Delivery control system and method based on large model and sound cloning technology

Abstract

The invention discloses a delivery control system and method based on a large model and sound cloning technology, and relates to the technical field of text delivery. The invention classifies texts that need to be delivered in history, generates templates for each text, and constructs a template database; inputs user data into a text template to automatically generate a delivery text, and constructs a large model for text generation; selects a receiving channel with the highest overall efficiency as a delivery channel, and delivers the text to a delivery target after processing the text according to the type of the delivery channel; generates an evidence chain using data in the delivery process, and determines whether the delivery task is successful based on the generated evidence chain; performs secondary delivery using an optimized delivery channel, and performs secondary judgment on the delivery task; communicates with a delivery target using personalized voice, explains the text content to the delivery target, and provides feedback after the delivery target has clarified the text content.

Description

Technical Field

[0001] The present invention relates to the technical field of text delivery, and in particular to a delivery control system and method based on a large model and sound cloning technology. Background Art

[0002] Traditional legal text delivery mainly relies on manual delivery, mailing or announcement delivery. Legal staff need to deliver the text to the parties by door-to-door or mailing. This method is time-consuming, costly, and difficult to guarantee the success rate of delivery. With the rapid development of information technology, the delivery method of legal texts has undergone a transformation from traditional manual delivery to modern intelligent and automated delivery. In particular, the introduction of technologies such as information technology, big data, and artificial intelligence (AI) in recent years has made the control of legal text delivery more efficient, safe, and standardized. With the popularization of mobile phones, SMS delivery has gradually become an efficient and convenient delivery method, especially for short legal texts such as summonses and reminder notices. When the parties are difficult to contact or evade delivery, the system can publish the legal text on a designated website through a network announcement to complete the delivery procedure. However, with the development of informatization, there are many delivery channels. It is difficult to choose the appropriate delivery channel when delivering the text and make real-time judgments on the delivery process. In addition, many delivery targets do not understand most legal texts. Therefore, it is crucial to improve the participation of delivery targets, make it easier to understand the text, and increase the efficiency of text delivery. Summary of the invention

[0003] The purpose of the present invention is to provide a delivery control system and method based on large models and sound cloning technology to solve the problems raised in the prior art.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A delivery control method based on a large model and sound cloning technology, the method comprising the following steps:

[0006] S100, collecting all texts that need to be delivered in historical work, extracting features of each text, classifying the texts that need to be delivered in history, analyzing the structure of each type of text, generating templates for each type of text and building a template database;

[0007] Furthermore, the specific steps of generating a template for each text and constructing a template database are as follows:

[0008] S101, collect all the texts that need to be delivered in the historical work, and extract the words in each text in the history , Indicates the 1st, 2nd, 3rd, ..., nth word in each text in the extracted history, where n is a positive integer; the number of each word extracted is calculated as ,

[0009] Indicates the number of occurrences of the 1st, 2nd, 3rd, ..., nth word in the calculated text. The number of occurrences of each word in the text is compared, and the word with the most occurrences is used as the text feature. The feature of each text in the history is screened as , Represents the features of the 1st, 2nd, 3rd, ..., nth texts in the collected history. All the texts in the collected history are judged and the texts with the same features are classified into the same category. After classification, the categories of all texts are , represents the 1st, 2nd, 3rd, ..., mth texts obtained by classification, where m is a positive integer;

[0010] S102, after classifying the historical texts, use scanning technology to scan each text image and extract the positions of all words in each text. , Indicates the position of the 1st, 2nd, 3rd, ..., pth word in the extracted text, where p is a positive integer; analyze the position of each word in the text, and when each word is adjacent in the horizontal direction, the adjacent words are judged to be the same part; when there are no adjacent words in the word position, it is judged to be an independent part; after judging all the words in the text, the parts of all the words in the text are obtained. , represents the 1st, 2nd, 3rd, ..., qth word parts obtained after text analysis, where q is a positive integer;

[0011] S103, after analyzing all word parts in the text, extract the word positions of the same type of text h times, and judge the extracted word positions. When the positions of the same words are the same in the h extractions, the corresponding words are judged to be fixed words; when the positions of the words are different in the h extractions, the corresponding words are judged to be replacement words; after judging, the fixed words in each word part in the text are , It means that the 1st, 2nd, 3rd, ..., i-th fixed words in each word part are obtained after judgment, where i is a positive integer; the replacement words for each part of the text are judged to be , It means that the 1st, 2nd, 3rd, ..., jth replacement words in each word part are obtained after judgment, j is a positive integer, and a replacement word set for each text is constructed using all replacement word types; all word parts and corresponding fixed words obtained by text analysis are combined to construct a template for each text, and then all templates are stored to obtain a template database.

[0012] S200, collecting user data of the delivery target and text features to be generated in real time, using the text features to determine the type of text to be generated, and selecting a template corresponding to the text type in the template database, inputting the user data into the text template to automatically generate the delivery text, and constructing a large model for text generation;

[0013] After classifying the text content in the history, a template for each type of text is generated, and the large model for automatic text generation is obtained through training; using the large model, only replacement words in the user data need to be extracted for filling, and the required text is automatically generated; replacing a large amount of manual writing work, saving time and labor costs, and automatically generated text avoids typos, grammatical errors or logical inconsistencies that may occur in manual writing, ensuring the consistency and accuracy of the output. In information-intensive scenarios, the automatic generation system can quickly summarize data and information and generate meaningful reports.

