violation detection device is provided, the call recording violation detection device comprising:
[0009] A voice data acquisition module is used to obtain the call recording data to be detected;
[0010] A text data acquisition module, configured to perform voice conversion processing on the call recording data to obtain text data of the call recording data;
[0011] The violation level determination module is used to perform keyword matching on the text data to determine whether the text data includes preset keywords. If the preset keywords are included, the violation level of the text data is determined to be a serious violation. If the preset keywords are not included, the text data is detected using a pre-trained violation detection model to obtain the violation level of the text data, which includes no violation, minor violation and serious violation.
[0012] 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. When the processor executes the computer program, the steps of the above-mentioned call recording violation detection method are implemented.
[0013] In a fourth aspect, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned call recording violation detection method are implemented.
[0014] In the solution implemented by the aforementioned call recording violation detection method, apparatus, computer device, and storage medium, call recording data is converted into text data after speech processing. This text data is then subjected to keyword matching to determine whether the text data contains preset keywords. If so, the text data's violation level is determined to be severe. If not, the text data is tested using a pre-trained violation detection model to determine the text data's violation level. Using speech conversion technology, keyword matching, and a violation detection model to detect violations in call recordings solves the problems of traditional manual detection methods, which suffer from insufficient objectivity and comprehensiveness, high error rates, and heavy workload, significantly improving the efficiency and accuracy of call recording violation detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0016] Figure 1 2 is a schematic diagram of an application environment of a call recording violation detection method provided by an embodiment of the present invention.
[0017] Figure 2 The figure is a flow chart of an implementation method of a call recording violation detection method provided by an embodiment of the present invention.
[0018] Figure 3 yes Figure 2Schematic diagram of a flow chart of a specific implementation of step S30.
[0019] Figure 4 yes Figure 2 FIG. 1 is a flow chart of another specific implementation of step S30.
[0020] Figure 5 yes Figure 4 A flowchart of a specific implementation method after step S35.
[0021] Figure 6 yes Figure 4 A flowchart of a specific implementation of step S33 is shown in FIG.
[0022] Figure 7 yes Figure 2 A flowchart of a specific implementation of step S20 is shown in FIG.
[0023] Figure 8 This is a flow chart of another implementation of a call recording violation detection method provided by an embodiment of the present invention.
[0024] Figure 9 This is a schematic diagram of a call recording violation detection device according to an embodiment of the present invention.
[0025] Figure 10 It is a structural diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION
[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.
[0027] The call recording violation detection method provided by the embodiment of the present invention can be applied to Figure 1 In the application environment, Figure 1 A schematic diagram of an application environment of a call recording violation detection method provided by an embodiment of the present invention is shown as follows: Figure 1As shown, in this application environment, a terminal 10 and a server 20 are included. The terminal 10 is a terminal device used by customer service, and the server 20 is a database terminal. The terminal 10 and the server 20 jointly execute the call recording violation detection method. It should be noted that the terminal 10 and the server 20 can be smart phones, tablet computers, laptop computers, desktop computers, etc., but are not limited to these. The terminal 10 and the server 20 can be connected via Bluetooth, USB (Universal Serial Bus) or other communication connection methods, and the present invention is not limited thereto.
[0028] Figure 2 A flowchart of an implementation method of a call recording violation detection method provided by an embodiment of the present invention is shown as follows: Figure 2 As shown, the call recording violation detection method may specifically include the following steps:
[0029] S10: Obtain call recording data to be detected.
[0030] In this step, the call recording data to be detected is the call recording data between the customer service and the customer. The call recording data between the customer service and the customer is collected and stored in a database, and the call recording data between the customer service and the customer is directly obtained from the database.
[0031] S20: Perform voice conversion on the call recording data to obtain text data of the call recording data.
