Artificial intelligence-based verbal skill generation method and device, computer equipment and medium

Abstract

The invention belongs to the technical field of artificial intelligence, relates to an artificial intelligence-based verbal skill generation method and device, computer equipment and a storage medium, is applied to the field of financial science and technology, and comprises the following steps: if a verbal skill generation request input by a user is received, extracting insurance type product information and scene description information from the verbal skill generation request; calling a preset verbal skill generation model; wherein the verbal skill generation model is a model generated by performing training and fine tuning processing on a pre-training language model based on a preset fine tuning sample set and a fine tuning algorithm; processing the insurance type product information and the scene description information based on a verbal skill generation model to obtain a verbal skill text; performing adjustment processing on the verbal skill text based on the adjustment strategy to obtain a target verbal skill text; and returning the target verbal skill text to the user. In addition, the invention also relates to a block chain technology, and the target verbal skill text can be stored in a block chain. Based on the use of the verbal skill generation model, the generation efficiency and the generation accuracy of the verbal skill text are effectively improved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence development technology and financial technology, and in particular to artificial intelligence-based speech generation methods, devices, computer equipment and storage media. Background Art

[0002] In the insurance industry, with the continuous increase in the types of insurance products and the increasing diversification of market demand, insurance companies are facing huge challenges, one of which is how to efficiently and accurately convey product information to target customers and promote sales. Traditionally, insurance companies rely on manually compiled scripts to deal with different products and scenarios, but this approach has significant limitations.

[0003] First, manual script writing is inefficient. There are many types of insurance products, each with its own unique coverage, clause details and applicable scenarios, which requires script writers to have deep professional knowledge and rich industry experience. However, even so, in the face of rapidly changing market demands and the continuous introduction of new products, the speed of manual script writing is still difficult to meet actual business needs.

[0004] Secondly, it is difficult to guarantee the accuracy of manually written scripts. Different writers may have different understandings and expressions of the same product, which may lead to deviations in the script content or omission of important information. This not only affects the customer experience, but may also cause unnecessary misunderstandings and disputes. Summary of the invention

[0005] The purpose of the embodiments of the present application is to propose a method, device, computer equipment and storage medium for generating speech based on artificial intelligence, so as to solve the technical problems that the existing method of manually compiling speech in the insurance industry is inefficient and difficult to ensure accuracy.

[0006] In order to solve the above technical problems, the embodiment of the present application provides a method for generating speech based on artificial intelligence, which adopts the following technical solution:

[0007] Determine whether a speech generation request input by a user is received; wherein the speech generation request carries insurance product information and scenario description information;

[0008] If yes, extract the insurance product information and the scenario description information from the speech generation request;

[0009] Calling a preset speech generation model; wherein the speech generation model is a model generated by training and fine-tuning a pre-trained language model based on a preset fine-tuning sample set and a fine-tuning algorithm;

[0010] Processing the insurance product information and the scenario description information based on the speech generation model to obtain corresponding speech text;

[0011] Adjust the speech text based on a preset adjustment strategy to obtain a corresponding target speech text;

[0012] The target speech text is returned to the user.

[0013] Furthermore, the step of adjusting the speech text based on a preset adjustment strategy to obtain a corresponding target speech text specifically includes:

[0014] Performing word verification on the speech text to obtain corresponding word verification results;

[0015] Adjust the speech text based on the word verification result to obtain a corresponding first speech text;

[0016] Performing a compliance check on the first script text to obtain a corresponding compliance check result;

[0017] Adjusting the first speech text based on the compliance check result to obtain a corresponding second speech text;

[0018] Performing sentiment analysis on the second speech text to obtain a corresponding sentiment analysis result;

[0019] Adjust the second speech text based on the sentiment analysis result to obtain a corresponding third speech text;

[0020] The third speech text is used as the target speech text.

[0021] Furthermore, the step of adjusting the second speech text based on the sentiment analysis result to obtain a corresponding third speech text specifically includes:

[0022] Determining whether the content of the sentiment analysis result is negative;

[0023] If yes, perform word replacement processing on the second speech text to obtain the corresponding first specified speech text;

[0024] Perform sentence replacement processing on the first specified speech text to obtain a corresponding second specified speech text;

[0025] The second designated speech text is used as the third speech text.

[0026] Furthermore, before the step of calling the preset speech generation model, the method further includes:

[0027] Obtaining a first iteration speech set corresponding to the pre-trained language model;

[0028] Get the preset general question and answer pairs;

[0029] Constructing the corresponding fine-tuning sample set based on the first iterative speech set and the general question-answer pair;

[0030] Calling the pre-trained language model;

[0031] Based on a low-rank adaptive fine-tuning algorithm, the pre-trained language model is trained and fine-tuned using the fine-tuning sample set to obtain a specified language model that meets preset construction conditions;

[0032] The specified language model is used as the speech generation model.

[0033] Furthermore, after the step of returning the target speech text to the user, the method further includes:

[0034] Call the preset speech scoring model;

[0035] Generate a target speech score corresponding to the target speech text based on a preset speech scoring model;

[0036] Determine whether the target speech score is greater than a preset score threshold;

[0037] If yes, call the preset initial script library;

[0038] The target speech text is stored in the initial speech library to generate a corresponding high-quality speech library.

