A control method and device for intelligent voice assistant

Abstract

The present invention relates to the field of artificial intelligence technology, and discloses a control method and device for an intelligent voice assistant. When receiving a first voice, the method performs a voice-to-text operation on the first voice in combination with a voice recognition correction model to form a first sentence; performs intent matching on the first sentence in combination with an intention recognition correction model to obtain a first intention, and obtains a first intention processing function corresponding to the first intention; extracts several first parameters related to the first intention in the first sentence in combination with a parameter correction model; executes the first intention processing function according to several first parameters, and obtains a voice response result. The present invention simplifies the interaction process between users and various systems by using voice input; improves the accuracy of intention matching by adopting a correction model; and effectively controls data circulation by calling multiple algorithms in series, thereby breaking data silos and lowering the threshold for use.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a control method and device for an intelligent voice assistant. Background Art

[0002] In the current artificial intelligence interaction, one system is used to implement one function. The various marketing systems currently on the market will cause marketers to fall into a common problem, that is, the systems operate independently and lack synergy. Users need to switch between different systems, which will make the interaction cumbersome and affect the user experience. Moreover, each independent system is limited to a specific application scenario and is aimed at staff from different departments. If these systems are used in series at the same time, a high threshold for use is required, because only when users have a deep understanding of the operation mode and characteristics of each system can they be effectively used in series. Furthermore, each system runs in an independent data environment, making it difficult to share and cross-use information. This leads to limitations in user personalized services, and user needs cannot be seamlessly transmitted and responded to between systems. This data isolation not only limits the breadth of services, but also affects the system's deep understanding of global information, hindering the system's potential to provide more intelligent and personalized services.

[0003] Therefore, the current AI interactive marketing system has problems such as fragmented systems, cumbersome interactions, high barriers to serial use, data silos, and information fragmentation, which urgently require a solution. Summary of the invention

[0004] The present invention provides a control method and device for an intelligent voice assistant, which can optimize the interactive experience by integrating independent systems, lower the threshold for using intelligent services through voice input, and improve the accuracy of voice recognition matching through a correction model.

[0005] In order to solve the above technical problems, the present invention provides a control method of an intelligent voice assistant, comprising:

[0006] When a first voice is received, performing a voice-to-text operation on the first voice in combination with a voice recognition correction model to form a first sentence;

[0007] Performing intent matching on the first sentence in combination with the intent recognition and correction model to obtain a first intent, and acquiring a first intent processing function corresponding to the first intent;

[0008] Extracting a plurality of first parameters related to the first intention in the first sentence in combination with the parameter correction model;

[0009] According to the plurality of the first parameters, the first intention processing function is executed, and a voice response result is obtained.

[0010] Furthermore, when the first voice is received, the voice recognition correction model is combined to perform a voice-to-text operation on the first voice to form a first sentence, which is specifically:

[0011] Using speech recognition technology to perform speech-to-text conversion on the first speech to form a sentence to be processed;

[0012] Convert the text of the sentence to be processed into pinyin to form the pinyin to be processed;

[0013] Compare the pinyin to be processed with the brand mapping table pre-stored in the speech recognition correction model to obtain a number of matching words;

[0014] The sentence to be processed is corrected according to a plurality of the matching words to obtain a first sentence.

[0015] Furthermore, the combined intent recognition and correction model performs intent matching on the first sentence to obtain a first intent, and obtains a first intent processing function corresponding to the first intent, specifically:

[0016] Performing a first intent match on the first sentence by using a plurality of intent matching rules pre-stored in the intent recognition and correction model;

[0017] When a first intention matching result is obtained through the first intention matching, determining the first intention matching result as the first intention;

[0018] When the second intent matching result cannot be obtained through the first intent matching, the AI ​​big model is used to match the first sentence with intent;

[0019] A second intention matching result obtained through the second intention matching is determined as the first intention.

[0020] Furthermore, the combined parameter correction model extracts several first parameters related to the first intent in the first sentence, specifically:

[0021] Extracting several intent parameters related to the first intent from the first sentence using the AI ​​big model;

[0022] Using a plurality of parameter correction rules pre-stored in the parameter correction model, a plurality of the intended parameters are corrected to obtain a plurality of first parameters.

[0023] Further, the first intention processing function is executed according to the plurality of the first parameters, and a voice response result is obtained, specifically:

[0024] When the first intention is an exponential search, performing an exponential search function based on a plurality of the first parameters to perform data screening, obtaining a data screening result, and determining the data screening result as a voice response result;

[0025] When the first intention is to generate a creative idea, executing a creative idea generation function based on the plurality of the first parameters to generate a plurality of creative ideas, and determining the plurality of the creative ideas as voice response results;

[0026] When the first intention is creativity scoring, executing a creativity scoring function based on the plurality of the first parameters to obtain creativity scores corresponding to the plurality of creativity, and determining the plurality of the creativity scores as voice response results;

[0027] When the first intention is to generate a note, executing a note generation function based on the plurality of the first parameters, generating a note generation result, and determining the note generation result as a voice response result;

[0028] When the first intention is note optimization, executing a note optimization function based on the plurality of first parameters, generating a note optimization result, and determining the note optimization result as a voice response result;

[0029] When the first intention is expert recommendation, executing an expert screening function based on the first parameters, generating an expert screening list, and determining the expert screening list as a voice response result;

[0030] When no intent is matched or the first parameter lacks parameters required to execute the first intent processing function, a reminder reply language is formed and the reminder reply language is determined as a voice response result.

