Artificial intelligence-based verbal skill quality inspection method, apparatus and device, and storage medium

Abstract

The invention belongs to the field of artificial intelligence and big data, and relates to an artificial intelligence-based verbal skill quality inspection method, which comprises the following steps of: obtaining session information, and converting the session information into to-be-identified text information; based on a pre-trained sentiment analysis model, analyzing whether the to-be-recognized text information has a sentiment guiding behavior or not; if the to-be-identified text information does not exist, performing illegal behavior identification on the to-be-identified text information based on an illegal rule base and a pre-trained quality inspection model, and judging whether an illegal behavior exists or not; and if the emotion guiding behavior or the illegal behavior exists, sending out an illegal behavior warning. The invention further provides a verbal skill quality inspection device and equipment based on artificial intelligence and a storage medium. In addition, the invention further relates to a block chain technology, and accessed data can be stored in a block chain. The method can be applied to a verbal skill quality inspection scene in the financial field, multi-aspect verbal skill violation recognition is performed by utilizing artificial intelligence in combination with multiple models, and the recognition efficiency and accuracy of violation verbal skill in a business scene can be improved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and big data technology, and in particular to a speech quality inspection method, device, equipment and storage medium based on artificial intelligence. Background Art

[0002] Artificial intelligence and big data technologies have developed rapidly in recent years. Artificial intelligence simulates human intelligence through technologies such as machine learning and deep learning to achieve automated decision-making and intelligent services; big data mines the value of data through massive data processing and analysis to provide a scientific basis for decision-making. The combination of the two is profoundly changing the production methods and business models of all walks of life, especially in the financial field, where they are widely used in insurance, banking, etc., bringing about tremendous changes in the financial industry.

[0003] Bank or insurance salespersons usually communicate with customers through corporate WeChat, and the communication methods include voice, text, pictures, etc. In this process, in order to achieve better performance, salespersons will make some illegal speeches, provide false or inaccurate information, conceal important information, etc., such as exaggerating product features, unconfirmed return promises, hiding costs or risks, concealing contract terms, cost structure or possible risks, so that customers can make informed decisions. These behaviors may involve violations. The daily communication data is massive, and the traditional quality inspection method is highly dependent on manual labor, but manual quality inspection has many problems such as low efficiency and poor timeliness. The prior art also proposes to identify speech based on keyword matching, but it needs to rely on manual experience to summarize keywords. Incomplete keyword summary will still cause illegal speech to be omitted. In addition, it is easy to ignore semantic information and cause misjudgment. In addition, the prior art also proposes training and identification based on natural language models, but these models have the problems of too large parameters and slow training and reasoning speed, and are not applicable in the application scenarios of illegal speech identification. It can be seen that the prior art has the problems of low efficiency and accuracy when identifying illegal speech in industries such as banking and insurance. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose a speech quality inspection method, device, equipment and storage medium based on artificial intelligence to solve the problem of low efficiency and accuracy in the existing technology when identifying illegal speech in industries such as banking and insurance.

[0005] In order to solve the above technical problems, the present application embodiment provides a speech quality inspection method based on artificial intelligence, which adopts the following technical solutions:

[0006] Acquiring conversation information, and performing text conversion on the conversation information to obtain text information to be recognized, wherein the conversation information includes at least one type of speech information;

[0007] Based on the pre-trained sentiment analysis model, the text information to be identified is subjected to sentiment analysis to determine whether there is any sentiment-guiding behavior;

[0008] If the emotion-guiding behavior does not exist, then based on the violation rule library and the pre-trained quality inspection model, the text information to be identified is subjected to violation identification to determine whether there is a violation, wherein the pre-trained quality inspection model is trained based on a historical violation data set;

[0009] If the aforementioned emotionally-guided behavior or the aforementioned illegal behavior exists, a violation warning will be issued.

[0010] Furthermore, the conversation information includes voice information, and the step of acquiring the conversation information and converting the conversation information into text to obtain the text information to be recognized includes:

[0011] Acquire the session information, and determine whether the session information includes the voice information;

[0012] If the conversation information includes the voice information, performing audio preprocessing on the voice information to obtain preprocessed audio information;

[0013] The pre-processed audio information is subjected to audio feature extraction based on a speech recognition model, and the audio feature is subjected to text conversion through the speech recognition model to obtain the text information to be recognized.

[0014] Furthermore, the session information also includes picture information, and the step of obtaining the session information and converting the session information into text to obtain the text information to be identified includes:

[0015] Acquire the session information, and determine whether the session information includes the picture information;

[0016] If the session information includes the picture information, detecting whether there is picture modification information in the picture information based on a computer vision model, wherein the picture modification information includes text;

[0017] If the image change information exists, the image information is preprocessed, and the image change information in the preprocessed image information is located and the text is segmented to obtain the text to be extracted;

[0018] The text features of the text to be extracted are extracted by an optical character recognition algorithm to obtain the text information to be recognized.

[0019] Furthermore, the session information also includes text information, and the step of obtaining the session information and converting the session information into text to obtain the text information to be identified includes:

[0020] Acquire the session information, and determine whether the session information includes the text information;

[0021] If the session information includes the text information, the text information is preprocessed to obtain the text information to be identified.

[0022] Furthermore, the step of performing sentiment analysis on the text information to be identified based on the pre-trained sentiment analysis model to determine whether there is any sentiment-guiding behavior includes:

[0023] The text information to be identified is identified by the pre-trained sentiment analysis model to extract key sentiment features, wherein the pre-trained sentiment analysis model is trained based on a data set with marked sentiment categories;

[0024] Determining whether there is a preset limited emotion category based on the key emotion features;

[0025] If the preset limited emotion category exists, it is determined that the emotion guiding behavior exists.

[0026] Furthermore, the step of identifying the violation behavior of the to-be-identified text information based on the violation rule library and the pre-trained quality inspection model and determining whether there is any violation behavior includes:

[0027] Acquire the service type corresponding to the text information to be identified, where different service types correspond to different violation rules in the violation rule library;

[0028] Reading a target violation rule corresponding to the service type of the text information to be identified in the violation rule library through a rule engine, detecting the text information to be identified based on the target violation rule to determine whether it complies with the target violation rule;

[0029] If the target violation rule is met, then the text information to be identified is judged to have the violation behavior;

[0030] If the target violation rule is not met, it is determined that the text information to be identified does not contain the violation, and the violation is identified on the text information to be identified through the pre-trained quality inspection model to determine whether the violation exists.