[0014] Furthermore, the specific steps for constructing a large text generation model are:

[0015] S201, collecting the features of the text to be generated in real time as Ts, using the collected real-time text features to search for a text template with the same features in a template database, and after finding the text template requiring the real-time text features, extracting a replacement word set in the text template;

[0016] S202: Real-time collection of user data delivered to the target: , represents the first, second, third, ..., vth user data of the collected delivery target, where v is a positive integer; using the word types in the extracted replacement word set to extract the same type of data in the collected user data to obtain input words for generating text, extracting the position of each replacement word in S103, marking the input word using the word part position of the replacement word in the text, and setting the marked input word to , It indicates the 1st, 2nd, 3rd, ..., gth input data after marking. Where B represents the word part of the input data in the text, and W represents the specific position of the input data in the word part of the text; the marked input data is The location is input into the search text template to automatically generate real-time delivery text;

[0017] S203. Use machine learning to train S201 classification real-time feature search and S202 text generation. After training, a large text generation model is obtained. The large text generation model is used to automatically generate text that needs to be generated in real time during work.

[0018] S300, collecting all information receiving channels used by the delivery target in history, calculating the comprehensive efficiency of each receiving channel in the delivery target history according to the number of times each receiving channel is used, the number of successful receptions, and the receiving time, selecting the receiving channel with the highest comprehensive efficiency as the delivery channel, and processing the text according to the type of the delivery channel and delivering it to the delivery target;

[0019] Furthermore, the specific steps of processing the text according to the type of delivery channel and then delivering it to the delivery target are as follows:

[0020] S301, collect all the information receiving channels used by the delivery target in the history , Indicates the first, second, third, ..., rth information receiving channel used by the delivery target in the collected history, where r is a positive integer; corresponding to the usage data of each information receiving channel by the delivery target in the collected history;

[0021] S302, extracting the usage times, successful reception times and reception time of each information receiving channel from the collected usage data of each information receiving channel, and calculating the comprehensive efficiency of each information receiving channel, the formula is:

[0022] ;

[0023] In the formula, Ef represents the comprehensive efficiency of each information receiving channel calculated, C represents the number of times each information receiving channel is used. represents the total number of times all information receiving channels are used, Su represents the number of successful receptions after each information receiving channel is used, and Tt represents the receiving time of each information receiving channel. After calculation, the comprehensive efficiency of each information receiving channel is , Indicates the calculated comprehensive efficiency of the first, second, third, ..., r information receiving channels; compares the r comprehensive efficiencies and selects the information receiving channel with the largest comprehensive efficiency as the delivery channel for the delivered text;

[0024] S303. After selecting the delivery channel for the delivery text, data processing is performed on the delivery text according to the transmission type of the delivery channel, and after converting it into an information type suitable for delivery, the delivery text is delivered.

[0025] Predict the most effective delivery method and time based on the target's behavioral preferences, historical response time, and location, thereby improving delivery success and efficiency;

[0026] S400, recording all data in the text delivery process in real time, using the data in the delivery process to generate an evidence chain, and judging whether the delivery task is successful based on the generated evidence chain;

[0027] Furthermore, the specific steps to determine whether the delivery task is successful based on the generated evidence chain are as follows:

[0028] S401: Analyze the entire delivery process of the delivery text, record each processing and analysis node of the delivery text, and build a delivery task chain of text generation-channel selection-text conversion-sending-delivery node - receive feedback, Represents each node reached during the text delivery process, Indicates delivery to the target node;

[0029] S402, when executing the delivery task for the delivery text, record the time of each step in the delivery task chain, mark the delivery task chain with a timestamp to generate an evidence chain; collect the time Tj when the information receiving channel selected in the history successfully receives the information, and calculate the information delivery time threshold, the formula is: , where Ty represents the time threshold of the information receiving channel. It indicates the average time when the selected information receiving channel successfully receives information in the collected history. It represents the standard deviation of the time when the selected information receiving channel successfully receives the information in the collected history;

[0030] S403, when the text is delivered, the timestamp of the sending stage in the evidence chain is extracted as Tf, the real-time collection time is Tm, and the delivery task is judged by the information delivery time threshold. If no feedback is received, the delivery task is considered to have failed and the delivery target is the received text. When feedback is received, it is determined that the delivery task is successful and the delivery target receives the text.

[0031] The system has a built-in full-process tracking function, which automatically generates a chain of evidence for the delivery process, including the time of each voice delivery, the recipient's feedback, the record of the voice content and the switching process of the delivery method, to ensure the traceability of the delivery; all delivery behaviors and feedback information are recorded securely and unalterably.