[0031] Further, after executing the first intention processing function according to the plurality of the first parameters and obtaining a voice response result, the method further includes:

[0032] When the first intent is index search, use a page that supports filtering to display voice response results;

[0033] When the first intention is to generate ideas, generate notes, or optimize notes, use a pop-up window on the page to display the voice response results;

[0034] When the first intent is creativity rating, the creativity rating tab is used to display the voice response results;

[0035] When the first intention is expert recommendation, the voice response result is displayed in combination with a page pop-up box and a table; wherein the table includes the recommendation reasons of each expert.

[0036] Furthermore, when the first speech is received, the speech recognition correction model is combined to perform speech-to-text operation on the first speech to form the first sentence, and the method further includes:

[0037] Use monitors to collect sound signals in real time;

[0038] When a preset keyword is recognized in the sound signal, a voice input function is started.

[0039] The present invention provides a control method for an intelligent voice assistant. When a first voice is received, a voice-to-text operation is performed on the first voice in combination with a voice recognition correction model to form a first sentence; an intent recognition correction model is used to match the first sentence to obtain a first intent, and a first intent processing function corresponding to the first intent is obtained; a parameter correction model is used to extract a plurality of first parameters related to the first intent in the first sentence; the first intent processing function is executed according to the plurality of first parameters, and a voice response result is obtained. The present invention replaces the cumbersome operations of manual input and screening with voice input, thereby simplifying the interaction process between users and various systems; a correction model is used in the process of processing user voice to improve the accuracy of intent matching; multiple algorithms are called in series to effectively control data circulation, thereby breaking data silos and lowering the threshold for use.

[0040] Accordingly, the present invention provides a control device for an intelligent voice assistant, comprising: a voice recognition module, an intention matching module, a parameter extraction module and a response module;

[0041] The speech recognition module is used for, when receiving a first speech, performing speech-to-text conversion on the first speech in combination with the speech recognition correction model to form a first sentence;

[0042] The intention matching module is used to perform intention matching on the first sentence in combination with the intention recognition and correction model to obtain a first intention, and obtain a first intention processing function corresponding to the first intention;

[0043] The parameter extraction module is used to extract a plurality of first parameters related to the first intention in the first sentence in combination with the parameter correction model;

[0044] The response module is used to execute the first intention processing function according to the first parameters and obtain a voice response result.

[0045] Furthermore, the speech recognition module includes: a text conversion unit, a pinyin conversion unit, a comparison unit and a correction unit;

[0046] The text-to-speech unit is used to perform a speech-to-text operation on the first speech using speech recognition technology to form a sentence to be processed;

[0047] The pinyin conversion unit is used to convert the text of the sentence to be processed into pinyin to form the pinyin to be processed;

[0048] The comparison unit is used to compare the pinyin to be processed with the brand mapping table pre-stored in the speech recognition correction model to obtain a plurality of matching words;

[0049] The correction unit is used to correct the sentence to be processed according to a plurality of the matching words to obtain a first sentence.

[0050] Further, the intention matching module includes: a first matching unit, a first determining unit, a second matching unit and a second determining unit;

[0051] The first matching unit is used to perform a first intent matching on the first sentence by using a plurality of intent matching rules pre-stored in the intent recognition and correction model;

[0052] The first determining unit is configured to determine the first intention matching result as the first intention when a first intention matching result is obtained through the first intention matching;

[0053] The second matching unit is used to use the AI ​​big model to perform intent matching on the first sentence when the second intent matching result cannot be obtained through the first intent matching;

[0054] The second determination unit is used to determine a second intention matching result obtained through the second intention matching as the first intention.

[0055] The present invention provides a control device for an intelligent voice assistant, which is based on the organic combination of modules. When a first voice is received, a voice recognition correction model is combined to perform voice-to-text operation on the first voice to form a first sentence; an intention recognition correction model is combined to perform intent matching on the first sentence to obtain a first intention, and a first intention processing function corresponding to the first intention is obtained; a parameter correction model is combined to extract several first parameters related to the first intention in the first sentence; according to several first parameters, the first intention processing function is executed, and a voice response result is obtained. The present invention uses voice input to replace the cumbersome operations of manual input and screening, simplifying the interaction process between users and various systems; a correction model is used in the process of processing user voice to improve the accuracy of intention matching; multiple algorithms are called in series to effectively control data circulation, thereby breaking data islands and lowering the threshold for use. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A flowchart of an embodiment of a method for controlling an intelligent voice assistant provided by the present invention;

[0057] Figure 2 A flowchart of another embodiment of the control method of the intelligent voice assistant provided by the present invention;

[0058] Figure 3 A schematic structural diagram of an embodiment of a control device for an intelligent voice assistant provided by the present invention. DETAILED DESCRIPTION

[0059] 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.