[0039] Furthermore, before the step of calling the preset speech scoring model, the method further includes:

[0040] Obtaining a second iteration speech set corresponding to the pre-trained language model;

[0041] Scoring the speech set based on a preset scoring strategy to obtain a speech score corresponding to the second iterative speech set;

[0042] Constructing a corresponding speech score data set based on the speech score and the second iterative speech set;

[0043] Call the preset initial model;

[0044] Training and evaluating the initial model based on the speech score data set to obtain a target model that meets preset performance requirements;

[0045] The target model is used as the speech scoring model.

[0046] Furthermore, the step of scoring the speech set based on a preset scoring strategy to obtain a speech score corresponding to the second iterative speech set specifically includes:

[0047] Obtaining preset scoring dimensions; wherein the scoring dimensions at least include language logic, language fluency, compliance of speech, persuasiveness of speech, and rationality of case;

[0048] A preset weight generation strategy is used to generate dimension weights corresponding to the scoring dimensions;

[0049] Obtaining a specified speech in the speech set; wherein the specified speech is any one of all the speech included in the speech set;

[0050] Obtain the dimension score of the specified speech under the scoring dimension;

[0051] The dimension score of the designated speech under the scoring dimension is scored based on a preset scoring formula, and the dimension weight corresponding to the scoring dimension is calculated and processed to obtain the designated speech score of the designated speech.

[0052] In order to solve the above technical problems, the embodiment of the present application also provides a speech generation device based on artificial intelligence, which adopts the following technical solution:

[0053] The first judgment module is used to judge whether a speech generation request input by the user is received; wherein the speech generation request carries insurance product information and scenario description information;

[0054] An extraction module, configured to extract the insurance product information and the scenario description information from the speech generation request;

[0055] A first calling module is used to call a preset speech generation model; wherein the speech generation model is a model generated by training and fine-tuning a pre-trained language model based on a preset fine-tuning sample set and a fine-tuning algorithm;

[0056] A processing module, used to process the insurance product information and the scenario description information based on the speech generation model to obtain a corresponding speech text;

[0057] An adjustment module, used to adjust the speech text based on a preset adjustment strategy to obtain a corresponding target speech text;

[0058] A return module is used to return the target speech text to the user.

[0059] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0060] Determine whether a speech generation request input by a user is received; wherein the speech generation request carries insurance product information and scenario description information;

[0061] If yes, extract the insurance product information and the scenario description information from the speech generation request;

[0062] Calling a preset speech generation model; wherein the speech generation model is a model generated by training and fine-tuning a pre-trained language model based on a preset fine-tuning sample set and a fine-tuning algorithm;

[0063] Processing the insurance product information and the scenario description information based on the speech generation model to obtain corresponding speech text;

[0064] Adjust the speech text based on a preset adjustment strategy to obtain a corresponding target speech text;

[0065] The target speech text is returned to the user.

[0066] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0067] Determine whether a speech generation request input by a user is received; wherein the speech generation request carries insurance product information and scenario description information;

[0068] If yes, extract the insurance product information and the scenario description information from the speech generation request;

[0069] Calling a preset speech generation model; wherein the speech generation model is a model generated by training and fine-tuning a pre-trained language model based on a preset fine-tuning sample set and a fine-tuning algorithm;

[0070] Processing the insurance product information and the scenario description information based on the speech generation model to obtain corresponding speech text;

[0071] Adjust the speech text based on a preset adjustment strategy to obtain a corresponding target speech text;

[0072] The target speech text is returned to the user.

[0073] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0074] The present application first determines whether a speech generation request input by the user is received; if so, the insurance product information and the scene description information are extracted from the speech generation request; then the preset speech generation model is called; wherein the speech generation model is a model generated by training and fine-tuning a pre-trained language model based on a preset fine-tuning sample set and a fine-tuning algorithm; then the insurance product information and the scene description information are processed based on the speech generation model to obtain a corresponding speech text; subsequently, the speech text is adjusted based on a preset adjustment strategy to obtain a corresponding target speech text; finally, the target speech text is returned to the user. After receiving the speech generation request input by the user, the present application extracts the insurance product information and the scene description information from the speech generation request, and then uses the speech generation model generated by training and fine-tuning a pre-trained language model with a preset fine-tuning sample set and a fine-tuning algorithm to process the insurance product information and the scene description information, thereby realizing the rapid and accurate generation of matching speech texts, effectively improving the efficiency of speech text generation, and ensuring the accuracy of the obtained speech texts. In addition, the adjustment strategy will be intelligently used to adjust the speech text to obtain the target speech text, so as to further improve the accuracy of the generated target speech text, so that the target speech text can be returned to the user later, which is conducive to improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] 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.

[0076] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0077] Figure 2 A flowchart of an embodiment of a method for generating speech based on artificial intelligence according to the present application;

[0078] Figure 3 It is a structural schematic diagram of an embodiment of an artificial intelligence-based speech generation device according to the present application;

[0079] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0080] 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.

[0081] 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.

[0082] 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.

[0083] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0084] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0085] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to the laptop computer 1011, tablet computer 1012 or mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0086] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0087] It should be noted that the artificial intelligence-based speech generation method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the artificial intelligence-based speech generation device is generally set in the server / terminal device.

[0088] 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 implementation requirements.