[0032] In this step, the call recording data is subjected to speech conversion processing to obtain text data of the call recording data. Advanced automatic speech recognition (ASR) technology can be used to convert the call recording data into text data. Specifically, a deep learning model can be used for automatic speech recognition, such as a speech recognition model based on the Transformer architecture (for example, Wav2Vec 2.0). Transformer is superior to traditional models in modeling complex speech features. Wav2Vec 2.0 can reduce dependence on labeled data through self-supervised learning, while improving adaptability to noise and accents, making it more suitable for the practical application requirements of call recording data.
[0033] S30. Perform keyword matching on the text data to determine whether the text data includes preset keywords. If the text data includes the preset keywords, the violation level of the text data is determined to be a severe violation. If the text data does not include the preset keywords, a pre-trained violation detection model is used to detect the text data to obtain the violation level of the text data, where the violation level includes no violation, minor violation, and severe violation.
[0034] In this step, the text data is first keyword-matched. The keyword matching system is pre-established, and keywords include insulting and sensitive words, which are absolutely prohibited in customer service. Therefore, when such words are found in the text data during keyword matching, the violation level of the text data is directly determined to be a serious violation. If the keyword is not found in the data during keyword matching, the text data is tested using a pre-trained violation detection model to obtain the violation level of the text data. The violation detection model is a pre-trained classification model, and the violation levels obtained by the violation detection model include no violation, minor violation, and serious violation.
[0035] In a specific embodiment, in the medical field, keywords can be set as medical effect promises, privacy information, and related words that belittle patient choices. In the financial field, keywords can be set as related words such as belittlement of financial status, privacy leakage risks, misleading statements, and violation promises. In the medical field application scenario, keywords may include "cure all diseases", "absolutely safe", "zero risk", etc. In the financial field application scenario, keywords may include "100% compensation", "guaranteed profit", "guaranteed high returns" and "inside information", etc. In the above two application scenarios, since some words will mislead customers to make judgments, thereby causing adverse consequences and damaging the interests of customers and companies, they will be directly judged as serious violations when detected by the keyword matching system.
[0036] The call recording violation detection method of this embodiment performs violation detection on call recordings through voice conversion technology, keyword matching and violation detection models, solving the problems of traditional manual detection methods that are not objective and comprehensive, have high error rates and heavy workload, and greatly improves the efficiency and accuracy of call recording violation detection.
[0037] Figure 3 yes Figure 2 A flow chart of a specific implementation method in step S30 is shown in FIG. Figure 3 As shown, in some embodiments, the training steps of the violation detection model are as follows:
[0038] S31. Obtain historical call recording text data with marked categories to obtain a text training set, where the categories include a first category, a second category, and a third category, and the first category, the second category, and the third category correspond to labels of no violation, minor violation, and serious violation, respectively.
[0039] In this step, we first collect historical real call recordings of all customer service call scenarios, use speech-to-text technology to convert the historical real call recordings into historical call recording text, and then preprocess the historical call recording text to obtain the historical call recording text data. Preprocessing includes text cleaning, text segmentation, word form restoration, and annotation of the historical call recording text. Among them, text cleaning includes removing noise, invalid characters, repeated sentences, and garbled characters caused by background noise. When performing text segmentation, the historical call recording text is recorded as X = {x1, x2, ..., x n}, where x i is the i-th word, n is the length of the text, and the historical call recording text is split into subword sequences T = {t1, t2, ..., t m}, where m may not equal n. Labeling can be done manually by experts, or pre-labeled using a rule-based model followed by manual review. The preprocessed historical call recording text data is used as the text training set.
[0040] S32, inputting the text training set into the initial classification model, training the initial classification model based on the loss function and the optimizer to obtain a violation detection model; the loss function includes Where L is the loss value, N is the number of samples, C is the number of categories, and y ij is the one-hot encoding of the true label, The probability that the i-th sample belongs to the j-th category predicted by the violation detection model; the optimizer is Adamw, and the learning rate is 2e-5.