[0032] S500: When it is determined that the delivery task has failed, the evidence chain is used to trace the failure of the delivery task, and the text delivery channel is fully optimized according to the tracing result. The optimized delivery channel is used for secondary delivery, and the delivery task is re-judged;

[0033] Furthermore, the specific steps for making a secondary judgment on the delivery task are as follows:

[0034] S501. When it is determined that the delivery task has failed, extract the evidence chain in the delivery process, check the evidence chain for source tracing, extract the delivery node in the evidence chain, and use the last delivery node recorded as the source point of the delivery failure;

[0035] S502, screening in S302 in descending order of comprehensive efficiency, checking the delivery nodes that need to be passed in each information receiving channel, extracting the information receiving channels without the source of delivery failure in the delivery channels, and selecting the information receiving channels with the highest comprehensive efficiency among the extracted information receiving channels as the optimized delivery channel;

[0036] S503, using the optimized delivery channel to deliver the delivery text, and performing a secondary judgment on the optimized delivery task according to the method in S403.

[0037] By using the delivery nodes in the evidence chain, we can clearly trace the cause of failure and obtain the source of failure. We can use the source of failure to optimize the delivery channel and conduct a new round of delivery, which greatly enhances the reliability and efficiency of delivery and reduces the time for troubleshooting the cause of failure.

[0038] S600. When it is determined that the delivery task is completed and the text is delivered successfully, the delivery target uses the client to read different types of delivery texts in different ways. The client uses sound cloning technology to generate personalized voice according to the text type, uses the personalized voice to communicate with the delivery target, explains the text content to the delivery target, and the delivery target provides feedback after clarifying the text content.

[0039] Furthermore, the specific steps for providing feedback after the target text content is clearly defined are:

[0040] S601. When it is determined that the delivery task is completed and the text is delivered successfully, the client collects the type of the delivered text in real time, and reads the delivered text using a reading method corresponding to the type. After reading the text, the client generates a personalized voice according to the text type using voice cloning technology, and communicates with the delivery target using the personalized voice. During the communication process, the transformer algorithm is used to process the questions raised by the user, and the personalized voice answers the questions raised by the user.

[0041] S602: When the user understands the text content and responds according to the text content, user feedback is provided through the client to complete the delivery task.

[0042] When the system delivers notifications via voice, users can conduct real-time interactive inquiries through voice or text input to understand text details or supplement relevant information. This multimodal interactive method can make the delivery process more efficient and flexible, and enhance the parties' sense of participation. Using voice cloning technology to generate different types of voices according to the type of text, serious voices are generated when the text is important and urgent, and relaxed voices are generated when the text is non-urgent, so that the delivery target can personally feel the content of the text and accelerate the delivery target's understanding of the text content.

[0043] The delivery control system based on large model and voice cloning technology includes data collection module, text generation module, channel selection module, task judgment module, optimization module and personalized voice module;

[0044] The data collection module is used to collect information receiving channels and usage data used by the delivery target in history;

[0045] The text generation module is used to utilize the texts generated in the historical work, classify the texts, construct templates for each type of text, and automatically generate texts using the text templates;

[0046] The channel selection module is used to analyze the information receiving channels used by the delivery target in the collected history, calculate the comprehensive efficiency, and select the delivery channel for the delivered text;

[0047] The task judgment module is used to record the delivery process to generate a timestamp to form an evidence chain, and calculate the information reception time threshold based on the historical reception time, and use the time threshold and the evidence chain to determine whether the delivery task is successful;

[0048] The optimization module is used to trace the failure by using the evidence chain after determining that the delivery task has failed, obtain the source of the delivery failure, analyze the source of the delivery failure, and optimize the delivery channel to re-deliver;

[0049] The personalized voice module is used to generate personalized voice according to the text type using sound cloning technology after the text is delivered, and communicate with the delivery target using the personalized voice. During the communication process, the transformer algorithm is used to process the questions raised by the user, and the personalized voice answers the questions raised by the user.

[0050] The text generation module includes a classification unit, a template construction unit and a text generation large model unit;

[0051] The classification unit is used to extract the features of all texts generated in history and classify the texts according to the features of each text;

[0052] The template construction unit is used to scan the text, find the word parts, fixed words and replacement words of the text, and construct a template for each text;

[0053] The text generation model unit is used to collect user data of the delivery target, extract replacement words in the user data as input words, mark the position of the input words, input them into the template to automatically generate text, and use machine learning training to generate a text generation model.

[0054] The task judgment module includes a delivery task chain unit, an evidence chain unit, a threshold unit and a judgment unit;

[0055] The delivery task chain unit is used to analyze the delivery process and generate a delivery task chain;

[0056] The evidence chain unit is used to extract the time of delivery to each node in the task chain, generate a timestamp mark, and construct an evidence chain;

[0057] The threshold unit is used to calculate the information delivery time threshold according to the time when the transport channel receives the information in history;

[0058] The judgment unit is used to extract the timestamp and real time of the sending stage in the evidence chain, and use the time threshold to judge the delivery task.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] 1. The present invention generates a template for each text after classifying the text content in the history, and trains to obtain a large model for automatic text generation; using the large model, only replacement words in the user data need to be extracted for filling, and the required text is automatically generated; replacing a large amount of manual writing work, saving time and labor costs, and the automatically generated text avoids typos, grammatical errors or logical inconsistencies that may occur in manual writing, ensuring the consistency and accuracy of the output.

[0061] 2. The present invention automatically generates a chain of evidence for the delivery process, including the time of each voice delivery, the recipient's feedback, the recording of the voice content, and the switching process of the delivery method, to ensure the legality and traceability of the delivery; all delivery behaviors and feedback information are recorded securely and unalterably as evidence in future litigation or disputes, ensuring the transparency and fairness of legal delivery.