[0089] Continue to refer Figure 2 , showing a flow chart of an embodiment of the method for generating speech 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, and some steps can be omitted. The method for generating speech based on artificial intelligence provided in the embodiment of the present application can be applied to any scenario where speech generation is required, and the method for generating speech based on artificial intelligence can be applied to products in these scenarios, for example, insurance speech generation in the field of finance and insurance. The method for generating speech based on artificial intelligence includes the following steps:

[0090] Step S201, determining whether a speech generation request input by a user is received; wherein the speech generation request carries insurance product information and scenario description information.

[0091] In this embodiment, the electronic device (for example, Figure 1The server / terminal device shown in the figure) can obtain the speech generation request through a wired connection or a wireless connection. It should be pointed out that the above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future. The executor of this application is a speech generation system, or simply referred to as the system. The above-mentioned speech generation request is a request triggered by the user through the system interface of the speech generation system to request the generation of matching insurance speech based on the input insurance product information and scenario description information. Among them, the above-mentioned insurance product information may include product name, coverage, premium and other information. The above-mentioned scenario description information may include claims scenarios, critical illness insurance scenarios, and so on.

[0092] Step S202: If yes, extract the insurance product information and the scenario description information from the script generation request.

[0093] In this embodiment, the above-mentioned speech generation request can be subjected to information parsing processing to extract the insurance product information and scenario description information carried in the speech generation request.

[0094] Step S203, calling the preset speech generation model.

[0095] In this embodiment, the above-mentioned speech generation model is a model generated by training and fine-tuning the pre-trained language model based on a preset fine-tuning sample set and fine-tuning algorithm. The specific process of constructing the speech generation model will be further described in detail in the subsequent specific embodiments of this application, and will not be elaborated on here.

[0096] Step S204: Process the insurance product information and the scenario description information based on the speech generation model to obtain corresponding speech text.

[0097] In this embodiment, by inputting the insurance product information and the scenario description information into the above-mentioned speech generation model, the speech generation model will perform reasoning processing based on the insurance product information and the scenario description information, and automatically generate corresponding speech text. The generated speech may include a detailed introduction to the product, answers to customer needs, and sentences that guide customers to further understand or purchase.

[0098] Step S205, adjusting the speech text based on a preset adjustment strategy to obtain a corresponding target speech text.

[0099] In the present embodiment, the above-mentioned specific implementation process of adjusting the speech text based on the preset adjustment strategy to obtain the corresponding target speech text will be described in further detail in the subsequent specific embodiments of the present application, and will not be elaborated in detail here.

[0100] Step S206, returning the target speech text to the user.

[0101] In this embodiment, the target speech text can be returned to the user by sending by email, sending by SMS, displaying on a system interface, etc.

[0102] The present application first determines whether a speech generation request input by the user is received; if so, the insurance product information and the scene description information are extracted from the speech generation request; then the preset speech generation model is called; wherein the speech generation model is a model generated by training and fine-tuning a pre-trained language model based on a preset fine-tuning sample set and a fine-tuning algorithm; then the insurance product information and the scene description information are processed based on the speech generation model to obtain a corresponding speech text; subsequently, the speech text is adjusted based on a preset adjustment strategy to obtain a corresponding target speech text; finally, the target speech text is returned to the user. After receiving the speech generation request input by the user, the present application extracts the insurance product information and the scene description information from the speech generation request, and then uses the speech generation model generated by training and fine-tuning a pre-trained language model with a preset fine-tuning sample set and a fine-tuning algorithm to process the insurance product information and the scene description information, thereby realizing the rapid and accurate generation of matching speech texts, effectively improving the efficiency of speech text generation, and ensuring the accuracy of the obtained speech texts. In addition, the adjustment strategy will be intelligently used to adjust the speech text to obtain the target speech text, so as to further improve the accuracy of the generated target speech text, so that the target speech text can be returned to the user later, which is conducive to improving the user experience.

[0103] In some optional implementations, step S205 includes the following steps:

[0104] Perform word verification on the speech text to obtain corresponding word verification results.

[0105] In this embodiment, the speech text can be cleaned first, including removing unnecessary spaces, line breaks, special characters, etc., to ensure the uniformity of the text format, and obtain the preprocessed designated speech text. Then, the designated speech text is subjected to grammar checking and semantic verification to obtain the corresponding word verification results. Among them, the designated speech text can be subjected to lexical analysis (word segmentation, part-of-speech tagging) and syntactic analysis (constructing a dependency syntax tree) using natural language processing tools to check whether there are grammatical errors, such as inconsistency between subject and predicate, incomplete sentence structure, etc. In addition, semantic role labeling or textual entailment judgment based on a pre-trained model is used to help identify logical contradictions, unclear meanings or ambiguous expressions in speech. In addition, the above-mentioned word verification results include passing both grammar checking and semantic verification, or failing to pass both grammar checking and semantic verification.

[0106] The speech text is adjusted based on the word verification result to obtain a corresponding first speech text.

[0107] In this embodiment, if the content of the word verification result fails to pass both the grammatical check and the semantic check, rule replacement (such as replacing "she run" with "she runs") or an error correction model based on machine learning is used to automatically correct the dialogue text to obtain the corresponding first dialogue text.

[0108] Perform a compliance check on the first speech text to obtain a corresponding compliance check result.

[0109] In this embodiment, the first speech text can be matched with the rules in the preset rule library to check whether there is any illegal content. Among them, the matching process can be performed by string search, regular expression matching, etc. In addition, the laws, regulations and internal regulations of the insurance industry can be collected and converted into executable rules, such as keyword lists, regular expression patterns, etc., and the corresponding rule library can be constructed. In addition, the content of the above-mentioned compliance check results includes passing the compliance check or failing the compliance check.