[0031] Furthermore, the step of identifying the illegal behavior of the text information to be identified by the pre-trained quality inspection model and determining whether the illegal behavior exists includes:

[0032] The text information to be identified is converted into a text vector by using the pre-trained quality inspection model, wherein the pre-trained quality inspection model is modeled based on a deep learning algorithm;

[0033] Extracting text feature vectors from the text vectors based on a deep learning algorithm;

[0034] Extracting text keywords based on the text content and text structure of the text information to be identified;

[0035] Classify based on the text feature vector and the text keywords, and determine whether there is any violation according to the classification result, wherein the classification result includes violation and non-violation;

[0036] If it is determined that the violation exists, the text information to be identified is marked with a violation label.

[0037] In order to solve the above technical problems, the embodiment of the present application further provides a speech quality inspection device based on artificial intelligence, the device comprising:

[0038] A conversion module, used to obtain session information, perform text conversion on the session information, and obtain text information to be recognized, wherein the session information includes at least one type of speech information;

[0039] An analysis module is used to perform sentiment analysis on the text information to be identified based on a pre-trained sentiment analysis model to determine whether there is any sentiment-guiding behavior;

[0040] An identification module, for identifying illegal behaviors on the text information to be identified based on the illegal rule library and the pre-trained quality inspection model to determine whether there are illegal behaviors if the emotional guidance behaviors do not exist, wherein the pre-trained quality inspection model is trained based on the historical illegal behavior data set;

[0041] The sending module is used to issue a violation warning if the emotion-guiding behavior or the violation exists.

[0042] In order to solve the above technical problems, an embodiment of the present application also provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the artificial intelligence-based speech quality inspection method are implemented.

[0043] In order to solve the above technical problems, an embodiment of the present application also provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the steps of the artificial intelligence-based speech quality inspection method are implemented.

[0044] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0045] The conversation information obtained by the present application includes multiple types. Through text conversion, recognition is performed based on text, which can improve the efficiency of information processing, and the automated text conversion process reduces the need for manual input and proofreading, thereby improving accuracy. After the information is uniformly converted into a text format, it is more convenient to store and manage, avoiding the incompatibility problem between information in different formats. In combination with a pre-trained sentiment analysis model, sentiment analysis is performed on the text information to be recognized, which can accurately and quickly determine from an emotional perspective whether the salesperson has emotional guidance behavior, which is more conducive to improving the accuracy and efficiency of speech quality inspection. In the absence of emotional guidance behavior, in combination with a violation rule library and a pre-trained quality inspection model, violation identification is continued. Based on the violation rule library, speech that violates some defined rules can be quickly and accurately identified, which is conducive to quickly warning the salesperson. Based on the pre-trained quality inspection model, identification can continue to analyze violations when the violations cannot be accurately identified through the defined violation rules, which is more conducive to improving the accuracy of violation identification, and timely warning the salesperson, ensuring that the salesperson provides real information to the customer and helping the customer make wise decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0048] Figure 2 It is a flowchart of an embodiment of a speech quality inspection method based on artificial intelligence according to the present application;

[0049] Figure 3 It is a flowchart of another embodiment of the speech quality inspection method based on artificial intelligence of the present application;

[0050] Figure 4 It is a structural schematic diagram of an embodiment of a speech quality inspection device based on artificial intelligence according to the present application;

[0051] Figure 5 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0053] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0054] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0055] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0056] Clients can use terminal devices 101, 102, 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0057] Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, desktop computers, etc.

[0058] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0059] It should be noted that the artificial intelligence-based speech quality inspection method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the artificial intelligence-based speech quality inspection device is generally set in the server / terminal device.

[0060] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0061] Continue to refer Figure 2 , shows a flow chart of an embodiment of the speech quality inspection method based on artificial intelligence according to the present application. According to different requirements, the order of the steps in the flow chart can be changed. The speech quality inspection method based on artificial intelligence provided by the embodiment of the present application can be applied to any scenario where speech quality inspection is required. The speech quality inspection method based on artificial intelligence includes the following steps:

[0062] Step S201: Acquire session information, perform text conversion on the session information, and obtain text information to be recognized, wherein the session information includes at least one type of speech information.

[0063] In this embodiment, the electronic device on which the speech quality inspection method based on artificial intelligence runs can perform data access and data transmission during the speech quality inspection process through a wired connection or a wireless connection. It should be noted that the above wireless connection method may include but is not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wide band) connection, and other wireless connection methods currently known or to be developed in the future.

[0064] Specifically, when this application is used in the banking and insurance industries, salesmen can communicate with customers through WeChat for Business. During the communication process, both salesmen and customers can send voice, text, pictures, videos and other speech content. In order to more completely analyze the conversation information, the acquired conversation information can be uniformly converted into text form through text conversion to obtain the text information to be identified, and unified analysis and identification are performed in text form to improve information processing efficiency and data management efficiency. Among them, due to the different service types of the acquired conversation information, the corresponding text conversion method is also different.

[0065] Furthermore, the system used by salesmen to communicate with customers has a conversation archiving function, which allows enterprises to save chat records between salesmen and customers. When the conversation archiving function is turned on, the chat records between salesmen and customers can be obtained in real time based on the conversation archiving function, that is, the conversation information between salesmen and customers can be obtained in real time, including obtaining text, pictures, voice, video and other forms of speech information. Of course, the conversation information between salesmen and customers can also be obtained by querying historical chat records.

[0066] Step S202: Perform sentiment analysis on the text information to be identified based on a pre-trained sentiment analysis model to determine whether there is any emotion-guiding behavior.

[0067] The pre-trained sentiment analysis model can be trained based on a dataset with labeled sentiment categories that is used by salespeople in banking or insurance industries in customer marketing and service scenarios. By performing sentiment analysis on the text information to be identified using the pre-trained sentiment analysis model, it is possible to determine whether the salesperson has any emotional guidance behavior towards the customer from the perspective of sentiment analysis.