[0060] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0061] Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0062] Example 1

[0063] See also Figure 1 , is a flow chart of an embodiment of a control method of an intelligent voice assistant provided by the present invention, the method includes steps 101 to 104, each step is specifically as follows:

[0064] Step 101: When a first voice is received, a voice-to-text operation is performed on the first voice in combination with a voice recognition correction model to form a first sentence.

[0065] Further, in the first embodiment of the present invention, when the first speech is received, the speech recognition correction model is combined to perform speech-to-text operation on the first speech to form the first sentence, and the method further includes:

[0066] Use monitors to collect sound signals in real time;

[0067] When a preset keyword is recognized in the sound signal, a voice input function is started.

[0068] In the first embodiment of the present invention, the user wakes up the intelligent voice assistant through voice. Specifically, the sound signal around the intelligent voice assistant is collected in real time by using a monitor, and the collected sound signal is detected. When the preset keyword is detected in the sound signal, the voice input function of the intelligent voice assistant is awakened, so that the intelligent voice assistant receives the user's voice input and realizes the user's needs according to the user's voice input. By setting the preset keyword for waking up the intelligent voice assistant, the present invention can prevent the intelligent voice assistant from receiving the keyword when it is not awakened, and automatically running when the user does not need to use the intelligent voice assistant, causing a waste of resources.

[0069] Further, in the first embodiment of the present invention, when a first voice is received, a voice-to-text operation is performed on the first voice in combination with the voice recognition correction model to form a first sentence, specifically:

[0070] Using speech recognition technology to perform speech-to-text conversion on the first speech to form a sentence to be processed;

[0071] Convert the text of the sentence to be processed into pinyin to form the pinyin to be processed;

[0072] Compare the pinyin to be processed with the brand mapping table pre-stored in the speech recognition correction model to obtain a number of matching words;

[0073] The sentence to be processed is corrected according to a plurality of the matching words to obtain a first sentence.

[0074] In the first embodiment of the present invention, when the user describes the demand in colloquial language, some proprietary special words, such as brand words, etc., are usually carried. If the speech recognition API does not set these proprietary words in advance, the text conversion will not meet the user's expectations. For example: when the user uses voice input "help me check the delivery situation of Cui Shengsheng's grass planting", if the speech recognition API does not have the built-in brand word "Cui Shengsheng", the speech-to-text result may be "help me check the delivery situation of Cui Shengsheng's grass planting", which generates erroneous text information. Therefore, by setting a brand mapping table in the speech recognition correction model, with pinyin as the key and brand words as the value, the accuracy of speech-to-text conversion can be further improved. The present invention converts the voice file into text, and intelligently corrects the converted text by establishing a brand mapping table, so as to timely identify and process the special brands mentioned by the user to avoid misunderstandings. It not only improves the accuracy of the speech recognition system, but also accurately identifies the user's spoken needs, improves the user experience, and provides users with a more personalized and meticulous service experience.

[0075] As an example of the first embodiment of the present invention, when a first voice input by a user is received, the first voice is converted into a sentence to be processed using voice recognition technology, and then the sentence to be processed is converted into pinyin to be processed, and then this pinyin is queried in a set brand mapping table. If the target brand word is found in the brand mapping table, the matched brand word will be compared with the words in the sentence to be processed. If the two are the same, the sentence to be processed will not be modified; if the two are different, the matched brand word will replace the words in the sentence to be processed. Assume that the brand mapping table in the voice recognition correction model records "cuishengsheng-->cuishengsheng; yagao-->yagao" and so on. The words in the sentence to be processed include cuishengsheng and toothpaste doctor, then the correction process of these two words is "cuishengsheng-->cuishengsheng-->cuishengsheng", and "toothpaste doctor-->yagaoyisheng-->yagao doctor".

[0076] In the first embodiment of the present invention, after the user turns on the voice input function of the intelligent voice assistant, he or she can read out the question he wants to query in colloquial language. The intelligent voice assistant can receive the complete voice signal, call the voice recognition interface, perform voice-to-text operation on the voice segment, and then perform text correction through the voice recognition correction model. At this time, the intelligent voice assistant can display the text information after conversion and correction, and the user can view and edit it. After the user confirms that the information is correct, the text information after the voice conversion will be used as the parameter of the next step for subsequent intent matching. The present invention uses a voice-driven method, and users do not need to log in to different systems to operate. As long as they describe their own problems to the intelligent voice system, the user's needs can be translated and the text can be corrected with the help of the self-built voice recognition correction model, so as to understand and match the system that the user expects to call later, thereby solving the problems of system dispersion and cumbersome interaction, and building a unified service system.

