CN-122019556-A - Intelligent voice recognition number query method and system based on AI Agent

CN122019556ACN 122019556 ACN122019556 ACN 122019556ACN-122019556-A

Abstract

The invention discloses an intelligent voice recognition number query method and system based on an AI Agent. The method comprises the steps of receiving user voice number checking information, carrying out voice text conversion and extracting user names, carrying out word segmentation processing on the user names, enabling an AI Agent to inquire in a history record base according to user name word segmentation priority, outputting the user name word segmentation priority, inquiring step by step according to the same and similar sequences of names, pinyin and initial consonants in a contact person information base based on the user name word segmentation if the user name word segmentation exists, outputting candidate numbers according to basic matching degree, feeding the inquired number results back to a user, and storing logs and updating the history record base according to confirmation of the number results by the user. The invention effectively solves the problems of multi-phonetic characters and front and back nasal recognition errors in voice recognition by dividing words by the user name, combining multidimensional query of a history record library and a contact person information library and dynamically updating the history record based on feedback, and improves the number checking accuracy.

Inventors

PENG JUNJIAN
Long Qingjun

Assignees

北京易讯正通网络通信技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. An intelligent voice recognition number query method based on AI Agent is characterized by comprising the following steps: receiving voice directory information of a user, performing voice conversion from the voice directory information to text and extracting a user name; performing word segmentation processing on the user name to obtain user name word segmentation; The AI Agent divides words according to the user name, preferentially inquires in a history record library, and if the corresponding number is inquired in the history record library, the AI Agent outputs an inquiry result; If no corresponding number is queried in the history record library, continuing to query step by step in a contact information library by the AI Agent based on the user name word segmentation according to the same and similar sequence of the name, the pinyin and the initial consonant, and outputting a candidate number according to the basic matching degree; Feeding back the output number result to the user; and according to the confirmation of the user to the number result, saving a log and updating the history record library.
2. The method of claim 1, wherein the step of word segmentation of the user name to obtain a user name word segmentation comprises: performing word segmentation and part-of-speech tagging on the user name through natural language processing, cleaning and keyword extraction based on preset rules, outputting the user name word segmentation, and transmitting the user name word segmentation to an AI Agent; the preset rule is used for carrying out fault tolerance processing on the polyphones and the front and rear nasal sounds, and comprises the steps of accurately labeling the polyphones based on a common Chinese character pinyin library and contextual semantic analysis, and carrying out fuzzy matching correction on the front and rear nasal characters through a pronunciation feature library.
3. The method of claim 1, wherein the AI Agent queries in a history repository according to the user name segmentation priority, and if a corresponding number is queried in the history repository, outputs a query result, comprising: The AI Agent receives the user name word segmentation, preferentially queries one or more candidate numbers matched with the user name word segmentation in the history record library, and if the candidate numbers with the hit times not lower than a preset threshold value exist, the candidate numbers are arranged in descending order according to the hit times and then output; the preset threshold is the accurate number of times of the historical query result, and is determined by user-defined configuration of a system administrator, wherein the hit number is the accumulated count of the word segmentation and number combination after each time the user confirms that the query result is correct.
4. The method of claim 1, wherein if no corresponding number is queried in the history repository, the AI Agent continues to query step by step in the contact information repository in the same and similar order of name, pinyin, and initial consonant based on the user name segmentation, and outputs candidate numbers according to a basic matching degree, specifically comprising: If no corresponding number is queried in the history record library, continuing to query the contact person information library by the AI Agent based on the user name segmentation, and sequentially according to the six dimensions of the name, the pinyin and the initial consonant, wherein the six dimensions are the same and similar; if one or more candidate contact numbers are queried in any dimension, terminating the query, arranging the query in descending order according to the basic matching degree scores corresponding to the preset scoring rule table, outputting the first three candidate numbers, and outputting all query results when the query results are less than three.
5. The method according to claim 4, wherein the basic matching degree specifically comprises: The basic matching degree refers to an original credibility score which is only given according to the rule of the current query dimension in the process of querying the contact person, and additional factors such as context, use frequency, user preference and the like are not considered.
6. The method according to claim 1, characterized in that feeding back the output number result to the user, in particular comprises: and converting the number result into voice information and then broadcasting the voice information to the user for the user to select or confirm.
7. The method of claim 1, wherein maintaining a log and updating the historian based on the user's validation of the number results comprises: And according to the selection of the number result by the user or the confirmation of the accuracy of the number result, a log is saved, and the corresponding relation between the user name segmentation and the number result and the hit frequency are recorded to the history record library.
8. The method according to claim 3 and 4, further comprising prompting the user to provide the voice query information again and repeating the query process if the corresponding number is not queried by the same and similar names, pinyin and initial consonants through the user name word segmentation in the contact information base and/or if the user feedback number is inaccurate, outputting a preset failure prompt and ending the query process if no matching result exists after the voice query information is continuously obtained again and repeating the query process.
9. An AI Agent-based intelligent voice recognition number query system for realizing the AI Agent-based intelligent voice recognition number query method as set forth in any one of claims 1-8, characterized by comprising a voice processing module, a word segmentation module, a first query module, a second query module, a feedback module and a storage updating module, wherein the first query module and the second query module are scheduled and executed by the AI Agent, the first query module preferentially calls a history record library, and the second query module sequentially starts the same and similar queries of names, pinyin and initial consonants when no history record exists; the voice processing module is used for receiving voice directory information of a user, converting the voice directory information into a text and extracting a user name; The word segmentation module is used for carrying out word segmentation processing on the user name to obtain user name word segmentation; The first query module is used for utilizing the AI Agent to segment words according to the user name, preferentially querying in a history record library, and outputting a query result if a corresponding number is queried in the history record library; The second query module is used for querying step by step according to the same and similar sequences of names, pinyin and initials based on the user name segmentation in the contact person information base by utilizing AI Agent when no corresponding number is queried in the history record base, and outputting candidate numbers according to the basic matching degree; The feedback module is used for feeding back the output number result to the user; and the storage updating module is used for storing a log and updating the history record library according to the confirmation of the user to the number result.
10. The system of claim 9, wherein the first query module and the second query module, in particular, further comprise: And the retry control module is used for controlling to reacquire the voice directory information of the user to repeat the query flow when the corresponding number is not queried in the contact person information base or the user feedback number is inaccurate, and controlling to report query failure information to the user if no matching result exists after the voice directory information is reacquired twice continuously and the query flow is repeated.