[0110] The first speech text is adjusted based on the compliance check result to obtain a corresponding second speech text.

[0111] In this embodiment, if the content of the above compliance check result is that the compliance check has not passed, the illegal content detected in the first speech text is corrected or marked according to the guiding principles in the rule base to obtain the corresponding second speech text.

[0112] Perform sentiment analysis on the second speech text to obtain corresponding sentiment analysis results.

[0113] In this embodiment, the sentiment analysis model pre-trained can be used to analyze the sentiment tendency of the second speech text, and the corresponding sentiment analysis result can be output, wherein the content of the sentiment analysis result can include positive or negative.

[0114] The second speech text is adjusted based on the sentiment analysis result to obtain a corresponding third speech text.

[0115] In this embodiment, the specific implementation process of adjusting the second speech text based on the sentiment analysis result to obtain the corresponding third speech text will be described in further detail in subsequent specific embodiments of the present application and will not be elaborated here.

[0116] The third speech text is used as the target speech text.

[0117] This application performs word verification on the speech text to obtain the corresponding word verification result; then adjusts the speech text based on the word verification result to obtain the corresponding first speech text; and performs compliance check on the first speech text to obtain the corresponding compliance check result; then adjusts the first speech text based on the compliance check result to obtain the corresponding second speech text; and performs sentiment analysis on the second speech text to obtain the corresponding sentiment analysis result; subsequently adjusts the second speech text based on the sentiment analysis result to obtain the corresponding third speech text; and uses the third speech text as the target speech text. This application performs word verification, compliance check and sentiment analysis on the speech text, and then adjusts the speech text based on the obtained word verification result, compliance check result and sentiment analysis result, which can effectively improve the accuracy and intelligence of the generated target speech text, and is conducive to providing more powerful support for insurance sales and insurance services.

[0118] In some optional implementations of this embodiment, adjusting the second speech text based on the sentiment analysis result to obtain a corresponding third speech text includes the following steps:

[0119] Determine whether the content of the sentiment analysis result is negative.

[0120] In this embodiment, the content of the above sentiment analysis result may include positive or negative.

[0121] If so, perform word replacement processing on the second speech text to obtain the corresponding first specified speech text.

[0122] In this embodiment, if the content of the sentiment analysis result is negative, the uncertain words in the second speech text are replaced with certain words, for example, "may" is replaced with "definitely" to increase certainty.

[0123] The first designated speech text is subjected to sentence replacement processing to obtain a corresponding second designated speech text.

[0124] In this embodiment, the second designated speech text can be obtained by replacing the interrogative sentence in the first designated speech text with an affirmative sentence to enhance confidence.

[0125] The second designated speech text is used as the third speech text.

[0126] This application determines whether the content of the sentiment analysis result is negative; if so, performs word replacement processing on the second speech text to obtain the corresponding first designated speech text; then performs sentence replacement processing on the first designated speech text to obtain the corresponding second designated speech text; and subsequently uses the second designated speech text as the third speech text. When this application detects that the content of the sentiment analysis result is negative, it will intelligently perform word replacement processing and sentence replacement processing on the second speech text to achieve rapid and accurate adjustment of the second speech text, which is conducive to improving the certainty and intelligence of the generated third speech text.

[0127] In some optional implementations, before step S203, the electronic device may further perform the following steps:

[0128] Obtain a first iteration speech set corresponding to the pre-trained language model.

[0129] In this embodiment, the construction process of the above-mentioned pre-trained language model includes: 1. Data acquisition and cleaning. 1.1, obtain product information of each type of insurance, including insurance age limit, premium, insured amount, number of packages, insurance liability, etc. 1.2, obtain product posters from insurance websites. Since there are many missing fields in product information, in order to supplement product information, features, etc., obtain product posters from insurance websites, use OCR technology to identify text information in posters, and polish and supplement the identified text information in combination with large language models to improve product information. 1.3, mine excellent speech in the existing speech library, summarize product selling points, features, and advantages for different types of insurance (such as the 0 deductible of Baiwanyi is its product advantage, which needs to be highlighted), and design prompt cases for initial speech based on existing speech. 2. Construction process of pre-trained language model. 2.1 Large model speech generation-business personnel feedback. 2.1.1, design prompt engineering according to different theme scenarios and business requirements, and standardize prompt engineering components. Prompt engineering uniformly uses: role + scenario + requirement + product information + case form. 2.1.2. Use the one-shot method of the big model (qwen model) to generate insurance scripts for different topics and different types of insurance based on the designed prompt words. Then conduct a preliminary evaluation of the generated insurance scripts, modify the prompt words based on the evaluation results, and perform iterative optimization. 2.1.3. Submit the script results generated by the big model to business personnel for review, accumulate feedback on the scripts, and accumulate a compliance matters library and a banned word library based on business feedback for subsequent script optimization. 2.1.4. Repeat steps 2.1.1-2.1.3 until the scripts generated by the big model meet the business standards to obtain a big model for the vertical field related to insurance, that is, the above-mentioned pre-trained language model. Among them, the above-mentioned first iteration script set refers to the script set accumulated during the script iteration process in step 2.1.