[0077] Step 102: perform intent matching on the first sentence in combination with the intent recognition and correction model to obtain a first intent, and acquire a first intent processing function corresponding to the first intent.

[0078] Further, in the first embodiment of the present invention, the first sentence is matched with intent in combination with the intent recognition and correction model to obtain a first intent, and a first intent processing function corresponding to the first intent is obtained, specifically:

[0079] Performing a first intent match on the first sentence by using a plurality of intent matching rules pre-stored in the intent recognition and correction model;

[0080] When a first intention matching result is obtained through the first intention matching, determining the first intention matching result as the first intention;

[0081] When the second intent matching result cannot be obtained through the first intent matching, the AI ​​big model is used to match the first sentence with intent;

[0082] A second intention matching result obtained through the second intention matching is determined as the first intention.

[0083] In the first embodiment of the present invention, when users describe their needs, there are various forms of expression, and there may even be overly simple descriptions or key words with other intentions, which will cause AI to be inaccurate when matching user intentions. The present invention solves this problem by setting intention matching rules in the intention recognition and correction model. The present invention uses the self-built intention matching rules in the AI ​​big model and the intention recognition and correction model to match user intentions. If the results of the two matches are consistent, the matching result is determined to be the user's intention. If the results of the two matches are inconsistent, the matching result obtained by matching the intention recognition and correction model is the user's intention. If the user's intention cannot be matched using the intention recognition and correction model, the matching result obtained using the AI ​​big model is determined as the user's intention. The present invention adopts a dual strategy for intent matching, which can effectively improve the accuracy of intent matching and ensure the consistency of the system's understanding and feedback of user intentions.

[0084] As an example of the first embodiment of the present invention, a series of intent matching rules are set in the intent recognition and correction model to supplement and verify the intent matching of the AI ​​large model. For example, when the recognition result of the user's voice contains both the words "generate" and "creative", it can be considered that the user's intention is to perform a creative generation operation. According to similar logic, a series of intent matching rules can be established to ensure accurate understanding of user intentions. Among them, the intent matching rules in the intent recognition and correction model include but are not limited to the mapping relationships of the following situations:

[0085] "Planting" or "delivery" or "market" or "marketing" -> index search;

[0086] "Filter" or "Screen" or "Find" or "View only" -> Index filtering;

[0087] "Generate" or "recommend" or "give" and "idea" -> idea generation;

[0088] "Contrast" or "Compare" and "Creativity" -> Creativity Rating;

[0089] "Generate" or "Create" and "Note" or "Article" -> Note generation;

[0090] "SEO" or "Optimization" -> SEO optimization;

[0091] "Recommend" or "Find" or "Filter" and "Expert" or "Blogger" -> Recommend experts.

[0092] Step 103: extracting a plurality of first parameters related to the first intention from the first sentence in combination with the parameter correction model.

[0093] Further, in the first embodiment of the present invention, a plurality of first parameters related to the first intent are extracted from the first sentence in combination with the parameter correction model, specifically:

[0094] Extracting several intent parameters related to the first intent from the first sentence using the AI ​​big model;

[0095] Using a plurality of parameter correction rules pre-stored in the parameter correction model, a plurality of the intended parameters are corrected to obtain a plurality of first parameters.

[0096] In the first embodiment of the present invention, after obtaining the user's intention through intent matching, necessary parameters are extracted from the first sentence converted from the user's voice, and parameter correction model is used to perform parameter verification, so that the intelligent voice assistant can understand the user's intention more comprehensively and deeply, ensuring that each match is based on detailed information, and the user does not need to perform tedious manual operations to input parameter information and condition screening, further solving the problem of cumbersome interaction process and high usage threshold.

[0097] As an example of the first embodiment of the present invention, assuming that the first sentence after voice conversion is "Help me check the planting and delivery situation of Lay's potato chips last year", through intention matching, it can be concluded that the user's intention is "query the planting and delivery situation", and it can be concluded that the intention processing function at this time is an exponential search algorithm, and the parameters such as the brand "Lay's", the product "potato chips", and the time interval "last year" are extracted from the first sentence, and then the parameter correction rules in the parameter correction model are used to perform parameter verification and correction to complete the subsequent algorithm call execution. For example, the mapping relationships in the parameter correction model include "first half of last year -> January 1 of the previous year of the current year to June 30 of the previous year of the current year", "this year -> January 1 of the current year to today", etc.

[0098] Step 104: Execute the first intention processing function according to the plurality of the first parameters, and obtain a voice response result.