[0068] In some examples, the pre-trained sentiment analysis model can be constructed based on a sentiment dictionary, and the sentiment tendency of the text can be determined by matching sentiment words in the text information to be identified based on the sentiment dictionary. The sentiment dictionary can include sentiment word categories such as positive, negative, and neutral, and the sentiment intensity corresponding to different words.

[0069] In other examples, the above-mentioned pre-trained sentiment analysis model can be constructed based on a machine learning algorithm, and the sentiment analysis model is constructed by training a large amount of text data with labeled sentiment categories. Among them, the machine learning algorithm may include Naive Bayes, Support Vector Machine, Random Forest, etc. The sentiment analysis model constructed according to the machine learning algorithm can automatically learn the sentiment features in the text, extract keywords, phrases or patterns and other information according to semantics for analysis, and can perform sentiment classification on the words issued by salesmen to customers in industries such as banking or insurance. Through the judgment of sentiment categories, it can be determined whether the salesmen have emotional guidance behavior towards customers. For example: extract the product characteristics of marketing products, the activity arrangements of marketing products, the vocabulary used by salesmen to express emotions, etc., and detect whether the salesmen use overly positive or negative emotional expressions. The vocabulary used to express emotions includes but is not limited to thank you, sorry, good / bad attitude, slow / fast speed, etc.

[0070] In other examples, the above-mentioned pre-trained sentiment analysis model can also be constructed based on a deep learning algorithm, wherein the deep learning algorithm can include a convolutional neural network, a recursive neural network, a long short-term memory network, etc. The data set with labeled sentiment categories is divided into a training set and a test set, and the sentiment analysis model is trained and evaluated based on the deep learning algorithm to obtain the final sentiment analysis model. Training the sentiment analysis model based on the deep learning algorithm can make full use of the contextual information of the text in the process of communication between the salesperson and the customer, actively learn the characteristics of texts containing different emotions, ensure that the analysis results can make judgments based on the overall sentiment, and have a higher recognition accuracy.

[0071] In other examples, multiple algorithms can be integrated to build sentiment analysis models to improve the overall performance. For example, a sentiment dictionary-based method is combined with a machine learning or deep learning-based method to train a sentiment analysis model, thereby improving the overall analysis performance of the sentiment analysis model.

[0072] Step S203: If the emotion-guiding behavior does not exist, the text information to be identified is subjected to violation identification based on the violation rule library and the pre-trained quality inspection model to determine whether there is any violation, wherein the pre-trained quality inspection model is trained based on a historical violation data set.

[0073] In this embodiment, if there is no emotional guidance behavior, there may be violations in other aspects. In this regard, the rules in the violation rule library can be extracted and matched with the text information to be identified to determine whether there is an illegal speech. If there is an illegal speech, it is determined that there is an illegal behavior. Considering that there are many kinds of illegal speech, in addition to the illegal speech that can be determined by matching rules, more cannot be determined by exact rules. In this regard, in this embodiment, according to the historical violation data set, the violation can be identified by training a supervised learning model (pre-trained quality inspection model), and the supervised learning model can be used to classify the behavior, so as to determine whether there is a violation. Among them, the historical violation data set is a data set of known violation cases, including some new potential violations, such as: privately promising customers, frequently sending advertising and marketing information, leaking customer privacy, etc. Among them, the pre-trained quality inspection model can be a model trained based on a convolutional neural network, a recurrent neural network or a Transformer, etc., which can perform in-depth analysis on the input text information to be identified through steps such as semantic understanding, sentiment analysis and / or keyword extraction, so as to realize the identification of dialogue violations.

[0074] In some examples, the above-mentioned violation rule library can be built for illegal speech that meets a series of rules and conditions, allowing administrators to define and manage a set of rules for illegal behaviors, such as: specific words or sentences that exaggerate product features, prohibited speech, etc. The violation rule library has flexible rule management and configuration functions. Administrators can add, edit or delete rules as needed without modifying the underlying application code. The violation rule library has high real-time performance and can quickly make judgments on illegal behaviors. For illegal speech that meets the rules in the violation rule library, the salesperson can be warned immediately.

[0075] Step S204: If the emotion-guiding behavior or the illegal behavior exists, a violation warning is issued.

[0076] In some embodiments, template matching can be performed based on the business service type between the salesperson and the customer. Different business service types correspond to different warning templates. For example, in insurance services, the business service types of salespersons include insurance knowledge popularization, insurance product interpretation, insurance plan design, insurance, renewal, claims, loans, etc.; in banking services, the business service types of salespersons include account management, deposits and withdrawals, loans, transfers and remittances, investment and financial management, risk assessment, etc. The system can store business warning templates of various business service types, which can be extracted from them according to the matched templates, and the templates can be filled in according to the violations of the bank or insurance salesperson, and then the filled templates can be sent to the salesperson and / or other managers. The templates may include business service time, salesperson personal information, customer name, violation type, violation content, violation level, punishment, etc. Among them, the notification method may include in-site information, email, SMS, group notification, etc.

[0077] Furthermore, by matching templates according to business service types to fill in warning information and sending them to corresponding personnel, the corresponding alarm information can be generated quickly and conveniently, and the template matching method can be better managed in a unified manner, and the warning content is more comprehensive, which is convenient for obtaining warning content in a concise and clear manner, thereby improving the intelligence of the alarm method.

[0078] In some embodiments, the salesperson's leader or other reviewer is allowed to serve as a monitor for speech quality inspection, and the monitor can spot check and mark potential violations, and regularly update the algorithms involved in the speech quality inspection process based on the marked data to adapt to new violations and changing market conditions.