[0130] Get preset common question-answer pairs.

[0131] In this embodiment, the general question-answer pair refers to the general question-answer pair that matches the first iterative speech set.

[0132] The corresponding fine-tuning sample set is constructed based on the first iterative speech set and the general question and answer pair.

[0133] In this embodiment, the first iteration speech set is annotated, including scoring of language logic, fluency, compliance, etc., and the annotated speech is combined with general question and answer pairs to construct a corresponding fine-tuning sample set.

[0134] The pre-trained language model is called.

[0135] Based on a low-rank adaptive fine-tuning algorithm, the pre-trained language model is trained and fine-tuned using the fine-tuning sample set to obtain a specified language model that meets preset construction conditions.

[0136] In this embodiment, the pre-trained language model is fine-tuned by using LoRA technology (Low-Rank Adaptation of Large Language Models). LoRA achieves rapid adaptation to the model by adding a low-rank matrix to the model parameter matrix without retraining the entire model. The fine-tuning sample set is then divided into a training data set and a test data set. The training data set is input into the fine-tuned model for training. During the training process, the performance indicators of the model (such as accuracy, loss value, etc.) are monitored to evaluate the training effect of the model. The fine-tuned model is then evaluated using the test data set to check whether its performance on the new task meets the requirements, and a specified language model that is qualified is obtained. The qualified model will then be deployed to the production environment for generating insurance speech.

[0137] The specified language model is used as the speech generation model.

[0138] The present application obtains the first iterative speech set corresponding to the pre-trained language model; and obtains the preset general question-answer pair; then constructs the corresponding fine-tuning sample set based on the first iterative speech set and the general question-answer pair; subsequently calls the pre-trained language model; further trains and fine-tunes the pre-trained language model based on the fine-tuning algorithm with low-rank adaptation, and obtains a specified language model that meets the preset construction conditions; finally, the specified language model is used as the speech generation model. The present application constructs a fine-tuning sample set based on the first iterative speech set corresponding to the pre-trained language model and the preset general question-answer pair, and then trains and fine-tunes the pre-trained language model based on the fine-tuning algorithm with low-rank adaptation, so that a speech generation model that meets the requirements can be constructed, which improves the construction efficiency of the speech generation model and ensures the model effect of the speech generation model.

[0139] In some optional implementations, after step S206, the electronic device may further perform the following steps:

[0140] Call the preset speech scoring model.

[0141] In this embodiment, the construction process of the above-mentioned speech scoring model will be described in further detail in subsequent specific embodiments, and will not be elaborated here.

[0142] A target speech score corresponding to the target speech text is generated based on a preset speech scoring model.

[0143] In this embodiment, the target speech text is input into the speech scoring model, and the target speech text is scored according to the speech scoring model, so as to output the corresponding target speech score.

[0144] Determine whether the target speech score is greater than a preset score threshold.

[0145] In this embodiment, there is no specific limitation on the value of the score threshold, which can be set according to the actual definition requirements of high-quality speech. If the target speech score is greater than the score threshold, the target speech text is determined to be a high-quality speech. If the target speech score is less than the score threshold, the target speech text is determined not to be a high-quality speech.

[0146] If so, call the preset initial script library.

[0147] In this embodiment, the above-mentioned initial script library is a pre-built database for storing high-quality insurance scripts.

[0148] The target speech text is stored in the initial speech library to generate a corresponding high-quality speech library.

[0149] In this embodiment, the target speech text can be stored in the initial speech library, thereby completing the construction process of a high-quality insurance speech database in the insurance field.

[0150] This application calls a preset speech scoring model; then generates a target speech score corresponding to the target speech text based on the preset speech scoring model; then determines whether the target speech score is greater than the preset score threshold; if so, calls the preset initial speech library; and subsequently stores the target speech text in the initial speech library to generate a corresponding high-quality speech library. After generating the target speech text based on the speech generation model, this application will also intelligently generate a target speech score corresponding to the target speech text based on the use of the speech scoring model, so that the evaluation process of the target speech text can be automatically and accurately completed, the processing efficiency of the speech evaluation is improved, and the accuracy of the target speech score is guaranteed. In addition, when it is detected that the target speech score of the target speech text is greater than the preset score threshold, the target speech text will be intelligently stored in the initial speech library to generate a corresponding high-quality speech library to realize the automatic recording and processing of high-quality insurance speech, and improve the construction intelligence and efficiency of the high-quality speech library.

[0151] In some optional implementations of this embodiment, before the step of calling the preset speech scoring model, the electronic device may further perform the following steps:

[0152] Obtain a second iterative speech set corresponding to the pre-trained language model.

[0153] In this embodiment, the second iterative speech set refers to the speech set accumulated by the pre-trained language model during the speech iteration process, and the first iterative speech set may be referred to.

[0154] The speech set is scored based on a preset scoring strategy to obtain a speech score corresponding to the second iterative speech set.

[0155] In this embodiment, the specific implementation process of scoring the speech set based on the preset scoring strategy to obtain the speech score corresponding to the second iterative speech set will be described in further detail in subsequent specific embodiments of the present application and will not be elaborated here.

[0156] A corresponding speech score data set is constructed based on the speech score and the second iterative speech set.