[0099] Further, in the first embodiment of the present invention, the first intention processing function is executed according to a plurality of the first parameters, and a voice response result is obtained, specifically:

[0100] When the first intention is an exponential search, performing an exponential search function based on a plurality of the first parameters to perform data screening, obtaining a data screening result, and determining the data screening result as a voice response result;

[0101] When the first intention is to generate a creative idea, executing a creative idea generation function based on the plurality of the first parameters to generate a plurality of creative ideas, and determining the plurality of the creative ideas as voice response results;

[0102] When the first intention is creativity scoring, executing a creativity scoring function based on the plurality of the first parameters to obtain creativity scores corresponding to the plurality of creativity, and determining the plurality of the creativity scores as voice response results;

[0103] When the first intention is to generate a note, executing a note generation function based on the plurality of the first parameters, generating a note generation result, and determining the note generation result as a voice response result;

[0104] When the first intention is note optimization, executing a note optimization function based on the plurality of first parameters, generating a note optimization result, and determining the note optimization result as a voice response result;

[0105] When the first intention is expert recommendation, executing an expert screening function based on the first parameters, generating an expert screening list, and determining the expert screening list as a voice response result;

[0106] When no intent is matched or the first parameter lacks parameters required to execute the first intent processing function, a reminder reply language is formed and the reminder reply language is determined as a voice response result.

[0107] In the first embodiment of the present invention, after the user's intention is obtained through intent matching, the intent processing function corresponding to the intention and the parameters required by the function are obtained at the same time. The corresponding intent processing function is executed based on the obtained parameters to realize the user's intention. Among them, the processing method varies depending on the intention. Intelligent calling of the intent processing function is not just a simple hit algorithm, but a deep calculation process. When calling the intent processing function, it is sometimes necessary to use the algorithm and necessary parameters used in the last intent match, combined with the rich data set accumulated when the previous algorithm was called, to complete the in-depth calculation work. This process involves many aspects of intelligent decision-making, including but not limited to:

[0108] Contextual association: Function calls not only look at the current matching parameters, but also consider the overall context of the user interaction to ensure that a call is based on comprehensive information.

[0109] Historical learning: After accumulating data on users’ historical behaviors, preferences, etc., deep learning technology is used to mine patterns from historical data to provide more targeted suggestions for the current algorithm call.

[0110] Dynamic adjustment: Dynamically adjust the algorithm's weights and parameters based on real-time user feedback and changing environments to maintain the system's adaptability and flexibility.

[0111] Therefore, the algorithm call of the present invention is a deep calculation process, which not only ensures that the algorithm call is not just a mechanical match, but also an intelligent decision. Through the context-related query method, data is continuously accumulated and transmitted throughout the process, effectively solving the problem of data sharing and information fragmentation between independent systems in the past, enhancing the availability and applicability of the system, and thus providing users with more personalized, intelligent and comprehensive services.

[0112] As an example of the first embodiment of the present invention, when the creative generation function is called, it is necessary to use the index search data of brands, products, etc. obtained by calling the index search function, combined with the keyword parameters obtained in the user's voice to generate several creative ideas. When the creative scoring function is called, it is necessary to use several creative ideas obtained by the creative generation function to score these ideas. When the note generation function is called, it is necessary to select creative ideas in combination with the creative scores obtained by the creative scoring function, and generate notes based on the ideas. When the note optimization function is called, it is necessary to use the notes obtained by calling the note generation function to optimize the notes. When the expert recommendation function is called, it is necessary to use the index search data of brands, products, etc. obtained by calling the index search function, and the optimized notes obtained by calling the note optimization function to screen and sort the experts.

[0113] Furthermore, in the first embodiment of the present invention, after executing the first intention processing function according to the plurality of the first parameters and obtaining a voice response result, the method further includes:

[0114] When the first intent is index search, use a page that supports filtering to display voice response results;

[0115] When the first intention is to generate ideas, generate notes, or optimize notes, use a pop-up window on the page to display the voice response results;

[0116] When the first intent is creativity rating, the creativity rating tab is used to display the voice response results;

[0117] When the first intention is expert recommendation, the voice response result is displayed in combination with a page pop-up box and a table; wherein the table includes the recommendation reasons of each expert.

[0118] In the first embodiment of the present invention, executing different intent processing functions will obtain voice response results in different data forms. Therefore, for the voice response results obtained according to different intent processing functions, different display methods are used to display the voice response results, so that users can understand and accept the information obtained more easily. The present invention provides users with personalized and intelligent experience by flexibly selecting different result display forms, thereby meeting the diverse needs of users, improving user satisfaction, and further improving the usability and user stickiness of the intelligent voice assistant.

[0119] As an example of the first embodiment of the present invention, the index search algorithm obtains a summary of a batch of data. At this time, the user may prefer to see a concise data summary, so the current page can be used to display the data results. The page can be filtered and filtered, and the user can quickly obtain accurate information from it. The creative generation algorithm, note generation algorithm and note optimization algorithm obtain a piece of copy processed by the AI ​​large model. At this time, the user may be more concerned about the quality and accuracy of the copy, so the current page is displayed in a pop-up box to intuitively present a variety of content information, such as generated creativity, generated notes, optimized notes and optimized references. When executing the creative scoring algorithm, a newly opened tab is involved, and other pages are opened. Therefore, the pre-set integrated creative scoring system and the newly opened tab bar can be used to display multi-dimensional scoring results. The execution of the expert recommendation algorithm involves table display, so a pop-up box combined with a table form can be used, and the recommendation reasons can be combined to provide users with more comprehensive and specific recommendation information. The present invention adopts different result presentation forms for different data forms to better meet the diverse needs of users and provide a more personalized and intelligent service experience, thereby improving the user experience.