[0079] In an embodiment of the present invention, the conversation information obtained by the present application includes multiple types, and is identified based on text through text conversion, which can improve the efficiency of information processing, and the automated text conversion process reduces the need for manual input and proofreading, and improves accuracy. After the information is uniformly converted into a text format, it is more convenient to store and manage, and the incompatibility problem between information in different formats is avoided. In combination with the pre-trained sentiment analysis model, sentiment analysis can be performed on the text information to be identified, and it can accurately and quickly judge whether the salesperson has emotional guidance behavior from an emotional perspective, which is more conducive to improving the accuracy and efficiency of speech quality inspection. In the absence of emotional guidance behavior, the violation rule library and the pre-trained quality inspection model are combined to continue to identify violations. Based on the violation rule library, the speech that violates some defined rules can be quickly and accurately identified, and it can be timely determined whether to warn the salesperson, reducing the possibility of adverse effects on customers. Based on the pre-trained quality inspection model for identification, the violation behavior analysis can be continued when the violation behavior cannot be accurately identified by the defined violation rules, which is more conducive to improving the accuracy of violation identification, and timely warning the salesperson, ensuring that the salesperson provides real information to the customer and helping the customer make wise decisions. Therefore, through text conversion, sentiment analysis, violation rule identification and multiple comprehensive identification of quality inspection models are carried out on the converted text information to be identified, which can improve the accuracy and efficiency of speech quality inspection and identification.

[0080] In some optional embodiments, combined with Figure 3 As shown, the above step S201 includes:

[0081] Acquire the session information, and determine whether the session information includes the voice information;

[0082] If the conversation information includes the voice information, performing audio preprocessing on the voice information to obtain preprocessed audio information;

[0083] The pre-processed audio information is subjected to audio feature extraction based on a speech recognition model, and the audio feature is subjected to text conversion through the speech recognition model to obtain the text information to be recognized.

[0084] Specifically, the above-mentioned voice information includes audio information and video information. When the conversation information is voice information, the voice can be preprocessed first to obtain preprocessed audio information. Among them, preprocessing includes voice noise reduction, noise removal, echo cancellation and other operations, and voice processing can be achieved by combining filters, spectral subtraction and acoustic models, which helps to reduce the interference of background noise and other non-voice signals, thereby improving the accuracy of voice recognition.

[0085] Furthermore, after preprocessing, audio feature information can be extracted from the preprocessed audio information for use in the speech recognition model. The extracted audio feature information may include Mel frequency cepstral coefficients, linear prediction coefficients, pitch periodicity, etc. These audio feature information can effectively represent the key information of the speech signal.

[0086] Among them, the above-mentioned speech recognition model can be trained based on model matching method, probability statistics method or discriminator classification method, among which the model matching method includes vector quantization, dynamic time warping, etc.; the probability statistics method includes Gaussian mixture model, hidden Markov model, etc.; the discriminator classification method includes support vector machine, artificial neural network and deep neural network, etc. and various combination methods. The speech recognition model can use speech recognition technology to recognize the extracted audio features and convert the speech into text to obtain the text information to be recognized. For example: a bank clerk handles loan procedures for customers through voice. After the extracted voice is pre-processed, the audio features are extracted, and then the audio features of the clerk are converted into text information of the loan procedures. After conversion to text, in order to ensure the coherence of the sentence and the rationality of grammar and semantics, the output text can also be subjected to post-processing operations such as grammar correction and punctuation processing to make it more in line with the expression habits of human language.

[0087] In this embodiment, converting speech into text facilitates direct and rapid browsing and analysis of conversation content, thereby efficiently identifying whether the salesperson has used illegal language; and being able to accurately convert speech into text reduces misjudgments caused by human transcription errors, which helps to ensure the accuracy and reliability of recognition results and reduce the risks and costs caused by misjudgments; unified conversion to text format is more convenient for storage and management, avoiding incompatibility issues between information in different formats.

[0088] In some optional embodiments, combined with Figure 3 As shown, the above step S201 further includes:

[0089] Acquire the session information, and determine whether the session information includes the picture information;

[0090] If the session information includes the picture information, detecting whether there is picture modification information in the picture information based on a computer vision model, wherein the picture modification information includes text;

[0091] If the image change information exists, the image information is preprocessed, and the image change information in the preprocessed image information is located and the text is segmented to obtain the text to be extracted;

[0092] The text features of the text to be extracted are extracted by an optical character recognition algorithm to obtain the text information to be recognized.

[0093] In this embodiment, some illegal information may be added to the picture and sent in the form of a picture. Therefore, the session information may also include picture information. When the picture information is identified, computer vision (CV) technology may be used to detect whether the picture has been modified. For example: detect whether there are traces of image editing in the picture, including picture cutting, synthesis or filters. When there is a picture modification, the picture modification information may be located and the text segmentation may be performed after the picture preprocessing, so as to find the text to be extracted from the picture, and further the text to be extracted may be extracted by optical character recognition (OCR) technology to obtain the text information to be identified.

[0094] Furthermore, the above-mentioned judgment of whether there is image modification information can be analyzed based on the image content, including through cloning detection, using tools such as Forensically, by identifying repeated pixels in the image to determine whether the image has been modified by the pixel cloning tool. When a large area of ​​adjacent pixels are identified to be exactly the same, it can be determined that there are cloning traces, and it is considered that there is image modification in the image. Error level analysis can also be used to highlight pixels that look similar to the naked eye but are very abrupt from a computer perspective, which helps to find tampered or added content, such as added objects or people, text, etc. Copy-move detection can also be used to use computer vision algorithms to find similar image content areas. When large blocks of identical content are detected in the same image, it may be determined that the image has been tampered with. Pixel resampling detection can also be used to analyze the correlation between the interpolated gap position and the original pixels around it, and it can be determined whether the image has been scaled, rotated, and other operations. Judgment can also be made based on machine learning and deep learning, such as image tampering detection using dual-stream Faster-RCNN. Considering that there are various ways to modify images, a single method may not be able to make a completely accurate judgment. Therefore, multiple methods can be combined for comprehensive judgment, and the final result can be obtained by combining the judgment results of multiple methods.

[0095] In some examples, when image recognition is performed in multiple ways, each recognition method can be assigned a weight and score, and the final output value can be calculated by combining the weight of each recognition method and the score corresponding to the recognition result, and whether the image information contains image modification information can be determined based on whether the final output value meets the threshold. This can improve the accuracy and reliability of detection.

[0096] Furthermore, the above-mentioned image preprocessing includes steps such as image enhancement, binarization, denoising, and tilt correction. After preprocessing, the image information can be clearer, less disturbed, and more accurately expressed. In the text positioning and segmentation of the image change information, the text positioning can locate the text area of ​​the changed part of the image through edge detection, connected domain analysis and other methods; the text segmentation can further divide the located text area into single characters or words to facilitate subsequent character recognition.