Description

Intelligent voice recognition number query method and system based on AI Agent Technical Field The invention relates to the technical field of voice recognition and artificial intelligence, in particular to an intelligent voice recognition number query method and system based on AI Agent. Background Traditional telephone number query services (e.g., 114 clan) rely primarily on manual agents or automatic voice response (IVR) systems based on keyword matching. In recent years, with the development of Automatic Speech Recognition (ASR) and Natural Language Processing (NLP) technologies, an automated directory searching scheme for performing database search after text conversion by speech has emerged. However, these schemes are still limited in recognition accuracy in the face of the specific polyphonic, front-to-back nasal sounds, and spoken language expressions of Chinese characters. For example, a user's speech input "Zhang Weifeng" (zhang wei feng), a traditional ASR may be identified as "Zhang Weifen" (zhang wei fen) (front-to-back nasal errors) or "Zhang Weifeng" (zhang wei feng, "Wei" is consistent with "pronunciation" but different in font), resulting in failure of an exact match query, whereas existing vector retrieval schemes can promote semantic understanding, but lack a step-by-step fault-tolerant query mechanism for speech recognition errors such as Chinese character polyphones, front-to-back nasal errors, etc., and do not dynamically optimize in combination with historical query behavior to continuously promote accuracy. Therefore, how to solve the recognition errors of multi-phonetic characters and front and rear nasal sounds of Chinese characters and to improve the query accuracy of the voice recognition number is a technical problem to be solved at present. Disclosure of Invention Based on the method and the system, the invention effectively solves the problems of multi-tone character and front-back nasal recognition errors in voice recognition and improves the number checking accuracy through multi-dimensional query of user name word segmentation, history record library and contact person information library and dynamic update of history record based on feedback. In a first aspect, an embodiment of the present invention provides an intelligent voice recognition number query method based on AI Agent, where the method includes: receiving voice directory information of a user, performing voice conversion from the voice directory information to text and extracting a user name; performing word segmentation processing on the user name to obtain user name word segmentation; The AI Agent divides words according to the user name, preferentially inquires in a history record library, and if the corresponding number is inquired in the history record library, the AI Agent outputs an inquiry result; if no corresponding number is queried in the history record library, continuing to query step by step in the contact information library by the AI Agent based on the user name word segmentation according to the same and similar sequences of the name, the pinyin and the initial consonant, and outputting a candidate number according to the basic matching degree; Feeding back the output number result to the user; and according to the confirmation of the user to the number result, saving a log and updating the history record library. In some embodiments of the present invention, the word segmentation processing is performed on the user name to obtain a user name word segmentation, which specifically includes: performing word segmentation and part-of-speech tagging on the user name through natural language processing, cleaning and keyword extraction based on preset rules, outputting the user name word segmentation, and transmitting the user name word segmentation to an AI Agent; the preset rule is used for carrying out fault tolerance processing on the polyphones and the front and rear nasal sounds, and comprises the steps of accurately labeling the polyphones based on a common Chinese character pinyin library and contextual semantic analysis, and carrying out fuzzy matching correction on the front and rear nasal characters through a pronunciation feature library. In some embodiments of the present invention, the AI Agent queries in a history repository according to the user name word segmentation priority, and if a corresponding number is queried in the history repository, outputs a query result, which specifically includes: The AI Agent receives the user name word segmentation, preferentially queries one or more candidate numbers matched with the user name word segmentation in the history record library, and if the candidate numbers with the hit times not lower than a preset threshold value exist, the candidate numbers are arranged in descending order according to the hit times and then output; the preset threshold is the accurate number of times of the historical query result, and is determined by user-defined con