[0157] In this embodiment, the speech score and the second iterative speech set can be sorted to ensure the accuracy and consistency of the data, and a speech score data set can be obtained. Then, duplicate, abnormal or invalid data records are removed to ensure the quality of the speech score data set. Subsequently, the sorted speech score data set is divided into a training set and a validation set, for example, with a ratio of 70% training set and 30% validation set.

[0158] Recall the preset initial model.

[0159] In this embodiment, the BERT (Bidirectional Encoder Representations from Transformers) model can be selected as the basic model for speech-score mapping, that is, the initial model, because it performs well in natural language processing tasks.

[0160] The initial model is trained and evaluated based on the speech score data set to obtain a target model that meets preset performance requirements.

[0161] In this embodiment, the parameters of the BERT model, such as learning rate, batch size, training rounds, etc., can be configured according to task requirements. Then the speech text in the training set is converted into a format that the BERT model can handle and used as input features. At the same time, the scores of each dimension and the total score are used as labels. The BERT model is then trained using the training set, and the model parameters are adjusted through the back propagation algorithm to minimize the difference between the predicted score and the actual score. The performance of the trained model, such as accuracy, recall rate, F1 score, etc., is evaluated on the validation set. The model parameters or training strategy are adjusted according to the evaluation results, and finally a target model that meets the preset performance requirements is obtained, that is, the above-mentioned speech scoring model.

[0162] The target model is used as the speech scoring model.

[0163] The present application obtains a second iterative speech set corresponding to the pre-trained language model; then scores the speech set based on a preset scoring strategy to obtain a speech score corresponding to the second iterative speech set; then builds a corresponding speech score data set based on the speech score and the second iterative speech set; subsequently calls a preset initial model; and trains and evaluates the initial model based on the speech score data set to obtain a target model that meets preset performance requirements; finally, the target model is used as the speech scoring model.

[0164] In some optional implementations of this embodiment, the scoring process of the speech set based on a preset scoring strategy to obtain a speech score corresponding to the second iterative speech set includes the following steps:

[0165] Obtain preset scoring dimensions; wherein the scoring dimensions include at least language logic, language fluency, compliance of speech, persuasiveness of speech and rationality of case.

[0166] In this embodiment, the scoring dimensions are determined by organizing business experts, linguists and data analysts to discuss together. Specifically, the scoring dimensions include at least language logic (checking the logical coherence between sentences), language fluency (evaluating the fluency and naturalness of sentences), compliance of speech (ensuring that the speech complies with industry norms and laws and regulations), persuasiveness of speech (evaluating the attractiveness and persuasiveness of speech to potential customers), and case rationality (checking whether the cited case is appropriate and relevant to the product).

[0167] A preset weight generation strategy is used to generate dimension weights corresponding to the scoring dimensions.

[0168] In this embodiment, according to the importance of each dimension, a weight is set for each scoring dimension by using an expert scoring method or a hierarchical analysis method, etc. The weight should ensure the comprehensiveness and fairness of the scoring system.

[0169] Get the specified speech in the speech set.

[0170] In this embodiment, the above-mentioned specified speech is any one of all the speech contained in the speech set.

[0171] Get the dimension score of the specified speech under the scoring dimension.

[0172] In this embodiment, during the iteration of the pre-trained language model, business personnel are regularly invited to provide feedback on the generated speech and record their comments. Then, based on the feedback from the business personnel, the generated speech is scored and the specific score of each scoring dimension is recorded. A spreadsheet or a dedicated scoring system can be used to record this information.

[0173] The dimension score of the designated speech under the scoring dimension is scored based on a preset scoring formula, and the dimension weight corresponding to the scoring dimension is calculated and processed to obtain the designated speech score of the designated speech.

[0174] In this embodiment, the above scoring formula includes: Where j represents the jth rating dimension, w j is the dimension weight of the j-th scoring dimension, S ikj It is the speech score of product i in the jth dimension under scenario k.

[0175] The present application obtains preset scoring dimensions; wherein the scoring dimensions include at least language logic, language fluency, compliance of speech, persuasiveness of speech and rationality of case; then generates dimension weights corresponding to the scoring dimensions based on a preset weight generation strategy; then obtains the specified speech in the speech set; subsequently obtains the dimension score of the specified speech under the scoring dimension; finally, based on a preset scoring formula, the dimension score of the specified speech under the scoring dimension and the dimension weight corresponding to the scoring dimension are calculated and processed to obtain the specified speech score of the specified speech. In this way, the present application calculates and processes the dimension score of the specified speech under the scoring dimension and the dimension weight corresponding to the scoring dimension based on the use of a scoring formula, so as to realize the rapid and accurate generation of the specified speech score of the specified speech, thereby improving the calculation efficiency of the speech score and ensuring the data accuracy of the obtained speech score.

[0176] In some optional implementations, the user information obtained is subject to the user's consent and complies with relevant laws and policies.

[0177] This application combines a variety of technical methods: OCR technology, large language model and BERT model to construct a method for generating and evaluating insurance speech, which makes up for the lack of large models in the vertical field of insurance speech. In addition, combined with multi-source data sets, it integrates product information, product posters, speech library, and business feedback compliance library, enriching the information input of the large language model, and can fully guarantee the richness of the speech, the accuracy of the information, and the compliance of the speech. In addition, based on actual business feedback, a speech-scoring evaluation system and model were constructed, which can effectively solve the pain points of large language models that are difficult to evaluate, and can score product speech in different scenarios in a targeted manner, and can better guide the subsequent fine-tuning of the speech generation of the large language model.