[0062] 3. The user of the present invention can conduct real-time interactive query through voice or text input to understand text details or supplement relevant information. The use of multi-modal interactive methods can make the delivery process more efficient and flexible, and enhance the sense of participation of the parties. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a module distribution diagram of the delivery control system based on the large model and sound cloning technology of the present invention;

[0064] Figure 2 A schematic diagram of the steps of the delivery control method based on a large model and sound cloning technology of the present invention;

[0065] Figure 3 It is a schematic diagram of a text template of the delivery control method based on a large model and sound cloning technology of the present invention. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] Example: Figure 1-Figure 3 As shown, the present invention provides a technical solution.

[0068] A delivery control method based on a large model and sound cloning technology, the method comprising the following steps:

[0069] S100, collecting all texts that need to be delivered in historical work, extracting features of each text, classifying the texts that need to be delivered in history, analyzing the structure of each type of text, generating templates for each type of text and building a template database;

[0070] The specific steps to generate templates for each text and build a template database are:

[0071] S101, collect all the texts that need to be delivered in the historical work, and extract the words in each text in the history , Indicates the 1st, 2nd, 3rd, ..., nth word in each text in the extracted history, where n is a positive integer; the number of each word extracted is calculated as ,

[0072] Indicates the number of occurrences of the 1st, 2nd, 3rd, ..., nth word in the calculated text. The number of occurrences of each word in the text is compared, and the word with the most occurrences is used as the text feature. The feature of each text in the history is screened as , Represents the features of the 1st, 2nd, 3rd, ..., nth texts in the collected history. All the texts in the collected history are judged and the texts with the same features are classified into the same category. After classification, the categories of all texts are , represents the 1st, 2nd, 3rd, ..., mth texts obtained by classification, where m is a positive integer;

[0073] S102, after classifying the historical texts, use scanning technology to scan each text image and extract the positions of all words in each text. , Indicates the position of the 1st, 2nd, 3rd, ..., pth word in the extracted text, where p is a positive integer; analyze the position of each word in the text, and when each word is adjacent in the horizontal direction, the adjacent words are judged to be the same part; when there are no adjacent words in the word position, it is judged to be an independent part; after judging all the words in the text, the parts of all the words in the text are obtained. , represents the 1st, 2nd, 3rd, ..., qth word parts obtained after text analysis, where q is a positive integer;

[0074] S103, after analyzing all word parts in the text, extract the word positions of the same type of text h times, and judge the extracted word positions. When the positions of the same words are the same in the h extractions, the corresponding words are judged to be fixed words; when the positions of the words are different in the h extractions, the corresponding words are judged to be replacement words; after judging, the fixed words in each word part in the text are , It means that the 1st, 2nd, 3rd, ..., i-th fixed words in each word part are obtained after judgment, where i is a positive integer; the replacement words for each part of the text are obtained after judgment. , It means that the 1st, 2nd, 3rd, ..., jth replacement words in each word part are obtained after judgment, j is a positive integer, and a replacement word set for each text is constructed using all replacement word types; all word parts and corresponding fixed words obtained by text analysis are combined to construct a template for each text, and then all templates are stored to obtain a template database.

[0075] S200, collecting user data of the delivery target and text features to be generated in real time, using the text features to determine the type of text to be generated, and selecting a template corresponding to the text type in the template database, inputting the user data into the text template to automatically generate the delivery text, and constructing a large model for text generation;

[0076] After classifying the text content in the history, a template for each type of text is generated, and the large model for automatic text generation is obtained through training; using the large model, only replacement words in the user data need to be extracted for filling, and the required text is automatically generated; replacing a large amount of manual writing work, saving time and labor costs, and automatically generated text avoids typos, grammatical errors or logical inconsistencies that may occur in manual writing, ensuring the consistency and accuracy of the output. In information-intensive scenarios, the automatic generation system can quickly summarize data and information and generate meaningful reports.

[0077] The specific steps to build a large text generation model are:

[0078] S201, collecting the features of the text to be generated in real time as Ts, using the collected real-time text features to search for a text template with the same features in a template database, and after finding the text template requiring the real-time text features, extracting a replacement word set in the text template;

[0079] S202: Real-time collection of user data delivered to the target: , represents the first, second, third, ..., vth user data of the collected delivery target, where v is a positive integer; using the word types in the extracted replacement word set to extract the same type of data in the collected user data to obtain input words for generating text, extracting the position of each replacement word in S103, marking the input word using the word part position of the replacement word in the text, and setting the marked input word to , It indicates the 1st, 2nd, 3rd, ..., gth input data after marking. Where B represents the word part of the input data in the text, and W represents the specific position of the input data in the word part of the text; the marked input data is The location is input into the search text template to automatically generate real-time delivery text;

[0080] S203. Use machine learning to train S201 classification real-time feature search and S202 text generation. After training, a large text generation model is obtained. The large text generation model is used to automatically generate text that needs to be generated in real time during work.

