A method for surname decomposition and confirmation in a voice environment

By combining dialogue context and a polyphonic character priority sorting algorithm, an updatable surname decomposition dictionary is constructed. A personalized adjustment mechanism is used to optimize surname recognition, which solves the problems of homophone confusion and insufficient personalization in surname recognition in a voice environment, and improves recognition accuracy and user experience.

CN120122911BActive Publication Date: 2026-11-10TENGCHUANG YIANG INFORMATION TECH (TAICANG) CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202510180426.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-02-19
Publication Date
2026-11-10
Estimated Expiration
2045-02-19

Smart Images

  • Figure CN120122911B_ABST
    Figure CN120122911B_ABST
Patent Text Reader

Abstract

This invention discloses a method for surname decomposition and confirmation in a voice environment. The method first receives user voice input and converts it into initial text. Then, it extracts dialogue context information and uses a priority ranking algorithm for polyphonic characters, dynamically adjusting weights based on corpus statistics, context information, and user historical interaction data. Next, it queries a pre-set surname decomposition dictionary to filter candidate surnames and retrieves user historical interaction data for personalized correction. If multiple candidate surnames exist or are ambiguous, a targeted confirmation question is generated. The final result is determined based on user feedback and stored for subsequent processing. Simultaneously, surname data is periodically synchronized from a public database to achieve automated version management of the dictionary. This method effectively handles homophone problems, improves the accuracy of surname recognition and user experience in voice interaction, and can be applied to voice interaction platforms. Through inter-module collaborative processing, it reduces the false positive rate.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The present invention relates to the technical field of natural language processing, and in particular to a method for disassembling and confirming surnames in a voice environment. BACKGROUND ART

[0002] In the current wave of digitalization and intelligence, voice interaction technology, as a cutting-edge method for human-computer communication, is being widely integrated into various application scenarios, bringing great convenience to people's lives and work from intelligent voice assistants to customer service hotlines. However, when it comes to the recognition of surnames as key information, voice interaction systems face many intractable problems, which severely restrict the accuracy of services and user experience.

[0003] At present, the technical means for surname recognition in a voice environment mainly include two types: traditional rule matching and machine learning model application. The traditional rule matching method performs query and comparison in a limited surname database according to pre-set simple rules. For example, the text after speech recognition is directly subjected to string matching with standard surnames in the database, and a successful match is determined as the recognition result. The machine learning model method uses a large amount of voice data to train the model, so that the model learns the mapping relationship between voice features and surnames, thereby realizing surname recognition. For example, a deep neural network is used to perform complex feature extraction and pattern recognition on acoustic features of voice signals, so as to determine the surname.

[0004] However, the prior art has obvious defects. On the one hand, the traditional rule matching method is too simple and mechanical. Due to the rich and diverse Chinese surnames and the widespread phenomenon of homophones and polyphonic characters, this method relying only on simple comparison is difficult to cope with complex voice environments. For example, different pronunciations of "Zhang (representing the surname 张)", "Zhang (representing the surname 章)" and "Qin (the surname 覃 with multiple pronunciations)" are extremely easy to be confused during speech recognition, and traditional rule matching cannot effectively distinguish them, leading to frequent recognition errors. At the same time, the limited database it relies on is difficult to cover all surname situations, and the problem of missed recognition often occurs for some rare surnames, regional variant surnames or newly emerged surnames. On the other hand, although machine learning models have achieved certain results in the overall field of speech recognition, they are insufficiently optimized for the specific task of surname recognition. In actual application scenarios, different fields (such as medical care, finance, education, etc.) have differences in the usage habits and frequency of surnames, and machine learning models often cannot fully consider these domain features. In addition, model training requires a large amount of high-quality labeled data, and if the data is biased or insufficient, the accuracy of surname recognition will be seriously affected. Furthermore, the prior art generally lacks the utilization of user personalized information, and cannot optimize the recognition result based on information such as the user's past interaction records, geographical location, and social data, so that the recognition process cannot adapt to the unique needs of different users. SUMMARY OF THE INVENTION