[0178] In addition, the non-Company software tools or components that appear in the embodiments of the present application are merely examples and do not represent actual use.

[0179] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0180] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned product conversion data, the above-mentioned product conversion data can also be stored in a node of a blockchain.

[0181] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, 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. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0182] The embodiments of the present application can acquire and process relevant data 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.

[0183] 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.

[0184] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0185] 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.

[0186] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a speech generation 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.

[0187] like Figure 3 As shown, the artificial intelligence-based speech generation device 300 described in this embodiment includes: a first judgment module 301, an extraction module 302, a first calling module 303, a processing module 304, an adjustment module 305 and a return module 306. Among them:

[0188] The first judgment module 301 is used to judge whether a speech generation request input by a user is received; wherein the speech generation request carries insurance product information and scenario description information;

[0189] An extraction module 302 is used to extract the insurance product information and the scenario description information from the speech generation request;

[0190] The first calling module 303 is used to call a preset speech generation model; wherein the speech generation model is a model generated by training and fine-tuning a pre-trained language model based on a preset fine-tuning sample set and a fine-tuning algorithm;

[0191] A processing module 304 is used to process the insurance product information and the scenario description information based on the speech generation model to obtain a corresponding speech text;

[0192] An adjustment module 305 is used to adjust the speech text based on a preset adjustment strategy to obtain a corresponding target speech text;

[0193] The return module 306 is used to return the target speech text to the user.

[0194] In some optional implementations of this embodiment, the adjustment module 305 includes:

[0195] The first verification submodule is used to perform word verification on the speech text to obtain a corresponding word verification result;

[0196] A first adjustment submodule, configured to adjust the speech text based on the word verification result to obtain a corresponding first speech text;

[0197] A checking submodule, used for performing a compliance check on the first speech text to obtain a corresponding compliance check result;

[0198] A second adjustment submodule, configured to adjust the first speech text based on the compliance check result to obtain a corresponding second speech text;

[0199] An analysis submodule, used for performing sentiment analysis on the second speech text to obtain a corresponding sentiment analysis result;

[0200] A third adjustment submodule, configured to adjust the second speech text based on the sentiment analysis result to obtain a corresponding third speech text;

[0201] A determination submodule is used to use the third speech text as the target speech text.

[0202] In some optional implementations of this embodiment, the third adjustment submodule includes:

[0203] A judgment unit, used to judge whether the content of the sentiment analysis result is negative;

[0204] A first replacement unit is used for, if yes, performing a word replacement process on the second speech text to obtain a corresponding first designated speech text;

[0205] A second replacement unit is used to perform sentence replacement processing on the first specified speech text to obtain a corresponding second specified speech text;

[0206] A determination unit is used to use the second specified speech text as the third speech text.

[0207] In some optional implementations of this embodiment, the speech generation device based on artificial intelligence further includes:

[0208] A first acquisition module, used to acquire a first iterative speech set corresponding to the pre-trained language model;

[0209] A second acquisition module is used to acquire a preset general question-answer pair;

[0210] A first construction module, configured to construct the corresponding fine-tuning sample set based on the first iterative speech set and the general question-answer pair;

[0211] A second calling module, used to call the pre-trained language model;

[0212] A second construction module is used to train and fine-tune the pre-trained language model using the fine-tuning sample set based on a low-rank adaptive fine-tuning algorithm to obtain a specified language model that meets preset construction conditions;

[0213] The first determination module is used to use the specified language model as the speech generation model.

[0214] In some optional implementations of this embodiment, the speech generation device based on artificial intelligence further includes:

[0215] The third calling module is used to call the preset speech scoring model;

[0216] A first generating module, configured to generate a target speech score corresponding to the target speech text based on a preset speech scoring model;

[0217] A second judgment module is used to judge whether the target speech score is greater than a preset score threshold;

[0218] The fourth calling module is used to call a preset initial speech library if yes;

[0219] The second generation module is used to store the target speech text into the initial speech library to generate a corresponding high-quality speech library.

[0220] In some optional implementations of this embodiment, the speech generation device based on artificial intelligence further includes:

[0221] A third acquisition module is used to acquire a second iterative speech set corresponding to the pre-trained language model;

[0222] A scoring module, used for scoring the speech set based on a preset scoring strategy to obtain a speech score corresponding to the second iterative speech set;

[0223] A third construction module is used to construct a corresponding speech score data set based on the speech score and the second iterative speech set;

[0224] The fifth calling module is used to call the preset initial model;

[0225] A fourth construction module is used to train and evaluate the initial model based on the speech score data set to obtain a target model that meets the preset performance requirements;

[0226] The second determination module is used to use the target model as the speech scoring model.

[0227] In some optional implementations of this embodiment, the scoring module includes:

[0228] The first acquisition submodule is used to acquire preset scoring dimensions; wherein the scoring dimensions at least include language logic, language fluency, compliance of speech, persuasiveness of speech and rationality of case;

[0229] A generating submodule, configured to generate dimension weights corresponding to the scoring dimensions according to a preset weight generating strategy;

[0230] A second acquisition submodule is used to acquire a specified speech in the speech set; wherein the specified speech is any one of all the speech included in the speech set;

[0231] The third acquisition submodule is used to obtain the dimension score of the specified speech under the scoring dimension;

[0232] The calculation submodule is used to calculate the dimension score of the specified speech under the scoring dimension based on a preset scoring formula, and the dimension weight corresponding to the scoring dimension to obtain the specified speech score of the specified speech.