[0097] Furthermore, text features can be extracted for the locked text to be extracted, and the extracted text features can be based on feature extraction of text structure or feature extraction based on statistics. Among them, feature extraction based on structure can be to extract text features by analyzing structural features of the text to be extracted, such as strokes, radicals, etc. Feature extraction based on statistics can be to extract text features by counting features such as pixel distribution and grayscale value of the text to be extracted. The extracted text features can be matched with pre-stored character templates through OCR technology, and the recognized text can be post-processed to improve the recognition accuracy, including error correction, typesetting and output, and finally output the text information to be recognized.

[0098] In this embodiment, by identifying the image information, it is possible to detect whether the salesperson has made secondary changes to the original image. When there is a change, the text feature extraction is combined with OCR technology to facilitate the subsequent analysis of the information changed by the salesperson. Thus, text analysis can be performed based on the information changed by the salesperson to determine whether there are some illegal behaviors caused by adding illegal information to the image and sending it to the customer in the form of an image. This avoids misjudgment due to human inability to recognize image changes, and improves the accuracy and reliability of image recognition.

[0099] In some optional embodiments, combined with Figure 3 As shown, the above step S201 further includes:

[0100] Acquire the session information, and determine whether the session information includes the text information;

[0101] If the session information includes the text information, the text information is preprocessed to obtain the text information to be identified.

[0102] In this embodiment, the session information can also be text information. When the communication form between the bank or insurance salesman and the customer is text information, the text information can be directly preprocessed, including removing irrelevant characters from the text information, such as: 中, 的, 了, etc. These characters may not be necessary for text recognition and may interfere with the recognition result. It also includes word segmentation of the text, removing duplicate content, spelling correction, text standardization processing, removing noise, text cleaning, etc. Among them, word segmentation can split continuous text into independent words or phrases, and correct word segmentation helps subsequent semantic understanding and recognition; duplicate removal can avoid generating redundant information during the recognition process; spelling correction can correct spelling mistakes in the text to improve the recognition accuracy; text standardization can convert the text into a unified format, for example: when the text information sent by the insurance salesman includes English text, all characters in the English text are converted to lowercase or translated into Chinese. After text preprocessing, the text information to be recognized will be obtained.

[0103] In some examples, when the session information is picture information or audio information, after processing to extract text, the above-mentioned text preprocessing process can also be performed on the extracted text to improve the subsequent recognition accuracy of the text information to be recognized.

[0104] The text information to be recognized in the above embodiments includes the content sent by the salesman and also the content sent by the user. Combining the content of both parties in the interaction for conversation analysis is beneficial to more accurately judge the meaning of the content sent by the salesman and reduce the possibility of misjudgment. For example: when the customer asks a question or makes a personal understanding to the salesman during the consultation of a loan business, and the salesman replies yes or no, based on semantic understanding, the yes / no at this time will be retained as key text content in the text information to be recognized and will not be deleted.

[0105] In this embodiment, when it is recognized that the communication between the salesman and the customer is through text, by performing text preprocessing on the text information, the text information can express the true meaning more clearly, concisely and uniformly, which is beneficial to improving the recognition accuracy and efficiency of the text information to be recognized.

[0106] In some alternative embodiments, the above step S202 includes:

[0107] Identifying the text information to be recognized through the pre-trained sentiment analysis model to extract sentiment key features, and the pre-trained sentiment analysis model is trained according to a dataset with pre-labeled sentiment categories;

[0108] Judging whether there is a preset limited sentiment category based on the sentiment key features;

[0109] If there is the preset limited sentiment category, it is judged that there is the sentiment guiding behavior.

[0110] In this embodiment, the sentiment analysis model trained based on the data set with marked sentiment categories can automatically identify the sentiment words contained in the text information to be identified. For this, the identified sentiment words can be extracted as sentiment key features. Different sentiment words correspond to different sentiment categories, and the sentiment categories can include positive, negative, and neutral. For example, the text that the salesperson replied to the customer is: I am very happy to hear that you are satisfied, then the extracted sentiment key features include satisfied and happy, and the sentiment key features at this time are positive emotions. For another example, the text that the insurance salesperson replied to the customer is: I am very disappointed with the claim process this time, and the extracted sentiment key features are very disappointed, then the sentiment key features are negative emotions. For another example, the text that the salesperson replied to the customer is: I will continue to communicate with you, and the extracted sentiment key features are continue and communicate, and these sentiment key features are neutral emotions. When the sentiment categories are divided into neutral, positive and negative, the negative can be used as a limited sentiment category.

[0111] In some embodiments, the above-mentioned emotion categories can also be graded according to the emotion degree, and whether they belong to the limited emotion category is determined according to the level corresponding to the extracted key emotion features, and then whether there is emotion-guiding behavior is determined.

[0112] As a possible implementation method, the sentiment analysis model can be divided into 5 levels according to the degree of sentiment, with level 1 being extremely positive, level 2 being positive, level 3 being neutral, level 4 being negative, and level 5 being extremely negative. Extremely positive and extremely negative are limited sentiment categories, and the sentiment analysis model can accurately analyze the levels corresponding to the key sentiment features after pre-training. For example, an insurance salesperson replies to a customer: This financial product will definitely have a large return, and you must buy one to try. The extracted key sentiment features include "definitely", "return", "must", and "buy". These key sentiment features are extremely positive and are judged to belong to level 1, belonging to the limited sentiment category, and the salesperson has sentiment-guiding behavior. For another example, an insurance salesperson replies to a customer: I have explained this financial product countless times, why don't you understand it? The corresponding extracted key sentiment features include "explain", "countless times", and "still don't understand". These key sentiment features are extremely negative and are judged to belong to level 5, belonging to the limited sentiment category, and the salesperson has sentiment-guiding behavior.

[0113] As another possible implementation, when there are multiple key emotional features, if there are extremely positive and / or extremely negative features, they are directly judged as corresponding to level 1 and / or level 5, confirming the presence of emotion-guiding behavior; if not, they are not considered to belong to the limited emotion category, and it is judged that there is no emotion-guiding behavior.