[0005] Purpose of the Invention: To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method for surname decomposition and confirmation in a voice environment. This method combines contextual information from the dialogue to assist in name parsing, uses a priority sorting algorithm for polyphonic characters to handle homophones, optimizes recognition based on user historical data using a personalized adjustment mechanism, constructs and maintains an updatable surname decomposition dictionary for rapid surname matching, and adopts a concise question-and-answer strategy to reduce the number of user interactions. This achieves efficient, accurate, and personalized surname decomposition and confirmation, improving the accuracy of surname recognition and user experience in voice interaction.

[0006] Technical Solution: To achieve the above objectives, this invention provides a method for surname decomposition and confirmation in a speech environment, comprising the following steps: S1: Receive user voice input and use a voice recognition engine to convert the voice input into initial text; S2: Extract dialogue context information, which includes current dialogue content, user geographical location, time information, and domain characteristics; S3: Based on the initial text, detect polyphonic characters and use a polyphonic character priority ranking algorithm that dynamically adjusts weights based on corpus statistics, context information and user historical interaction data to rank candidate Chinese characters. The dynamic weight adjustment includes combining geographical location, domain features in context information and high-frequency confirmation records in user historical interaction data. S4: Query the preset surname decomposition dictionary table, which stores predefined surnames and their decomposition combinations, and generates a priority score based on corpus statistics and user feedback data. Based on the initial text and the dictionary table, at least one candidate surname is selected. S5: Retrieve and analyze user historical interaction data, including the user's previously confirmed surname records, social information and user feedback information, and make personalized corrections to the candidate surnames in step S4 based on the data. S6: When there are multiple candidate surnames or the ambiguity still exists after correction, a targeted confirmation question is generated. The question is limited to high-probability error points. Its content includes a description of the decomposition and combination of candidate surnames. The decomposition and combination description is to append the Chinese character structure of the candidate surname to the surname in parentheses and output through the user interaction interface. S7: Receive feedback from the user on the confirmation question, determine the final surname analysis result based on the feedback, and if the feedback is negative, return to step S3 for reprocessing; S8: Store the final surname parsing result in the user's historical interaction data and use it for subsequent personalized corrections and dictionary table updates; S9: Periodically synchronize surname data from a public database and trigger real-time update tasks based on user feedback to automate version management of the surname decomposition dictionary table.

[0007] Furthermore, the polyphonic character priority ranking algorithm includes: generating an initial priority list based on the frequency of polyphonic character usage statistics from a corpus, which includes news, social media, and book data; adjusting the priority weights of regional and domain-related surnames using weighted coefficients, combined with the geographical location and domain features extracted in step S2; and dynamically optimizing the candidate ranking based on frequently confirmed surnames in user historical data and user feedback information. Through the analysis of a large amount of text data, the probability of different pronunciations appearing in various scenarios can be obtained, thus generating an initial priority list for polyphonic characters. This list provides a foundation for subsequent ranking, enabling the system to have a preliminary judgment basis when processing polyphonic characters. Different regions may have specific preferences for the pronunciation of certain surnames, and different fields (such as medicine, finance, etc.) may also have unique surname usage habits. By introducing weighted coefficients, the system can adjust the priority weights of regional and domain-related surnames according to these regional and domain characteristics, making the ranking results more consistent with reality. Each user may have their own unique surname pronunciation habits, and the system can dynamically optimize the candidate ranking based on users' past confirmation records and feedback. For example, if a user frequently confirms a specific pronunciation of a surname, that pronunciation will be given higher priority in subsequent processing.