[0041] In this step, the initial classification model is a pre-trained BERT model with a classification layer added. The cross-entropy loss function is used when training the violation detection model. This loss function compares the difference between the predicted probability distribution and the true distribution, penalizing misclassifications. This allows for faster training of the violation detection model and naturally supports multi-class classification tasks, eliminating the need for manual probability distribution manipulation like the mean squared error (MSE). The optimizer uses Adamw, which automatically adjusts the learning rate of each parameter to accelerate model convergence. The loss function provides a clear optimization objective, while the optimizer efficiently navigates the parameter space. The combination of the two allows for better training of the violation detection model.
[0042] Figure 4 yes Figure 2 Another specific implementation flow diagram of step S30 is as follows: Figure 4 As shown, in some embodiments, in step S30, the text data is detected using a pre-trained violation detection model to obtain the violation level of the text data, specifically including:
[0043] S33 . Preprocess the text data to obtain preprocessed text data, and encode the preprocessed text data to obtain a corresponding feature vector.
[0044] In this step, the text data is preprocessed, including text cleaning, word segmentation, and lemmatization. Text cleaning involves removing noise, invalid characters, repeated sentences, and garbled text caused by background noise. Lemmatization involves restoring verbs to their original form to facilitate subsequent analysis. The preprocessed text data is encoded to obtain a corresponding feature vector.
[0045] S34. Perform linear combination on the feature vectors to obtain an original prediction score for each category.
[0046] In this step, the feature vector is multiplied by the weight matrix and the bias vector is added to obtain the original prediction score logits for each category. The calculation method is: logits = BERT(x)·W+b, where BERT(x) is the feature vector of the input text sequence x after being processed by the BERT model, W is the weight matrix, and b is the bias parameter of the fully connected layer.
[0047] S35. Convert the original prediction score into a probability value for each category through a normalized exponential function.
[0048] In this step, the normalized exponential function is Here, P(y=c|x) represents the probability that the model predicts the output to be category c, given input x. e is the base of the natural logarithm, which is used in the normalized exponential function to convert logits to exponential form, thereby amplifying the differences between scores. Logits are the raw prediction scores, C represents the total number of categories, and j is the category index. The normalized exponential function converts raw prediction scores into probabilities, facilitating classification decisions.
[0049] S37: Select the label corresponding to the category with the largest probability value as the violation level of the text data.
[0050] In this step, you can The label corresponding to the category with the largest probability value is selected as the violation level of the text data. "argmax" is the abbreviation of "argument of the maximum", which means the value of the independent variable that makes the function reach the maximum value.
[0051] Figure 5 yes Figure 4 A specific implementation flow chart after step S35 is shown in FIG. Figure 5 As shown, in some embodiments, after step S35, the method further includes:
[0052] S36. When the probability value of the second category is within the first preset interval, it is determined that the violation level of the text data is a minor violation; when the probability value of the second category is within the second preset interval, it is determined that the violation level of the text data is a serious violation.
[0053] In this embodiment, the label corresponding to the second category is a minor violation. When using the pre-trained violation detection model to detect the text data, if the probability of the text data being slightly violated is too high, it means that there are too many minor violations in this segment of text data, and it should be determined as a serious violation. Therefore, two preset intervals are set for the second category, and by judging the preset interval where the probability value of the second category is located, the violation level of the text data is determined. In a specific embodiment, the first preset interval is [0.3, 0.7], and the second preset interval is (0.7, 1]. After a certain segment of text data is input into the violation detection model and converted into a probability value through S35, the probability values corresponding to the first category, the second category, and the third category are 0.1, 0.8, and 0.1 respectively. At this time, since the probability 0.8 corresponding to the second category is within the second preset interval (0.7, 1], it is directly determined that this segment of text data is a serious violation.
[0054] Figure 6 Yes Figure 4 A schematic flow diagram of a specific implementation manner of step S33 in Figure 6 As shown, step S33 includes:
[0055] S331. Clean the text data and perform text tokenization. The text cleaning includes filtering out modal particles and stop words, and the text tokenization includes splitting the text data into a sub-word sequence.