[0081] S300, collecting all information receiving channels used by the delivery target in history, calculating the comprehensive efficiency of each receiving channel in the delivery target history according to the number of times each receiving channel is used, the number of successful receptions, and the receiving time, selecting the receiving channel with the highest comprehensive efficiency as the delivery channel, and processing the text according to the type of the delivery channel and delivering it to the delivery target;

[0082] The specific steps for processing the text and delivering it to the destination according to the type of delivery channel are as follows:

[0083] S301, collect all the information receiving channels used by the delivery target in the history , Indicates the first, second, third, ..., rth information receiving channel used by the delivery target in the collected history, where r is a positive integer; corresponding to the usage data of each information receiving channel by the delivery target in the collected history;

[0084] S302, extracting the usage times, successful reception times and reception time of each information receiving channel from the collected usage data of each information receiving channel, and calculating the comprehensive efficiency of each information receiving channel, the formula is:

[0085] ;

[0086] In the formula, Ef represents the comprehensive efficiency of each information receiving channel calculated, C represents the number of times each information receiving channel is used. represents the total number of times all information receiving channels are used, Su represents the number of successful receptions after each information receiving channel is used, and Tt represents the receiving time of each information receiving channel. After calculation, the comprehensive efficiency of each information receiving channel is , Indicates the calculated comprehensive efficiency of the first, second, third, ..., r information receiving channels; compares the r comprehensive efficiencies and selects the information receiving channel with the largest comprehensive efficiency as the delivery channel for the delivered text;

[0087] S303. After selecting the delivery channel for the delivery text, data processing is performed on the delivery text according to the transmission type of the delivery channel, and after converting it into an information type suitable for delivery, the delivery text is delivered.

[0088] Predict the most effective delivery method and time based on the target's behavioral preferences, historical response time, and location, thereby improving delivery success and efficiency;

[0089] S400, recording all data in the text delivery process in real time, using the data in the delivery process to generate an evidence chain, and judging whether the delivery task is successful based on the generated evidence chain;

[0090] The specific steps to determine whether the delivery task is successful based on the generated evidence chain are:

[0091] S401: Analyze the entire delivery process of the delivery text, record each processing and analysis node of the delivery text, and build a delivery task chain of text generation-channel selection-text conversion-sending-delivery node - receive feedback, Represents each node reached during the text delivery process, Indicates delivery to the target node;

[0092] S402, when executing the delivery task for the delivery text, record the time of each step in the delivery task chain, mark the delivery task chain with a timestamp to generate an evidence chain; collect the time Tj when the information receiving channel selected in the history successfully receives the information, and calculate the information delivery time threshold, the formula is: , where Ty represents the time threshold of the information receiving channel. It indicates the average time when the selected information receiving channel successfully receives information in the collected history. It represents the standard deviation of the time when the selected information receiving channel successfully receives the information in the collected history;

[0093] S403, when the text is delivered, the timestamp of the sending stage in the evidence chain is extracted as Tf, the real-time collection time is Tm, and the delivery task is judged by the information delivery time threshold. If no feedback is received, the delivery task is considered to have failed and the delivery target is the received text. When feedback is received, it is determined that the delivery task is successful and the delivery target receives the text.

[0094] The system has a built-in full-process tracking function, which automatically generates a chain of evidence for the delivery process, including the time of each voice delivery, the recipient's feedback, the record of the voice content and the switching process of the delivery method, to ensure the legality and traceability of the delivery; all delivery actions and feedback information are recorded securely and unalterably as evidence in future litigation or disputes, ensuring the transparency and fairness of legal delivery.

[0095] S500: When it is determined that the delivery task has failed, the evidence chain is used to trace the failure of the delivery task, and the text delivery channel is fully optimized according to the tracing result. The optimized delivery channel is used for secondary delivery, and the delivery task is re-judged;

[0096] The specific steps for secondary judgment of the delivery task are as follows:

[0097] S501. When it is determined that the delivery task has failed, extract the evidence chain in the delivery process, check the evidence chain for source tracing, extract the delivery node in the evidence chain, and use the last delivery node recorded as the source point of the delivery failure;

[0098] S502, screening in S302 in descending order of comprehensive efficiency, checking the delivery nodes that need to be passed in each information receiving channel, extracting the information receiving channels without the source of delivery failure in the delivery channels, and selecting the information receiving channels with the highest comprehensive efficiency among the extracted information receiving channels as the optimized delivery channel;

[0099] S503, using the optimized delivery channel to deliver the delivery text, and performing a secondary judgment on the optimized delivery task according to the method in S403.

[0100] By using the delivery nodes in the evidence chain, we can clearly trace the cause of failure and obtain the source of failure. We can use the source of failure to optimize the delivery channel and conduct a new round of delivery, which greatly enhances the reliability and efficiency of delivery and reduces the time for troubleshooting the cause of failure.

[0101] S600. When it is determined that the delivery task is completed and the text is delivered successfully, the delivery target uses the client to read different types of delivery texts in different ways. The client uses sound cloning technology to generate personalized voice according to the text type, uses the personalized voice to communicate with the delivery target, explains the text content to the delivery target, and the delivery target provides feedback after clarifying the text content.

[0102] The specific steps for providing feedback after the target text content is clearly defined are:

[0103] S601. When it is determined that the delivery task is completed and the text is delivered successfully, the client collects the type of the delivered text in real time, and reads the delivered text using a reading method corresponding to the type. After reading the text, the client generates a personalized voice according to the text type using voice cloning technology, and communicates with the delivery target using the personalized voice. During the communication process, the transformer algorithm is used to process the questions raised by the user, and the personalized voice answers the questions raised by the user.