[0114] As another possible implementation, the sentiment analysis model can be scored based on sentiment key features. Different sentiment key features are assigned different scores and weights. The scores and weights of one or more sentiment key features extracted from the text information to be identified are used to calculate the scoring result. The sentiment category corresponding to the text information to be identified is determined based on the scoring result. Different sentiment categories correspond to different score ranges, and thus it can be determined whether the text information to be identified belongs to a limited sentiment category.

[0115] In this embodiment, by training the sentiment analysis model, when identifying the text to be identified, the key sentiment features are extracted, and based on the key sentiment features, it is determined whether the text information to be identified belongs to the specified sentiment category, and then it is determined whether the salesperson has any emotional guidance behavior towards the customer. Based on the sentiment analysis model, the process and results of the customer service conversation can be automatically analyzed from the direction of emotional expression, which helps to identify the emotional guidance or illegal behavior of the salesperson during the communication process, and promptly remind the salesperson to make adjustments to avoid unnecessary misunderstandings and conflicts, thereby improving service quality and efficiency.

[0116] In some optional embodiments, the above step S203 includes:

[0117] Acquire the service type corresponding to the text information to be identified, where different service types correspond to different violation rules in the violation rule library;

[0118] Reading a target violation rule corresponding to the service type of the text information to be identified in the violation rule library through a rule engine, detecting the text information to be identified based on the target violation rule to determine whether it complies with the target violation rule;

[0119] If the target violation rule is met, then the text information to be identified is judged to have the violation behavior;

[0120] If the target violation rule is not met, it is determined that the text information to be identified does not contain the violation, and the violation is identified on the text information to be identified through the pre-trained quality inspection model to determine whether the violation exists.

[0121] In this embodiment, the violation rule library may include a knowledge graph established based on different marketing product information. For example, for financial products, it may include interest rates, returns, risk ratings, suitable groups, etc. A rule engine can be created for illegal speech that meets a series of rules and conditions, and allows administrators to define and manage a set of rules for illegal behaviors, such as specific words or sentences that exaggerate product features, prohibited speech, etc. The rule engine has flexible rule management and configuration functions. Users can add, edit or delete rules in the violation rule library as needed without modifying the underlying application code. The rule engine has high real-time performance and can make judgments quickly. For speech that meets the violation rules in the violation rule library, it can immediately warn the salesperson.

[0122] Furthermore, because the types of services provided by the salesperson to the customers are different, and the preset violation rules corresponding to different service types are also different, the service type to which the text information to be identified belongs can be first determined, and then the corresponding target violation rules can be read from the rule database according to the service type through the rule engine, and the target violation rules can include at least one rule. The target violation rules are further compared and judged with the text information to be identified. If the text information to be identified violates any rule in the target violation rules, it is determined that there is a violation. If it is determined based on the target violation rules that the text information to be identified does not violate the rules and the salesperson does not violate the rules, the violation identification is further performed again based on the pre-trained quality inspection model.

[0123] In this embodiment, by constructing a rule engine, the rule engine is used to read the target violation rule corresponding to the service type of the text information to be identified in the violation rule library for rule matching to determine whether the salesperson has violated the rules. The rule engine has stronger real-time performance, which is conducive to making judgments quickly. In addition, rule matching through the violation rule library to make violation judgments can quickly and accurately identify the words that violate some defined rules, and promptly determine whether to warn the salesperson, thereby reducing the possibility of adverse effects on customers.

[0124] In some optional embodiments, the identifying of illegal behaviors on the text information to be identified by the pre-trained quality inspection model to determine whether the illegal behaviors exist includes:

[0125] The text information to be identified is converted into a text vector by using the pre-trained quality inspection model, wherein the pre-trained quality inspection model is modeled based on a deep learning algorithm;

[0126] Extracting text feature vectors from the text vectors based on a deep learning algorithm;

[0127] Extracting text keywords based on the text content and text structure of the text information to be identified;

[0128] Classify based on the text feature vector and the text keywords, and determine whether there is any violation according to the classification result, wherein the classification result includes violation and non-violation;

[0129] If it is determined that the violation exists, the text information to be identified is marked with a violation label.

[0130] In this embodiment, the above-mentioned text vector can capture the semantic information in the text communicated between the salesperson and the customer, so that the quality inspection model can understand the meaning of the text more accurately. Among them, the method of converting the text into a numerical vector includes unique hot encoding, bag of words model, TF-IDF (term frequency-inverse document frequency) and word embedding model. The above-mentioned deep learning algorithm includes convolutional neural network, recurrent neural network or Transformer. By building a quality inspection model based on a deep learning algorithm, it can be ensured that the quality inspection model captures the contextual information, sentence structure, grammatical relationship, etc. in the text communicated between the salesperson and the customer, thereby achieving an in-depth understanding of the text semantics.

[0131] In some examples, when the quality inspection model is built based on a recurrent neural network, during the process of extracting text feature vectors from text vectors, the convolutional layer in the recurrent neural network can extract local features in the text; the recurrent unit can capture the temporal information in the text. When the quality inspection model is built based on the Transformer, in the Transformer model, the self-attention mechanism can calculate the correlation between different positions in the text. Finally, the text feature vector is extracted from the text vector.

[0132] In this embodiment, the above-mentioned extraction of text keywords can be implemented by TF-IDF algorithm, TextRank algorithm, LDA algorithm, etc. Text keywords can be high-frequency words, important phrases, or terms in a specific field, etc. The above-mentioned different text feature vectors and different text keywords correspond to different classification results, and the classification results can include violations and non-violations. Based on the quality inspection model, the method of classifying according to the two results of text feature vectors and text keywords to determine whether the salesperson has violated the rules can improve the accuracy of recognition.

[0133] As a possible implementation method, the result of classification according to the text feature vector can be used as the first classification result, and the result of classification according to the text keywords can be used as the second classification result. The two classification results are combined to determine whether there is a violation in the end. For example, when both the first classification result and the second classification result are violations, it is determined that a violation exists.