[0233] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0234] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, 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.

[0235] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0236] The memory 41 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 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 4. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions based on the artificial intelligence speech generation method. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0237] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions stored in the memory 41 or process data, such as running computer-readable instructions of the artificial intelligence-based speech generation method.

[0238] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0239] 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 generation method.

[0240] 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, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0241] 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 method for generating speech based on artificial intelligence, characterized in that: The steps include: Determine whether a speech generation request input by a user is received; wherein the speech generation request carries insurance product information and scenario description information; If yes, extract the insurance product information and the scenario description information from the speech generation request; Calling a preset speech generation model; wherein the speech generation model is a model generated by training and fine-tuning a pre-trained language model based on a preset fine-tuning sample set and a fine-tuning algorithm; Processing the insurance product information and the scenario description information based on the speech generation model to obtain corresponding speech text; Adjust the speech text based on a preset adjustment strategy to obtain a corresponding target speech text; The target speech text is returned to the user.

2. The method for generating speech based on artificial intelligence according to claim 1, characterized in that: The step of adjusting the speech text based on a preset adjustment strategy to obtain a corresponding target speech text specifically includes: Performing word verification on the speech text to obtain corresponding word verification results; Adjust the speech text based on the word verification result to obtain a corresponding first speech text; Performing a compliance check on the first script text to obtain a corresponding compliance check result; Adjusting the first speech text based on the compliance check result to obtain a corresponding second speech text; Performing sentiment analysis on the second speech text to obtain a corresponding sentiment analysis result; Adjust the second speech text based on the sentiment analysis result to obtain a corresponding third speech text; The third speech text is used as the target speech text.

3. The method for generating speech based on artificial intelligence according to claim 2, characterized in that: The step of adjusting the second speech text based on the sentiment analysis result to obtain a corresponding third speech text specifically includes: Determining whether the content of the sentiment analysis result is negative; If yes, perform word replacement processing on the second speech text to obtain the corresponding first specified speech text; Perform sentence replacement processing on the first specified speech text to obtain a corresponding second specified speech text; The second designated speech text is used as the third speech text.

4. The method for generating speech based on artificial intelligence according to claim 1, characterized in that: Before the step of calling the preset speech generation model, the following steps are included: Obtaining a first iteration speech set corresponding to the pre-trained language model; Get the preset general question and answer pairs; Constructing the corresponding fine-tuning sample set based on the first iterative speech set and the general question-answer pair; Calling the pre-trained language model; Based on a low-rank adaptive fine-tuning algorithm, the pre-trained language model is trained and fine-tuned using the fine-tuning sample set to obtain a specified language model that meets preset construction conditions; The specified language model is used as the speech generation model.

5. The method for generating speech based on artificial intelligence according to claim 1, characterized in that: After the step of returning the target speech text to the user, the method further includes: Call the preset speech scoring model; Generate a target speech score corresponding to the target speech text based on a preset speech scoring model; Determine whether the target speech score is greater than a preset score threshold; If yes, call the preset initial script library; The target speech text is stored in the initial speech library to generate a corresponding high-quality speech library.

6. The method for generating speech based on artificial intelligence according to claim 5, characterized in that: Before the step of calling the preset speech scoring model, the method further includes: Obtaining a second iteration speech set corresponding to the pre-trained language model; Scoring the speech set based on a preset scoring strategy to obtain a speech score corresponding to the second iterative speech set; Constructing a corresponding speech score data set based on the speech score and the second iterative speech set; Call the preset initial model; Training and evaluating the initial model based on the speech score data set to obtain a target model that meets preset performance requirements; The target model is used as the speech scoring model.

7. The method for generating speech based on artificial intelligence according to claim 6, characterized in that: The step of scoring the speech set based on a preset scoring strategy to obtain a speech score corresponding to the second iterative speech set specifically includes: Obtaining preset scoring dimensions; wherein the scoring dimensions at least include language logic, language fluency, compliance of speech, persuasiveness of speech, and rationality of case; A preset weight generation strategy is used to generate dimension weights corresponding to the scoring dimensions; Obtaining a specified speech in the speech set; wherein the specified speech is any one of all the speech included in the speech set; Obtain the dimension score of the specified speech under the scoring dimension; The dimension score of the designated speech under the scoring dimension is scored based on a preset scoring formula, and the dimension weight corresponding to the scoring dimension is calculated and processed to obtain the designated speech score of the designated speech.

8. A speech generation device based on artificial intelligence, characterized in that: include: The first judgment module is used to judge whether a speech generation request input by the user is received; wherein the speech generation request carries insurance product information and scenario description information; An extraction module, for extracting the insurance product information and the scenario description information from the speech generation request; A first calling module is used to call a preset speech generation model; wherein the speech generation model is a model generated by training and fine-tuning a pre-trained language model based on a preset fine-tuning sample set and a fine-tuning algorithm; A processing module, used to process the insurance product information and the scenario description information based on the speech generation model to obtain a corresponding speech text; An adjustment module, used to adjust the speech text based on a preset adjustment strategy to obtain a corresponding target speech text; A return module is used to return the target speech text to the user.

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

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 method for generating speech based on artificial intelligence as described in any one of claims 1 to 7.

Images