[0008] Furthermore, the surname decomposition dictionary includes: decomposition combinations of standard surnames; regional variations and extended decomposition combinations of famous surnames, wherein the regional variations include surname decomposition combinations due to dialect or regional cultural differences; and a dynamic update mechanism implemented through user feedback, public database synchronization, and automatic update tasks. The surname decomposition dictionary stores decomposition combinations of standard surnames. This provides the system with a basic surname decomposition reference. When receiving possible surname input from a user, it can perform preliminary matching and filtering based on these standard decomposition combinations, ensuring correct processing of common and general surnames. Considering the variations in surname decomposition combinations due to dialect or regional cultural differences, different regions may have different decomposition methods for the same surname. This function enables the system to adapt to regional diversity, process surname input from users in different regions, avoid surname recognition errors caused by regional differences, and include extended decomposition combinations of famous surnames. Famous people's surnames may have some special decomposition methods due to their unique cultural background or historical origins. Incorporating this information into the dictionary table enriches the system's knowledge base of surnames and enhances the system's comprehensiveness in handling various surnames.

[0009] Furthermore, the personalized correction includes: when a user's historical data contains a confirmation record matching the current candidate surname, the corresponding decomposition and combination is directly adopted; when the user's social profile contains surname information, that information is used as the basis for correction; when user feedback data contains correction records for a specific surname, the corrected decomposition and combination is prioritized. Social profiles are usually filled in by the user and have high accuracy and reliability. The system can use this information to adjust the decomposition and combination of candidate surnames, further optimizing the surname parsing results. User feedback is an important basis for system improvement; by responding promptly to user corrections, the system can continuously learn and optimize, improving the accuracy of surname decomposition confirmation.

[0010] Furthermore, the rules for generating the confirmation question include: The system generates a confirmation question only when the difference in priority scores between candidate surnames is less than a preset threshold. This means that when the priorities of candidate surnames are close and difficult to determine directly, the system will ask the user to clarify the correct surname. For cases with large priority score differences, the system can directly determine the result, avoiding unnecessary user interaction and improving processing efficiency. The confirmation question includes a breakdown and combination description of the candidate surnames. This description appends the Chinese character structure of each candidate surname in parentheses after the surname. This method allows users to clearly understand the specific composition of each candidate surname, helping them accurately judge and select the correct surname and reducing incorrect choices due to unclear information. During the issue generation process, high-frequency confirmation records and user feedback information from user history data are given priority to reduce the number of user interactions.

[0011] Furthermore, the method includes recording user feedback data on surname recognition results in real time during the speech recognition process. This feedback data includes confirmation, correction, and newly added surname information. Confirmation indicates that the user acknowledges the recognition result provided by the system; correction refers to the user correcting any errors in the system's recognition; newly added surname information is supplemented by the user when the system encounters a surname not yet included in the database. This information is stored in real time in the user's historical database through a user interaction interface. This ensures that the feedback data is saved promptly, facilitating subsequent analysis and utilization by the system. Based on the feedback data, the method dynamically adjusts the polyphonic character priority sorting algorithm and the surname decomposition dictionary. For the polyphonic character priority sorting algorithm, the priority of different pronunciations is adjusted according to the correct pronunciation preferences reflected in the user feedback. For the surname decomposition dictionary, it is expanded based on newly added surname information, and the decomposition combinations of existing surnames are corrected based on correction information.

[0012] Furthermore, the method employs an artificial intelligence dynamic learning mechanism. Through a supervised learning model, based on user interaction data and public data updates, it periodically optimizes the weight parameters of the polyphonic character priority sorting algorithm and automatically optimizes the surname decomposition dictionary table, thereby improving the accuracy of speech recognition and user experience.