[0104] S602: When the user understands the text content and responds according to the text content, user feedback is provided through the client to complete the delivery task.

[0105] When the system delivers notifications via voice, users can conduct real-time interactive inquiries through voice or text input to understand text details or supplement relevant information. This multimodal interactive method can make the delivery process more efficient and flexible, and enhance the parties' sense of participation. Using voice cloning technology to generate different types of voices according to the type of text, serious voices are generated when the text is important and urgent, and relaxed voices are generated when the text is non-urgent, so that the delivery target can personally feel the content of the text and accelerate the delivery target's understanding of the text content.

[0106] The delivery control system based on large model and voice cloning technology includes data collection module, text generation module, channel selection module, task judgment module, optimization module and personalized voice module;

[0107] The data collection module is used to collect information receiving channels and usage data used by the delivery target in history;

[0108] The text generation module is used to utilize the texts generated in the historical work, classify the texts, construct templates for each type of text, and automatically generate texts using the text templates;

[0109] The channel selection module is used to analyze the information receiving channels used by the delivery target in the collected history, calculate the comprehensive efficiency, and select the delivery channel for the delivered text;

[0110] The task judgment module is used to record the delivery process to generate a timestamp to form an evidence chain, and calculate the information reception time threshold based on the historical reception time, and use the time threshold and the evidence chain to determine whether the delivery task is successful;

[0111] The optimization module is used to trace the failure by using the evidence chain after determining that the delivery task has failed, obtain the source of the delivery failure, analyze the source of the delivery failure, and optimize the delivery channel to re-deliver;

[0112] The personalized voice module is used to generate personalized voice according to the text type using sound cloning technology after the text is delivered, and communicate with the delivery target using the personalized voice. During the communication process, the transformer algorithm is used to process the questions raised by the user, and the personalized voice answers the questions raised by the user.

[0113] The text generation module includes a classification unit, a template construction unit and a text generation large model unit;

[0114] The classification unit is used to extract the features of all texts generated in history and classify the texts according to the features of each text;

[0115] The template construction unit is used to scan the text, find the word parts, fixed words and replacement words of the text, and construct a template for each text;

[0116] The text generation model unit is used to collect user data of the delivery target, extract replacement words in the user data as input words, mark the position of the input words, input them into the template to automatically generate text, and use machine learning training to generate a text generation model.

[0117] The task judgment module includes a delivery task chain unit, an evidence chain unit, a threshold unit and a judgment unit;

[0118] The delivery task chain unit is used to analyze the delivery process and generate a delivery task chain;

[0119] The evidence chain unit is used to extract the time of delivery to each node in the task chain, generate a timestamp mark, and construct an evidence chain;

[0120] The threshold unit is used to calculate the information delivery time threshold according to the time when the transport channel receives the information in history;

[0121] The judgment unit is used to extract the timestamp and real time of the sending stage in the evidence chain, and use the time threshold to judge the delivery task.

[0122] Example: Analyze the legal texts produced in history and classify the texts into civil complaints, power of attorney, etc.;

[0123] Take the authorization letter as an example, and generate a text template by scanning to determine the word parts, fixed words and replacement words. Figure 3 As shown, xxx represents a replacement word;

[0124] Extract the replacement words from the user data of the delivery target and input them to automatically generate an "authorization notice";

[0125] The information receiving channels used to deliver the target in the past are mailbox, express delivery, SMS, and telephone voice; the mailbox with the highest comprehensive efficiency is calculated; the authorization notice is delivered by mailbox;

[0126] The calculated delivery time threshold is 30 minutes; the sending time in the extracted evidence chain is 12:00, the real-time time is 13:00, and the judgment number system has not received feedback, so it is judged that the delivery failed;

[0127] By tracing the source of the failure using the chain of evidence, we found that the source of the failure was an incorrect email address. We then reselected telephone voice as the delivery channel and started a new round of delivery. We also converted the text into voice data for transmission.

[0128] After the delivery task is completed, the target receives the delivery text and uses the client to collect the text type, and obtains that the text type is voice. The client uses voice reading to read the "Authorization Notice" and uses sound cloning technology to generate personalized voice to explain the "Authorization Notice Content" to the target, helping the target to understand the text. When the target understands and executes according to the text, user feedback is provided in the client to complete the delivery.

[0129] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A delivery control method based on a large model and sound cloning technology, characterized in that: include: S100, collecting all texts that need to be delivered in historical work, extracting features of each text, classifying the texts that need to be delivered in history, analyzing the structure of each type of text, generating templates for each type of text and building a template database; S200, collecting user data of the delivery target and text features to be generated in real time, using the text features to determine the type of text to be generated, and selecting a template corresponding to the text type in the template database, inputting the user data into the text template to automatically generate the delivery text, and constructing a large model for text generation; S300, collecting all information receiving channels used by the delivery target in history, calculating the comprehensive efficiency of each receiving channel in the delivery target history according to the number of times each receiving channel is used, the number of successful receptions, and the receiving time, selecting the receiving channel with the highest comprehensive efficiency as the delivery channel, and processing the text according to the type of the delivery channel and delivering it to the delivery target; S400, recording all data in the text delivery process in real time, using the data in the delivery process to generate an evidence chain, and judging whether the delivery task is successful based on the generated evidence chain; S500: When it is determined that the delivery task has failed, the evidence chain is used to trace the failure of the delivery task, and the text delivery channel is fully optimized according to the tracing result. The optimized delivery channel is used for secondary delivery, and the delivery task is re-judged; S600. When it is determined that the delivery task is completed and the text is delivered successfully, the delivery target uses the client to read different types of delivery texts in different ways. The client uses sound cloning technology to generate personalized voice according to the text type, uses the personalized voice to communicate with the delivery target, explains the text content to the delivery target, and the delivery target provides feedback after clarifying the text content.