[0120] As an example of the first embodiment of the present invention, see Figure 2 , is a flow chart of another embodiment of the control method of the intelligent voice assistant provided by the present invention. When the listener receives the keyword voice sent by the user, it wakes up the voice input function of the intelligent voice assistant, receives the user's subsequent voice input, uses the voice recognition API to convert the voice file into text, and calls the model to match the user's intention. Based on the different matched user intentions, it selectively calls one or more intention processing functions in the index search algorithm, creative generation algorithm, creative scoring algorithm, note generation algorithm, note optimization algorithm, and expert recommendation algorithm to generate a voice response result, and uses different display methods according to the different data forms of the voice response result. The present invention simplifies the user operation process, improves the convenience of user experience, lowers the usage threshold, and provides users with a more comprehensive and personalized voice interaction service by integrating multiple algorithms and through voice recognition and intent matching.

[0121] In summary, the first embodiment of the present invention provides a control method for an intelligent voice assistant. When a first voice is received, a voice-to-text operation is performed on the first voice in combination with a voice recognition correction model to form a first sentence; an intent recognition correction model is used to match the first sentence to obtain a first intent, and a first intent processing function corresponding to the first intent is obtained; a parameter correction model is used to extract several first parameters related to the first intent in the first sentence; according to several first parameters, the first intent processing function is executed, and a voice response result is obtained. The present invention uses voice input to replace the cumbersome operations of manual input and screening, thereby simplifying the interaction process between users and various systems; a correction model is used in the process of processing user voice to improve the accuracy of intent matching; and multiple algorithms are called in series to effectively control data flow, thereby breaking data silos and lowering the threshold for use.

[0122] Example 2

[0123] See also Figure 3 , is a structural schematic diagram of an embodiment of a control device of an intelligent voice assistant provided by the present invention, the device includes a voice recognition module, an intention matching module, a parameter extraction module and a response module;

[0124] The speech recognition module 201 is used for, when receiving a first speech, performing speech-to-text conversion on the first speech in combination with a speech recognition correction model to form a first sentence;

[0125] The intention matching module 202 is used to perform intention matching on the first sentence in combination with the intention recognition and correction model to obtain a first intention and acquire a first intention processing function corresponding to the first intention;

[0126] The parameter extraction module 203 is used to extract a plurality of first parameters related to the first intention in the first sentence in combination with the parameter correction model;

[0127] The response module 204 is used to execute the first intention processing function according to the plurality of the first parameters and obtain a voice response result.

[0128] Further, in the second embodiment of the present invention, the speech recognition module 201 includes: an integration unit, a generation unit, a request unit and a display unit;

[0129] The text conversion unit is used to perform a speech-to-text conversion operation on the first speech using speech recognition technology to form a sentence to be processed;

[0130] The pinyin conversion unit is used to convert the text of the sentence to be processed into pinyin to form the pinyin to be processed;

[0131] The comparison unit is used to compare the pinyin to be processed with the brand mapping table pre-stored in the speech recognition correction model to obtain a number of matching words;

[0132] The correction unit is used to correct the sentence to be processed according to a plurality of the matching words to obtain a first sentence.

[0133] Further, in the second embodiment of the present invention, the intention matching module 202 includes: a first matching unit, a first determining unit, a second matching unit and a second determining unit;

[0134] The first matching unit is used to perform a first intent matching on the first sentence by using a plurality of intent matching rules pre-stored in the intent recognition and correction model;

[0135] The first determining unit is used for determining the first intention matching result as the first intention when a first intention matching result is obtained through the first intention matching;

[0136] The second matching unit is used to perform intent matching on the first sentence using the AI ​​big model when the second intent matching result cannot be obtained through the first intent matching;

[0137] The second determining unit is used to determine a second intention matching result obtained through the second intention matching as the first intention.

[0138] Further, in the second embodiment of the present invention, the parameter extraction module 203 includes: a parameter extraction unit and a parameter correction unit;

[0139] The parameter extraction unit is used to extract a plurality of intent parameters related to the first intent in the first sentence by using the AI ​​big model;

[0140] The parameter correction unit is used to correct a number of the intended parameters using a number of parameter correction rules pre-stored in the parameter correction model to obtain a number of first parameters.