[0056] In this embodiment, the text cleaning includes filtering out modal particles and stop words, removing stop words that are useless for analysis, such as "ah", "oh", "de", "le", etc. When performing text tokenization, the text data is split into a sub-word sequence T = {t1, t2,..., t m}, which is convenient for the violation detection model to identify the text data. Word form reduction includes reducing verbs to their original forms, which is convenient for the violation detection model to analyze.
[0057] Figure 7 Yes Figure 2 A schematic flow diagram of a specific implementation manner of step S20 in Figure 7 As shown, in some embodiments, step S20 may include:
[0058] S21. Extract language-related features from the call recording data through the ASR engine to obtain feature text; use the ASR engine to match the feature text with information in the built-in voice library to identify the voice content; convert the identified voice content into text form to form text data.
[0059] In this embodiment, the ASR (Automatic Speech Recognition) engine is a technical system that can convert speech signals into text. It recognizes the language content in speech by analyzing and processing the acoustic features of speech. The ASR engine first pre-processes the input call recording data, including operations such as noise removal and speech signal enhancement. Then, it extracts the acoustic features of the speech, such as Mel-Frequency Cepstral Coefficients (MFCC) and Linear Predictive Coding (LPC). These features can capture information such as the frequency, amplitude, and duration of the speech, and are the key basis for subsequent speech recognition. After feature extraction, the ASR engine will express the extracted features in some form to form feature text. This feature text is the basis for subsequent matching with the speech library.
[0060] ASR systems typically have a built-in speech library containing a large number of speech samples and their corresponding text annotations. These speech samples cover a wide range of pronunciations, accents, speaking speeds, and other conditions and are used to train and optimize the ASR model. The ASR engine compares and matches the extracted feature text with the information in the speech library. By calculating the similarity between features, it finds the speech sample that most closely matches the input features and determines the corresponding text content. Once matching is complete, the ASR engine converts the recognized speech content into its corresponding text form. This process extracts the matching text annotations from the speech library to form the final recognition result, namely the text data.
[0061] Figure 8 FIG. 1 is a flow chart of another implementation of a method for detecting call recording violations according to an embodiment of the present invention. Figure 8 As shown, in some embodiments, after step S30, the method further includes:
[0062] S40: When the violation level is no violation, do not process the call recording data;
[0063] When the violation level is a minor violation, generating a first warning message based on the call recording data, and sending the first warning message to a first responsible person corresponding to the call recording data;
[0064] When the violation level is a serious violation, a second warning message is generated according to the call recording data, and the second warning message is sent to a second responsible person corresponding to the call recording data.
[0065] In this step, different handling plans are generated for different violation levels. If the violation level of the call recording data is determined to be "No Violation," this means that the call content complies with regulations and no violations have occurred. Therefore, no action is required on the call recording data. If the violation level is determined to be "Minor Violation," a first warning message is generated based on the call recording data. This warning message typically includes relevant information such as the specific content, time, and location of the violation. This warning message is then sent to the corresponding primary responsible person, allowing them to understand the minor violation and take appropriate corrective measures. The primary responsible person includes those directly involved in the call, such as customer service personnel or sales staff. If the violation level is "Severe Violation," the system generates a second warning message. This second warning message not only includes relevant information such as the specific content, time, and location of the violation, but also includes an explanation of the consequences and corrective action requirements. This second warning message is sent to the corresponding secondary responsible person, typically someone with higher management authority, such as a department manager or supervisor. The secondary responsible person can conduct an in-depth investigation into serious violations and take strict corrective and disciplinary measures.