2. The delivery control method based on large model and sound cloning technology according to claim 1 is characterized in that: The specific steps of generating a template for each text and constructing a template database in S100 are: S101, collect all the texts that need to be delivered in the historical work, and extract the words in each text in the history , Indicates the 1st, 2nd, 3rd, ..., nth word in each text in the extracted history, where n is a positive integer; the number of each word extracted is calculated as , Indicates the number of occurrences of the 1st, 2nd, 3rd, ..., nth word in the calculated text. The number of occurrences of each word in the text is compared, and the word with the most occurrences is used as the text feature. The feature of each text in the history is screened as , Represents the features of the 1st, 2nd, 3rd, ..., nth texts in the collected history. All the texts in the collected history are judged and the texts with the same features are classified into the same category. After classification, the categories of all texts are , represents the 1st, 2nd, 3rd, ..., mth texts obtained by classification, where m is a positive integer; S102, after classifying the historical texts, use scanning technology to scan each text image and extract the positions of all words in each text. , Indicates the position of the 1st, 2nd, 3rd, ..., pth word in the extracted text, where p is a positive integer; analyze the position of each word in the text, and when each word is adjacent in the horizontal direction, the adjacent words are judged to be the same part; when there are no adjacent words in the word position, it is judged to be an independent part; after judging all the words in the text, the parts of all the words in the text are obtained. , represents the 1st, 2nd, 3rd, ..., qth word parts obtained after text analysis, where q is a positive integer; S103, after all word parts in the text are analyzed, the word positions of the same type of text are extracted h times, and the extracted word positions are judged. When the positions of the same words in the h extractions are the same, the corresponding words are judged to be fixed words; when the positions of the words in the h extractions are different, the corresponding words are judged to be replacement words; After judging, the fixed words in each word part of the text are , It means that the 1st, 2nd, 3rd, ..., i-th fixed words in each word part are obtained after judgment, where i is a positive integer; the replacement words for each part of the text are judged to be , It means that the 1st, 2nd, 3rd, ..., jth replacement words in each word part are obtained after judgment, j is a positive integer, and a replacement word set for each text is constructed using all replacement word types; all word parts and corresponding fixed words obtained by text analysis are combined to construct a template for each text, and then all templates are stored to obtain a template database.

3. The delivery control method based on large model and sound cloning technology according to claim 2 is characterized in that: The specific steps of constructing the large text generation model in S200 are: S201, collecting the features of the text to be generated in real time as Ts, using the collected real-time text features to search for a text template with the same features in a template database, and after finding the text template requiring the real-time text features, extracting a replacement word set in the text template; S202: Real-time collection of user data delivered to the target: , represents the first, second, third, ..., vth user data of the collected delivery target, where v is a positive integer; using the word types in the extracted replacement word set to extract the same type of data in the collected user data to obtain input words for generating text, extracting the position of each replacement word in S103, marking the input word using the word part position of the replacement word in the text, and setting the marked input word to , It indicates the 1st, 2nd, 3rd, ..., gth input data after marking. Where B represents the word part of the input data in the text, and W represents the specific position of the input data in the word part of the text; the marked input data is The location is input into the search text template to automatically generate real-time delivery text; S203. Use machine learning to train S201 classification real-time feature search and S202 text generation. After training, a large text generation model is obtained. The large text generation model is used to automatically generate text that needs to be generated in real time during work.

4. The delivery control method based on large model and sound cloning technology according to claim 1 is characterized in that: The specific steps of processing the text according to the type of the delivery channel and then delivering it to the delivery target in S300 are: S301, collect all the information receiving channels used by the delivery target in the history , Indicates the first, second, third, ..., rth information receiving channel used by the delivery target in the collected history, where r is a positive integer; corresponding to the usage data of each information receiving channel by the delivery target in the collected history; S302, extracting the usage times, successful reception times and reception time of each information receiving channel from the collected usage data of each information receiving channel, and calculating the comprehensive efficiency of each information receiving channel, the formula is: ； In the formula, Ef represents the comprehensive efficiency of each information receiving channel calculated, C represents the number of times each information receiving channel is used. represents the total number of times all information receiving channels are used, Su represents the number of successful receptions after each information receiving channel is used, and Tt represents the receiving time of each information receiving channel. After calculation, the comprehensive efficiency of each information receiving channel is , Indicates the calculated comprehensive efficiency of the first, second, third, ..., r information receiving channels; compares the r comprehensive efficiencies and selects the information receiving channel with the largest comprehensive efficiency as the delivery channel for the delivered text; S303. After selecting the delivery channel for the delivery text, data processing is performed on the delivery text according to the transmission type of the delivery channel, and after converting it into an information type suitable for delivery, the delivery text is delivered.