[0013] Furthermore, the method is applicable to voice interaction platforms. During real-time speech recognition, it reduces the misjudgment rate caused by homophones and improves the accuracy of surname decomposition confirmation through inter-module data collaborative processing, including real-time text transmission between the speech recognition module and the context understanding module, and dynamic query matching between the polyphonic character processing module and the surname decomposition dictionary module. This method can be integrated into the voice interaction platform, allowing the platform to use this surname decomposition confirmation method when processing voice input, providing users with more accurate surname recognition services and expanding the functionality and application scenarios of the voice interaction platform. The speech recognition module transmits the recognized initial text to the context understanding module in real time. The context understanding module can analyze the text by combining the current dialogue content, user geographical location, time information, and domain characteristics, providing richer semantic information for subsequent processing and helping the system better understand the user's intent. After detecting polyphonic characters in the initial text, the polyphonic character processing module dynamically queries the surname decomposition dictionary module. Through an algorithm that dynamically adjusts weights based on corpus statistics, context information, and user historical interaction data, it matches the information in the dictionary to determine the correct pronunciation of the polyphonic character and its corresponding surname. Through collaborative processing between modules, and by comprehensively considering factors such as contextual information, handling of homophones, and matching with the surname decomposition dictionary, the system can more accurately distinguish homophones. For example, by combining contextual information, it can determine which surname a homophone is more likely to represent in the current context, thereby reducing misjudgments caused by homophones.

[0014] As can be seen from the above technical solution, the present invention has the following beneficial effects: The method for surname decomposition and confirmation in a voice environment described in this invention combines dialogue context information to assist in name parsing, uses a polyphonic character priority sorting algorithm to handle homophones, utilizes a personalized adjustment mechanism to optimize recognition based on user historical data, constructs and maintains an updatable surname decomposition dictionary table to quickly match surnames, and adopts a concise question-and-answer strategy to reduce the number of user interactions, thereby achieving efficient, accurate, and personalized surname decomposition and confirmation, improving the accuracy of surname recognition and user experience in voice interaction. Attached Figure Description

[0015] Figure 1 A flowchart of a method for surname decomposition and confirmation in a voice environment; Figure 2 This is a software functional module structure diagram for applying this method of surname decomposition and confirmation in a voice environment. Detailed Description of the Embodiments

[0016] Embodiment 1 Scenario description: A user uses an intelligent voice assistant to make a medical appointment, and the system needs to accurately identify the user's surname to complete the appointment process. The voice input by the user is: "I am Doctor Zhang." Step 1: Receive the user's voice input and convert it into an initial text The system converts the voice input "I am Doctor Zhang" from the user into an initial text: "I am zhāng / zhàng Doctor" through a speech recognition engine.

[0017] Step 2: Extract dialog context information The system extracts the context information of the current dialog, including: Current dialog content: medical appointment scenario.

[0018] User's geographic location: Suzhou City.

[0019] Time information: 10 a.m. on February 17, 2025.

[0020] Domain feature: medical domain.

[0021] Step 3: Prioritize polyphonic characters The system detects that the character "Zhang" is a polyphonic character, and its possible pronunciations are "zhāng" or "zhàng".

[0022] The system dynamically adjusts the priority of the polyphonic character based on corpus statistics, context information and user historical interaction data: Corpus statistics show that the pronunciation "zhāng" of the character "Zhang" has a relatively high frequency in the medical domain.

[0023] The context information indicates that the current dialog is a medical appointment scenario, which further supports the priority of "zhāng".

[0024] In the user's historical data, the user has confirmed many times that the surname "Zhang" is pronounced as "zhāng".

[0025] The system takes "zhāng" as the candidate pronunciation with the highest priority.

[0026] Step 4: Query the surname decomposition dictionary table The system queries the preset surname decomposition dictionary table, and finds that the standard decomposition combination of the surname "Zhang" is "bow-long Zhang". Based on corpus statistics and user feedback data, the system generates a priority score of 95 points for the surname "Zhang" (full score is 100 points).

[0027] Step 5: Personalized correction The system retrieves the user's historical interaction data and finds that the user has confirmed on multiple occasions that the surname "Zhang" is pronounced as "zhāng", and the social profile also shows the surname as "Zhang".

[0028] The system directly adopts the confirmation record in the user's historical data to further confirm that the surname "Zhang" is pronounced as "zhāng".