[0134] In this embodiment, identification is performed based on the pre-trained quality inspection model, and the violation analysis can be continued when the violation cannot be accurately identified by the defined violation rules, which is conducive to improving the accuracy of violation identification and timely warning the salesperson, ensuring that the salesperson provides customers with real information and helps customers make wise decisions. And by conducting quality inspection on the salesperson's speech, it can ensure that the salesperson's speech is standardized, avoid potential risks, and provide better services to customers.

[0135] It should be emphasized that in order to further ensure the privacy and security of the data involved in the speech quality inspection process in the above-mentioned artificial intelligence-based speech quality inspection method, it can also be stored in a node of a blockchain. The blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain (Block chain) is essentially a decentralized database. It is a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. The blockchain can include the underlying blockchain platform, the platform product service layer, and the application service layer.

[0136] The embodiments of the present application can acquire and process the data related to the speech quality inspection method based on artificial intelligence technology. 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.

[0137] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0138] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0139] Further references Figure 4 , as a response to the above Figure 2 In order to realize the method shown in the figure, the present application provides an embodiment of a speech quality inspection device based on artificial intelligence, and the device embodiment is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0140] like Figure 4 As shown, the artificial intelligence-based speech quality inspection device 400 described in this embodiment includes: a conversion module 401, an analysis module 402, a recognition module 403 and a sending module 404.

[0141] in:

[0142] The conversion module 401 is used to obtain session information, perform text conversion on the session information, and obtain text information to be recognized, wherein the session information includes at least one type of speech information;

[0143] The analysis module 402 is used to perform sentiment analysis on the text information to be identified based on a pre-trained sentiment analysis model to determine whether there is any sentiment-guiding behavior;

[0144] The identification module 403 is used to identify the violation behavior of the text information to be identified based on the violation rule library and the pre-trained quality inspection model to determine whether there is a violation behavior if the emotion-guiding behavior does not exist, wherein the pre-trained quality inspection model is trained based on the historical violation behavior data set;

[0145] The sending module 404 is used to issue a violation warning if the emotion-guiding behavior or the violation exists.

[0146] In this embodiment, the operations performed by the above modules respectively correspond one-to-one to the steps of the artificial intelligence-based speech quality inspection method in the aforementioned implementation method, and will not be repeated here.

[0147] In some optional implementations of this embodiment, the conversion module includes:

[0148] The first judgment submodule is used to obtain the session information and judge whether the session information includes the voice information; the first preprocessing submodule is used to perform audio preprocessing on the voice information to obtain preprocessed audio information if the session information includes the voice information; the first text conversion submodule is used to extract audio features from the preprocessed audio information based on a speech recognition model, and perform text conversion on the audio features through the speech recognition model to obtain the text information to be recognized.

[0149] In some optional implementations of this embodiment, the conversion module also includes: a second judgment submodule, used to obtain the session information and determine whether the session information includes the picture information; a first detection submodule, used to detect whether there is picture change information in the picture information based on a computer vision model if the session information includes the picture information, and the picture change information includes text; a second preprocessing submodule, used to perform picture preprocessing on the picture information if the picture change information exists, and locate and segment the picture change information in the preprocessed picture information to obtain the text to be extracted; a first feature extraction submodule, used to perform text feature extraction on the text to be extracted by using an optical character recognition algorithm to obtain the text information to be recognized.

[0150] In some optional implementations of this embodiment, the conversion module also includes: a third judgment submodule, used to obtain the session information and determine whether the session information includes the text information; and a third preprocessing submodule, used to perform text preprocessing on the text information if the session information includes the text information to obtain the text information to be identified.

[0151] In some optional implementations of the present embodiment, the analysis module includes: a second feature extraction submodule, used to identify the text information to be identified through the pre-trained sentiment analysis model, and extract key sentiment features, wherein the pre-trained sentiment analysis model is trained based on a data set with labeled sentiment categories; a category judgment submodule, used to judge whether there is a preset limited sentiment category based on the sentiment key features; and a first judgment submodule, used to judge whether the sentiment-guiding behavior exists if the preset limited sentiment category exists.

[0152] In some optional implementations of the present embodiment, the identification module includes: an acquisition submodule, used to acquire the service type corresponding to the text information to be identified, different service types correspond to different violation rules in the violation rule library; a second detection submodule, which uses a rule engine to read the target violation rule corresponding to the service type of the text information to be identified in the violation rule library, and detects the text information to be identified based on the target violation rule to determine whether it complies with the target violation rule; a second determination submodule, which is used to determine that the text information to be identified has the violation if it complies with the target violation rule; and an identification submodule, which is used to determine that the text information to be identified does not have the violation if it does not comply with the target violation rule, and identifies the violation of the text information to be identified through the pre-trained quality inspection model to determine whether the violation exists.

[0153] In some optional implementations of the present embodiment, the recognition submodule is specifically used to: convert the text information to be recognized into a text vector through the pre-trained quality inspection model, and the pre-trained quality inspection model is modeled based on a deep learning algorithm; extract text feature vectors from the text vectors based on the deep learning algorithm; extract text keywords based on the text content and text structure of the text information to be recognized; perform classification based on the text feature vectors and the text keywords, and determine whether there is a violation based on the classification results, wherein the classification results include violation and no violation; if it is determined that the violation exists, the text information to be recognized is marked with a violation label.

[0154] In the above optional embodiments, the operations performed by the above sub-modules, units or sub-units respectively correspond one-to-one to the steps of the artificial intelligence-based speech quality inspection method in the aforementioned implementation mode, and will not be repeated here.

[0155] To solve the above technical problems, the present application also provides a computer device. Figure 5 , Figure 5It is a basic structural block diagram of the computer device of this embodiment. The computer device 5 includes a memory 51, a processor 52, and a network interface 53 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 5 with components 51-53, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0156] The computer device 5 may be a computing device such as a desktop computer, a notebook, a palm computer, a cloud server, etc. The computer device 5 may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0157] The memory 51 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 51 can be an internal storage unit of the computer device 5, such as a hard disk or memory of the computer device 5. In other embodiments, the memory 51 can also be an external storage device of the computer device 5, such as a plug-in hard disk equipped on the computer device 5, a smart memory card (Smart Media Card, SMC), etc. Of course, the memory 51 can also include both the internal storage unit of the computer device 5 and its external storage device. In this embodiment, the memory 51 is generally used to store the operating system and various application software installed on the computer device 5, such as computer-readable instructions of the speech quality inspection method based on artificial intelligence.