[0141] Further, in the second embodiment of the present invention, the response module 204 includes: a first response unit, a second response unit, a third response unit, a fourth response unit, a fifth response unit, a sixth response unit and a seventh response unit;

[0142] The first response unit is used for, when the first intention is an exponential search, performing an exponential search function based on a plurality of the first parameters to perform data screening, obtaining a data screening result, and determining the data screening result as a voice response result;

[0143] The second response unit is used for executing a creative generation function based on a plurality of the first parameters to generate a plurality of creatives when the first intention is creative generation, and determining the plurality of the creatives as voice response results;

[0144] The third response unit is used for executing a creativity scoring function based on a plurality of the first parameters to obtain creativity scores corresponding to a plurality of creativity when the first intention is creativity scoring, and determining the plurality of the creativity scores as voice response results;

[0145] The fourth response unit is used for executing a note generation function based on a plurality of the first parameters to generate a note generation result when the first intention is to generate a note, and determining the note generation result as a voice response result;

[0146] The fifth response unit is used for executing a note optimization function based on a plurality of the first parameters to generate a note optimization result, and determining the note optimization result as a voice response result when the first intention is note optimization;

[0147] The sixth response unit is used for executing an expert screening function based on a plurality of the first parameters to generate an expert screening list, and determining the expert screening list as a voice response result when the first intention is expert recommendation;

[0148] The seventh response unit is used to form a reminder reply language and determine the reminder reply language as a voice response result when the intent is not matched or the first parameter lacks the parameters required to execute the first intent processing function.

[0149] Furthermore, in the second embodiment of the present invention, after executing the first intention processing function according to the plurality of the first parameters and obtaining a voice response result, the method further includes:

[0150] When the first intent is index search, use a page that supports filtering to display voice response results;

[0151] When the first intention is to generate ideas, generate notes, or optimize notes, use a pop-up window on the page to display the voice response results;

[0152] When the first intent is creativity rating, the creativity rating tab is used to display the voice response results;

[0153] When the first intention is expert recommendation, the voice response result is displayed in combination with a page pop-up box and a table; wherein the table includes the recommendation reasons of each expert.

[0154] Further, in the second embodiment of the present invention, when the first speech is received, the speech recognition correction model is combined to perform speech-to-text operation on the first speech to form the first sentence, and the method further includes:

[0155] Use monitors to collect sound signals in real time;

[0156] When a preset keyword is recognized in the sound signal, a voice input function is started.

[0157] In summary, the second embodiment of the present invention provides a control device for an intelligent voice assistant, which is based on the organic combination of modules. When a first voice is received, the first voice is converted into text by combining the voice recognition correction model to form a first sentence; the first sentence is matched with the intention recognition correction model to obtain the first intention, and the first intention processing function corresponding to the first intention is obtained; the parameter correction model is combined to extract several first parameters related to the first intention in the first sentence; according to the several first parameters, the first intention processing function is executed, and a voice response result is obtained. The present invention uses voice input to replace the tedious operations of manual input and screening, which simplifies the interaction process between users and various systems; the correction model is used in the process of processing user voice to improve the accuracy of intention matching; and multiple algorithms are called in series to effectively control data flow, thereby breaking data silos and lowering the threshold for use.

[0158] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A control method for an intelligent voice assistant, characterized in that: include: When a first voice is received, performing a voice-to-text operation on the first voice in combination with a voice recognition correction model to form a first sentence; Performing intent matching on the first sentence in combination with the intent recognition and correction model to obtain a first intent, and acquiring a first intent processing function corresponding to the first intent; Extracting a plurality of first parameters related to the first intention in the first sentence in combination with the parameter correction model; Execute the first intention processing function according to the plurality of the first parameters, and obtain a voice response result; The first intention processing function is executed according to the plurality of the first parameters, and a voice response result is obtained, specifically: When the first intention is an exponential search, performing an exponential search function based on a plurality of the first parameters to perform data screening, obtaining a data screening result, and determining the data screening result as a voice response result; When the first intention is to generate a creative idea, executing a creative idea generation function based on the plurality of the first parameters to generate a plurality of creative ideas, and determining the plurality of the creative ideas as voice response results; When the first intention is creativity scoring, executing a creativity scoring function based on the plurality of the first parameters to obtain creativity scores corresponding to the plurality of creativity, and determining the plurality of the creativity scores as voice response results; When the first intention is to generate a note, executing a note generation function based on the plurality of the first parameters, generating a note generation result, and determining the note generation result as a voice response result; When the first intention is note optimization, executing a note optimization function based on the plurality of first parameters, generating a note optimization result, and determining the note optimization result as a voice response result; When the first intention is expert recommendation, executing an expert screening function based on the first parameters, generating an expert screening list, and determining the expert screening list as a voice response result; When no intent is matched or the first parameter lacks parameters required to execute the first intent processing function, a reminder reply language is formed and the reminder reply language is determined as a voice response result.

2. The control method of the intelligent voice assistant according to claim 1, characterized in that: When the first voice is received, the voice recognition correction model is combined to perform a voice-to-text operation on the first voice to form a first sentence, specifically: Using speech recognition technology to perform speech-to-text conversion on the first speech to form a sentence to be processed; Convert the text of the sentence to be processed into pinyin to form the pinyin to be processed; Compare the pinyin to be processed with the brand mapping table pre-stored in the speech recognition correction model to obtain a number of matching words; The sentence to be processed is corrected according to a plurality of the matching words to obtain a first sentence.