[0029] Step 6: Generate targeted confirmation questions Since the priority score of the surname "Zhang" is high and the user's historical data supports the pronunciation "zhāng", the system does not need to generate a confirmation question.

[0030] If there are multiple candidate surnames or there is still ambiguity after correction, the system will generate a confirmation question such as "Is your surname 'Zhang' (the 'Zhang' formed by the radical 'bow' and the character 'long')?".

[0031] Step 7: Receive user feedback and determine the final result The system presents the parsing result "Zhang (the 'Zhang' formed by the radical 'bow' and the character 'long')" to the user, and the user confirms that it is correct.

[0032] If the user's feedback is negative, the system will return to step 3 for reprocessing.

[0033] Step 8: Store the final parsing result The system stores the final parsing result "Zhang (the 'Zhang' formed by the radical 'bow' and the character 'long')" in the user's historical interaction data for subsequent personalized correction and dictionary table update.

[0034] Step 9: Regularly update the surname decomposition dictionary table The system regularly synchronizes the latest surname data from the public database, and triggers a real-time update task according to user feedback.

[0035] For example, if the user feeds back that the surname "Qin" should be pronounced as "qín" instead of "tán", the system will update the pronunciation priority of the surname "Qin" in the surname decomposition dictionary table.

[0036] As shown Figure 2 in the figure, this diagram is a structural diagram of software function modules applying the method for surname decomposition and confirmation in a speech environment, and the software applies the method for surname decomposition and confirmation in a speech environment as described Figure 1 above.

[0037] The main function modules include a speech recognition module, a context understanding module, a polyphone processing module, a common surname dictionary table module, a personalized adjustment module, a question and answer generation module, and a user interaction interface module.

[0038] Speech recognition module: responsible for converting the user's speech input into text.

[0039] Context understanding module: configured to analyze the dialogue context and other related clues.

[0040] Polyphonic character processing module: Includes a priority sorting algorithm for polyphonic characters.

[0041] Commonly Used Surnames Dictionary Module: Provides a maintainable table of surnames.

[0042] Personalized adjustment module: Optimizes name parsing using user historical data.

[0043] Question and answer generation module: responsible for generating short and targeted questions.

[0044] User interaction interface: Manages communication with users, including asking questions and displaying results.

[0045] Database: Stores all necessary data resources, such as a priority table for polyphonic characters and user history records.

[0046] Figure 2 The diagram depicts the core functional modules supporting this invention and their interrelationships. It illustrates how these modules work together to achieve efficient and accurate name resolution. In particular, the surname common dictionary module plays a crucial role in handling polyphonic characters, helping to quickly locate the correct Chinese character combinations. Technical effects of this embodiment: By combining contextual information, a priority sorting algorithm for homophones, and a personalized correction mechanism, the system can accurately identify users' surnames and reduce misjudgments caused by homophones.

[0047] The number of user interactions was significantly reduced, response time was shortened, and the user experience was improved.

[0048] The system has dynamic learning capabilities and can continuously optimize recognition accuracy based on user feedback and updates to public data.