[0158] The processor 52 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 52 is generally used to control the overall operation of the computer device 5. In this embodiment, the processor 52 is used to run computer-readable instructions stored in the memory 51 or process data, such as computer-readable instructions for running the artificial intelligence-based speech quality inspection method.

[0159] The network interface 53 may include a wireless network interface or a wired network interface. The network interface 53 is generally used to establish a communication connection between the computer device 5 and other electronic devices.

[0160] The present application also provides another embodiment, namely, providing a computer-readable storage medium, which stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the above-mentioned artificial intelligence-based speech quality inspection method.

[0161] The non-Company software tools or components appearing in the embodiments of this application are merely examples and do not represent actual use.

[0162] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), including a number of instructions for a terminal device to execute the method steps described in each embodiment of the present application.

[0163] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A speech quality inspection method based on artificial intelligence, characterized in that: The steps include: Acquiring conversation information, and performing text conversion on the conversation information to obtain text information to be recognized, wherein the conversation information includes at least one type of speech information; Based on the pre-trained sentiment analysis model, the text information to be identified is subjected to sentiment analysis to determine whether there is any sentiment-guiding behavior; If the emotion-guiding behavior does not exist, then based on the violation rule library and the pre-trained quality inspection model, the text information to be identified is subjected to violation identification to determine whether there is a violation, wherein the pre-trained quality inspection model is trained based on a historical violation data set; If the aforementioned emotionally-guided behavior or the aforementioned illegal behavior exists, a violation warning will be issued.

2. The method for quality inspection of speech based on artificial intelligence according to claim 1 is characterized in that: The conversation information includes voice information, and the step of acquiring the conversation information and converting the conversation information into text to obtain text information to be recognized includes: Acquire the session information, and determine whether the session information includes the voice information; If the conversation information includes the voice information, performing audio preprocessing on the voice information to obtain preprocessed audio information; The pre-processed audio information is subjected to audio feature extraction based on a speech recognition model, and the audio feature is subjected to text conversion through the speech recognition model to obtain the text information to be recognized.

3. The method for quality inspection of speech based on artificial intelligence according to claim 1 is characterized in that: The session information also includes picture information. The step of acquiring the session information and converting the session information into text to obtain the text information to be identified includes: Acquire the session information, and determine whether the session information includes the picture information; If the session information includes the picture information, detecting whether there is picture modification information in the picture information based on a computer vision model, wherein the picture modification information includes text; If the image change information exists, the image information is preprocessed, and the image change information in the preprocessed image information is located and the text is segmented to obtain the text to be extracted; The text features of the text to be extracted are extracted by an optical character recognition algorithm to obtain the text information to be recognized.

4. The method for quality inspection of speech based on artificial intelligence according to claim 1 is characterized in that: The session information also includes text information. The step of acquiring the session information and converting the session information into text to obtain the text information to be identified includes: Acquire the session information, and determine whether the session information includes the text information; If the session information includes the text information, the text information is preprocessed to obtain the text information to be identified.

5. The method for quality inspection of speech based on artificial intelligence according to claim 1 is characterized in that: The step of performing sentiment analysis on the text information to be identified based on the pre-trained sentiment analysis model to determine whether there is any sentiment-guiding behavior includes: The text information to be identified is identified by the pre-trained sentiment analysis model to extract key sentiment features, wherein the pre-trained sentiment analysis model is trained based on a data set with marked sentiment categories; Determining whether there is a preset limited emotion category based on the key emotion features; If the preset limited emotion category exists, it is determined that the emotion guiding behavior exists.

6. The method for quality inspection of speech based on artificial intelligence according to claim 1 is characterized in that: The step of identifying the violation behavior of the to-be-identified text information based on the violation rule library and the pre-trained quality inspection model and determining whether there is a violation behavior includes: Acquire the service type corresponding to the text information to be identified, where different service types correspond to different violation rules in the violation rule library; Reading a target violation rule corresponding to the service type of the text information to be identified in the violation rule library through a rule engine, detecting the text information to be identified based on the target violation rule to determine whether it complies with the target violation rule; If the target violation rule is met, then the text information to be identified is judged to have the violation behavior; If the target violation rule is not met, it is determined that the text information to be identified does not contain the violation, and the violation is identified on the text information to be identified through the pre-trained quality inspection model to determine whether the violation exists.

7. The method for quality inspection of speech based on artificial intelligence according to claim 6 is characterized in that: The step of identifying the illegal behavior of the to-be-identified text information through the pre-trained quality inspection model and determining whether the illegal behavior exists includes: The text information to be identified is converted into a text vector by using the pre-trained quality inspection model, wherein the pre-trained quality inspection model is modeled based on a deep learning algorithm; Extracting text feature vectors from the text vectors based on a deep learning algorithm; Extracting text keywords based on the text content and text structure of the text information to be identified; Classify based on the text feature vector and the text keywords, and determine whether there is any violation according to the classification result, wherein the classification result includes violation and non-violation; If it is determined that the violation exists, the text information to be identified is marked with a violation label.

8. A speech quality inspection device based on artificial intelligence, characterized in that: The device comprises: A conversion module, used to obtain session information, perform text conversion on the session information, and obtain text information to be recognized, wherein the session information includes at least one type of speech information; An analysis module is used to perform sentiment analysis on the text information to be identified based on a pre-trained sentiment analysis model to determine whether there is any sentiment-guiding behavior; An identification module, for identifying illegal behaviors on the text information to be identified based on the illegal rule library and the pre-trained quality inspection model to determine whether there are illegal behaviors if the emotional guidance behaviors do not exist, wherein the pre-trained quality inspection model is trained based on the historical illegal behavior data set; The sending module is used to issue a violation warning if the emotion-guiding behavior or the violation exists.

9. A computer device, characterized in that: It includes a memory and a processor, the memory stores computer-readable instructions, and the processor implements the steps of the artificial intelligence-based speech quality inspection method as described in any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the artificial intelligence-based speech quality inspection method as described in any one of claims 1 to 7.

Images