[0066] The call recording violation detection method provided by the embodiment of the present invention can be constructed based on artificial intelligence, and the relevant data can be acquired and processed based on artificial intelligence technology to realize unattended call recording violation detection. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0067] In one embodiment, a call recording violation detection device is provided, which corresponds to the call recording violation detection method in the above embodiment. Figure 9 As shown, the call recording violation detection device includes:
[0068] The voice data acquisition module 100 is used to obtain the call recording data to be detected;
[0069] The text data acquisition module 200 is used to perform voice conversion processing on the call recording data to obtain text data of the call recording data;
[0070] The violation level determination module 300 is used to perform keyword matching on the text data to determine whether the text data includes preset keywords. If the preset keywords are included, the violation level of the text data is determined to be a serious violation. If the preset keywords are not included, the text data is detected using a pre-trained violation detection model to obtain the violation level of the text data, which includes no violation, minor violation, and serious violation.
[0071] Specifically, the violation level determination module 300 includes a first model training unit and a second model training unit.
[0072] The first model training unit is used to obtain historical call recording text data with labeled categories to obtain a text training set, where the categories include a first category, a second category, and a third category, and the first category, the second category, and the third category correspond to labels of no violation, minor violation, and serious violation, respectively.
[0073] The second model training unit is used to input the text training set into the initial classification model, train the initial classification model based on the loss function and the optimizer, and obtain a violation detection model; the loss function includes Where L is the loss value, N is the number of samples, C is the number of categories, and y ij is the one-hot encoding of the true label, The probability that the i-th sample belongs to the j-th category predicted by the violation detection model; the optimizer is Adamw, and the learning rate is 2e-5.
[0074] Specifically, the violation level determination module 300 further includes a text processing unit, a score acquisition unit, a probability acquisition unit, and a first violation level determination unit.
[0075] The text processing unit is used to preprocess the text data to obtain preprocessed text data, and encode the preprocessed text data to obtain a corresponding feature vector.
[0076] The score acquisition unit is used to perform linear combination on the feature vectors to obtain an original prediction score for each category.
[0077] The probability acquisition unit is used to convert the original prediction score into a probability value for each category through a normalized exponential function.
[0078] The first violation level determination unit is configured to select a label corresponding to a category with a maximum probability value as the violation level of the text data.
[0079] Specifically, the violation level determination module 300 further includes a second violation level determination unit.
[0080] The second violation level determination unit is configured to determine that the violation level of the text data is a minor violation when the probability value of the second category is within a first preset interval; and to determine that the violation level of the text data is a serious violation when the probability value of the second category is within a second preset interval.
[0081] Specifically, the text processing unit is further used to perform text cleaning and text segmentation on the text data, wherein the text cleaning includes filtering modal particles and stop words, and the text segmentation includes splitting the text data into subword sequences.
[0082] Specifically, the text data acquisition module 200 includes a speech recognition unit.
[0083] The speech recognition unit is used to extract language-related features from call recording data through the ASR engine to obtain feature text; the ASR engine matches the feature text with the information in the built-in speech library to identify the speech content; and convert the recognized speech content into text form to form text data.
[0084] Specifically, the device also includes a violation processing module.
[0085] a violation processing module, configured to not process the call recording data when the violation level is no violation;
[0086] When the violation level is a minor violation, generating a first warning message based on the call recording data, and sending the first warning message to a first responsible person corresponding to the call recording data;
[0087] When the violation level is a serious violation, a second warning message is generated according to the call recording data, and the second warning message is sent to a second responsible person corresponding to the call recording data.
[0088] The specific limitations of the call recording violation detection device can be found in the limitations of the call recording violation detection method described above and will not be further elaborated here. Each module in the aforementioned call recording violation detection device may be implemented in whole or in part through software, hardware, or a combination thereof. Each of these modules may be embedded in or independent of a processor in a computer device in hardware form, or may be stored in a computer device memory in software form, allowing the processor to call and execute the corresponding operations of each module.
[0089] Figure 10 FIG. 1 is a schematic structural diagram of an embodiment of a computer device according to the present application, as shown in FIG. Figure 10 As shown, the present application also provides a computer device, including:
[0090] The memory and the processor store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes any step of the above-mentioned call recording violation detection method.