[0049] The above embodiments are exemplary and are intended to illustrate the technical concept and features of the present invention, so that those skilled in the art can understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made according to the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for surname decomposition and confirmation in a speech environment, characterized in that, Includes the following steps: S1: Receive user voice input and use a voice recognition engine to convert the voice input into initial text; S2: Extract dialogue context information, which includes current dialogue content, user geographical location, time information, and domain characteristics; S3: Based on the initial text, detect polyphonic characters and use a polyphonic character priority ranking algorithm that dynamically adjusts weights based on corpus statistics, context information and user historical interaction data to rank candidate Chinese characters. The dynamic weight adjustment includes combining geographical location, domain features in context information and high-frequency confirmation records in user historical interaction data. S4: Query the preset surname decomposition dictionary table, which stores predefined surnames and their decomposition combinations, and generates a priority score based on corpus statistics and user feedback data. Based on the initial text and the dictionary table, at least one candidate surname is selected. S5: Retrieve and analyze user historical interaction data, including previously confirmed surname records, social information and user feedback information, and make personalized corrections to the candidate surnames in step S4 based on the data. S6: When there are multiple candidate surnames or the ambiguity still exists after correction, a targeted confirmation question is generated. The question is limited to high-probability error points. Its content includes a description of the decomposition and combination of candidate surnames. The decomposition and combination description is to append the Chinese character structure of the candidate surname to the surname in parentheses and output through the user interaction interface. S7: Receive feedback from the user on the confirmation question, determine the final surname analysis result based on the feedback, and if the feedback is negative, return to step S3 for reprocessing; S8: Store the final surname parsing result in the user's historical interaction data and use it for subsequent personalized corrections and dictionary table updates; S9: Periodically synchronize surname data from a public database and trigger real-time update tasks based on user feedback to automate version management of the surname decomposition dictionary table.

2. The method according to claim 1, characterized in that, The priority sorting algorithm for polyphonic characters includes: An initial priority list is generated based on the frequency of use of polyphonic characters in a corpus, which includes news, social media, and book data. Combining the user's geographical location and domain features in the dialogue context information extracted in step S2, the priority weights of regional and domain-related surnames are adjusted by weighting coefficients; The candidate ranking is dynamically optimized based on frequently confirmed surnames in user history data and user feedback.

3. The method according to claim 1, characterized in that, The surname decomposition dictionary table includes: The decomposition and recombination of standard surnames; Regional variations and extended decomposition and combination of celebrity surnames, wherein the regional variations include surname decomposition and combination variations caused by dialect or regional cultural differences; A dynamic update mechanism is implemented through user feedback, public database synchronization, and automatic update tasks.

4. The method according to claim 1, characterized in that, The personalized modifications include: When there is a confirmed record in the user's historical data that matches the current candidate surname, the decomposition and combination corresponding to that record will be used directly. When a user's social profile contains surname information, that information will be used as the basis for correction; When user feedback data contains correction records for a specific surname, the corrected decomposition and combination will be used first.

5. The method according to claim 1, characterized in that, The rules for generating the confirmation question include: A confirmation question is only generated when the difference in priority scores between candidate surnames is less than a preset threshold. The content of the confirmation question includes a description of the breakdown and combination of candidate surnames, wherein the breakdown and combination description is to append the Chinese character structure of the candidate surname to the surname in parentheses. During the issue generation process, high-frequency confirmation records and user feedback information from user history data are given priority to reduce the number of user interactions.

6. The method according to claim 1, characterized in that, The method further includes recording user feedback data on surname recognition results in real time during the speech recognition process. The feedback data includes confirmation, correction, and addition of surname information, and is stored in the user's historical database in real time through the user interaction interface. The method also dynamically adjusts the polyphonic character priority sorting algorithm and the content of the surname decomposition dictionary table based on the feedback data.

7. The method according to claim 1, characterized in that, The method employs an artificial intelligence dynamic learning mechanism. Through a supervised learning model, it periodically optimizes the weight parameters of the polyphonic character priority sorting algorithm based on user interaction data and public data updates, and automatically optimizes the surname decomposition dictionary table, thereby improving the accuracy of speech recognition and user experience.

8. The method according to claim 1, characterized in that, The method is applicable to voice interaction platforms. During real-time voice recognition, it reduces the misjudgment rate caused by homophones and improves the accuracy of surname decomposition confirmation through inter-module data collaborative processing, including real-time text transmission between the voice recognition module and the context understanding module, and dynamic query matching between the polyphonic character processing module and the surname decomposition dictionary module.

Citation Information

Patent Citations

  • System and method for searching and matching data having ideogrammatic content

    AU2006318417A1

  • Name pronunciation by synthesizer

    CA2003565A1