5. The delivery control method based on large model and sound cloning technology according to claim 4 is characterized in that: The specific steps of determining whether the delivery task is successful according to the generated evidence chain in S400 are: S401: Analyze the entire delivery process of the delivery text, record each processing and analysis node of the delivery text, and build a delivery task chain of text generation-channel selection-text conversion-sending-delivery node - receive feedback, Represents each node reached during the text delivery process, Indicates delivery to the target node; S402, when executing the delivery task for the delivery text, record the time of each step in the delivery task chain, mark the delivery task chain with a timestamp to generate an evidence chain; collect the time Tj when the information receiving channel selected in the history successfully receives the information, and calculate the information delivery time threshold, the formula is: , where Ty represents the time threshold of the information receiving channel. It indicates the average time when the selected information receiving channel successfully receives information in the collected history. It represents the standard deviation of the time when the selected information receiving channel successfully receives the information in the collected history; S403, when the text is delivered, the timestamp of the sending stage in the evidence chain is extracted as Tf, the real-time collection time is Tm, and the delivery task is judged by the information delivery time threshold. If no feedback is received, the delivery task is considered to have failed and the delivery target is the received text. When feedback is received, it is determined that the delivery task is successful and the delivery target receives the text.

6. The delivery control method based on large model and sound cloning technology according to claim 5 is characterized in that: The specific steps of performing a secondary judgment on the transport task in S500 are: S501. When it is determined that the delivery task has failed, extract the evidence chain in the delivery process, check the evidence chain for source tracing, extract the delivery node in the evidence chain, and use the last delivery node recorded as the source point of the delivery failure; S502, screening in S302 in descending order of comprehensive efficiency, checking the delivery nodes that need to be passed in each information receiving channel, extracting the information receiving channels without the source of delivery failure in the delivery channels, and selecting the information receiving channels with the highest comprehensive efficiency among the extracted information receiving channels as the optimized delivery channel; S503, using the optimized delivery channel to deliver the delivery text, and performing a second judgment on the optimized delivery task according to the method in S403.

7. The delivery control method based on large model and sound cloning technology according to claim 6 is characterized in that: The specific steps of providing feedback after the target clear text content is delivered in S600 are: S601, when it is determined that the delivery task is completed and the text is delivered successfully, the client collects the type of the delivered text in real time, and reads the delivered text using a reading method corresponding to the type. After reading the text, the client generates a personalized voice according to the text type using voice cloning technology, and communicates with the delivery target using the personalized voice. During the communication process, the transformer algorithm is used to process the questions raised by the user, and the personalized voice answers the questions raised by the user; S602: When the user understands the text content and responds according to the text content, user feedback is provided through the client to complete the delivery task.

8. A delivery control system based on large model and sound cloning technology using the delivery control method based on large model and sound cloning technology as described in any one of claims 1 to 7, characterized in that: The delivery control system includes a data collection module, a text generation module, a channel selection module, a task judgment module, an optimization module and a personalized voice module; The data collection module is used to collect information receiving channels and usage data used by the delivery target in history; The text generation module is used to utilize the texts generated in the historical work, classify the texts, construct templates for each type of text, and automatically generate texts using the text templates; The channel selection module is used to analyze the information receiving channels used by the delivery target in the collected history, calculate the comprehensive efficiency, and select the delivery channel for the delivered text; The task judgment module is used to record the delivery process to generate a timestamp to form an evidence chain, and calculate the information reception time threshold based on the historical reception time, and use the time threshold and the evidence chain to determine whether the delivery task is successful; The optimization module is used to trace the failure by using the evidence chain after determining that the delivery task has failed, obtain the source of the delivery failure, analyze the source of the delivery failure, and optimize the delivery channel to re-deliver; The personalized voice module is used to generate personalized voice according to the text type by the client using sound cloning technology after the text is delivered, communicate with the delivery target using the personalized voice, process the questions raised by the user using the transformer algorithm during the communication process, and answer the questions raised by the user with the personalized voice.

9. The delivery control system based on large model and sound cloning technology according to claim 8 is characterized in that: The text generation module includes a classification unit, a template construction unit and a text generation large model unit; The classification unit is used to extract the features of all texts generated in history and classify the texts according to the features of each text; The template construction unit is used to scan the text, find the word parts, fixed words and replacement words of the text, and construct a template for each text; The text generation model unit is used to collect user data of the delivery target, extract replacement words in the user data as input words, mark the position of the input words, input them into the template to automatically generate text, and use machine learning training to generate a text generation model.

10. The delivery control system based on large model and sound cloning technology according to claim 8, characterized in that: The task judgment module includes a delivery task chain unit, an evidence chain unit, a threshold unit and a judgment unit; The delivery task chain unit is used to analyze the delivery process and generate a delivery task chain; The evidence chain unit is used to extract the time of delivery to each node in the task chain, generate a timestamp mark, and construct an evidence chain; The threshold unit is used to calculate the information delivery time threshold according to the time when the transport channel receives the information in history; The judgment unit is used to extract the timestamp and real time of the sending stage in the evidence chain, and use the time threshold to judge the delivery task.

Images