[0091] The present application also provides a computer-readable storage medium. When the computer-readable instructions are executed by one or more processors, the one or more processors execute any step in the call recording violation detection method. It can be understood that the readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.
[0092] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.
Claims
1. A method for detecting call recording violations, characterized in that: include: Obtain the call recording data to be tested; Performing voice conversion processing on the call recording data to obtain text data of the call recording data; Keyword matching is performed on the text data to determine whether the text data includes preset keywords. If the preset keywords are included, the violation level of the text data is determined to be a serious violation. If the preset keywords are not included, the text data is detected using a pre-trained violation detection model to obtain the violation level of the text data, which includes no violation, minor violation, and serious violation.
2. The call recording violation detection method according to claim 1, wherein: The training method of the violation detection model includes: Obtaining historical call recording text data labeled with categories to obtain a text training set, wherein the categories include a first category, a second category, and a third category, where the first category, the second category, and the third category correspond to labels of no violation, minor violation, and serious violation, respectively; The text training set is input into the initial classification model, and the initial classification model is trained based on the loss function and the optimizer to obtain a violation detection model; the loss function includes Where L is the loss value, N is the number of samples, C is the number of categories, and y ij is the one-hot encoding of the true label, The probability that the i-th sample belongs to the j-th category predicted by the violation detection model; the optimizer is Adamw, and the learning rate is 2e-5.
3. The call recording violation detection method according to claim 2, wherein: The detecting the text data using a pre-trained violation detection model to obtain a violation level of the text data includes: Preprocessing the text data to obtain preprocessed text data, encoding the preprocessed text data to obtain a corresponding feature vector; Performing a linear combination on the feature vectors to obtain the original prediction score for each category; Converting the raw prediction scores into probability values for each category through a normalized exponential function; The label corresponding to the category with the largest probability value is selected as the violation level of the text data.
4. The call recording violation detection method according to claim 3, wherein: After converting the original prediction score into a probability value for each category by a normalized exponential function, the method further includes: When the probability value of the second category is in the first preset interval, the violation level of the text data is determined to be a minor violation; when the probability value of the second category is in the second preset interval, the violation level of the text data is determined to be a serious violation.
5. The call recording violation detection method according to claim 3, wherein: The preprocessing of the text data includes: The text data is cleaned and segmented, wherein the text cleaning includes filtering modal particles and stop words, and the text segmentation includes splitting the text data into subword sequences.
6. The call recording violation detection method according to claim 1, wherein: The performing voice conversion processing on the call recording data to obtain text data of the call recording data specifically includes: The ASR engine extracts language-related features from call recording data to obtain feature text; the ASR engine matches the feature text with the information in the built-in voice library to identify the voice content; and the identified voice content is converted into text form to form text data.
7. The call recording violation detection method according to claim 1, wherein: After obtaining the violation level of the text data, the method further includes: When the violation level is no violation, the call recording data is not processed; When the violation level is a minor violation, generating a first warning message based on the call recording data, and sending the first warning message to a first responsible person corresponding to the call recording data; When the violation level is a serious violation, a second warning message is generated according to the call recording data, and the second warning message is sent to a second responsible person corresponding to the call recording data.
8. A call recording violation detection device, characterized in that: include: A voice data acquisition module is used to obtain the call recording data to be detected; A text data acquisition module, configured to perform voice conversion processing on the call recording data to obtain text data of the call recording data; The violation level determination module is used to perform keyword matching on the text data to determine whether the text data includes preset keywords. If the preset keywords are included, the violation level of the text data is determined to be a serious violation. If the preset keywords are not included, the text data is detected using a pre-trained violation detection model to obtain the violation level of the text data, which includes no violation, minor violation and serious violation.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the call recording violation detection method according to 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 call recording violation detection method according to any one of claims 1 to 7 are implemented.