3. The control method of the intelligent voice assistant according to claim 1, characterized in that: The combined intention recognition and correction model performs intent matching on the first sentence to obtain a first intent, and obtains a first intent processing function corresponding to the first intent, specifically: Performing a first intent match on the first sentence by using a plurality of intent matching rules pre-stored in the intent recognition and correction model; When a first intention matching result is obtained through the first intention matching, determining the first intention matching result as the first intention; When the second intent matching result cannot be obtained through the first intent matching, the AI ​​big model is used to match the first sentence with intent; A second intention matching result obtained through the second intention matching is determined as the first intention.

4. The control method of the intelligent voice assistant according to claim 1, characterized in that: The combined parameter correction model extracts several first parameters related to the first intention in the first sentence, specifically: Extracting several intent parameters related to the first intent from the first sentence using the AI ​​big model; Using a plurality of parameter correction rules pre-stored in the parameter correction model, a plurality of the intended parameters are corrected to obtain a plurality of first parameters.

5. The control method of the intelligent voice assistant according to claim 1, characterized in that: After executing the first intention processing function according to the plurality of the first parameters and obtaining a voice response result, the method further includes: When the first intent is index search, use a page that supports filtering to display voice response results; When the first intention is to generate ideas, generate notes, or optimize notes, use a pop-up window on the page to display the voice response results; When the first intent is creativity rating, the creativity rating tab is used to display the voice response results; When the first intention is expert recommendation, the voice response result is displayed in combination with a page pop-up box and a table; wherein the table includes the recommendation reasons of each expert.

6. The control method of the intelligent voice assistant according to claim 1, characterized in that: When the first speech is received, the speech-to-text operation is performed on the first speech in combination with the speech recognition correction model to form a first sentence, and the method further includes: Use monitors to collect sound signals in real time; When a preset keyword is recognized in the sound signal, a voice input function is started.

7. A control device for an intelligent voice assistant, characterized in that: include: Speech recognition module, intent matching module, parameter extraction module and response module; The speech recognition module is used for, when receiving a first speech, performing speech-to-text conversion on the first speech in combination with the speech recognition correction model to form a first sentence; The intention matching module is used to perform intention matching on the first sentence in combination with the intention recognition and correction model to obtain a first intention, and obtain a first intention processing function corresponding to the first intention; The parameter extraction module is used to extract a plurality of first parameters related to the first intention in the first sentence in combination with the parameter correction model; The response module is used to execute the first intention processing function according to the first parameters and obtain a voice response result; The first intention processing function is executed according to the plurality of the first parameters, and a voice response result is obtained, specifically: When the first intention is an exponential search, performing an exponential search function based on a plurality of the first parameters to perform data screening, obtaining a data screening result, and determining the data screening result as a voice response result; When the first intention is to generate a creative idea, executing a creative idea generation function based on the plurality of the first parameters to generate a plurality of creative ideas, and determining the plurality of the creative ideas as voice response results; When the first intention is creativity scoring, executing a creativity scoring function based on the plurality of the first parameters to obtain creativity scores corresponding to the plurality of creativity, and determining the plurality of the creativity scores as voice response results; When the first intention is to generate a note, executing a note generation function based on the plurality of the first parameters, generating a note generation result, and determining the note generation result as a voice response result; When the first intention is note optimization, executing a note optimization function based on the plurality of first parameters, generating a note optimization result, and determining the note optimization result as a voice response result; When the first intention is expert recommendation, executing an expert screening function based on the first parameters, generating an expert screening list, and determining the expert screening list as a voice response result; When no intent is matched or the first parameter lacks parameters required to execute the first intent processing function, a reminder reply language is formed and the reminder reply language is determined as a voice response result.

8. The control device of the intelligent voice assistant according to claim 7, characterized in that: The speech recognition module includes: a text conversion unit, a pinyin conversion unit, a comparison unit and a correction unit; The text-to-speech unit is used to perform a speech-to-text operation on the first speech using speech recognition technology to form a sentence to be processed; The pinyin conversion unit is used to convert the text of the sentence to be processed into pinyin to form the pinyin to be processed; The comparison unit is used to compare the pinyin to be processed with the brand mapping table pre-stored in the speech recognition correction model to obtain a plurality of matching words; The correction unit is used to correct the sentence to be processed according to a plurality of the matching words to obtain a first sentence.

9. The control device of the intelligent voice assistant according to claim 7, characterized in that: The intention matching module includes: a first matching unit, a first determining unit, a second matching unit, and a second determining unit; The first matching unit is used to perform a first intent matching on the first sentence by using a plurality of intent matching rules pre-stored in the intent recognition and correction model; The first determining unit is configured to determine the first intention matching result as the first intention when a first intention matching result is obtained through the first intention matching; The second matching unit is used to use the AI ​​big model to perform intent matching on the first sentence when the second intent matching result cannot be obtained through the first intent matching; The second determination unit is used to determine a second intention matching result obtained through the second intention matching as the first intention.

Images