Method and apparatus for improving accuracy of speech recognition, and electronic device

Abstract

The application provides a method and device for improving the accuracy of speech recognition and electronic equipment, and relates to the technical field of artificial intelligence. The method comprises the following steps: obtaining a speech recognition text; wherein the speech recognition text comprises a city name string to be recognized; performing multi-level matching processing on the city name string to be recognized based on a preset database, correcting a non-standard city name incorrectly recognized in the speech recognition text to a standard city name; wherein the preset database at least comprises a city name library in the field of transportation and a dynamically updated wrong word contrast table; and dynamically updating the wrong word contrast table according to the matching relationship generated in the correction process, so as to improve the correction accuracy of the city name in subsequent speech recognition. Through the method provided by the application, not only can the misrecognized words be effectively corrected to the correct place names, but also the wrong word contrast table can be continuously enriched in continuous use, so that the recognition accuracy of the system for the special words in the field of transportation can be gradually improved with the increase of the use frequency.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, and electronic device for improving the accuracy of speech recognition. Background Technology

[0002] Currently, speech recognition technology still has insufficient accuracy in recognizing proper nouns such as city names in scenarios such as ticket purchase and inquiry in the transportation sector. For example, it may misrecognize "Tonghua" as "call", which directly affects the user experience.

[0003] Existing solutions typically employ a general speech recognition model combined with a static error-correcting dictionary. However, these general models lack specific optimization for city names within the transportation sector, making it difficult to effectively distinguish easily confused words. Furthermore, static dictionaries cannot be dynamically updated based on actual recognition errors, resulting in repeated errors not being automatically corrected. Therefore, existing solutions not only have limited recognition accuracy but also lack self-learning capabilities, hindering the continuous improvement of accuracy over long-term use and limiting the deep application of voice interaction in the transportation vertical. Thus, a solution to address these issues is urgently needed. Summary of the Invention

[0004] This application provides a method, apparatus, and electronic device for improving the accuracy of speech recognition, in order to address the deficiencies existing in the prior art.

[0005] This application provides a method for improving the accuracy of speech recognition, including: Obtain the speech recognition text; wherein, the speech recognition text includes a string of the city name to be recognized; Based on a preset database, the city name string to be identified is subjected to multi-level matching processing to correct non-standard city names that are incorrectly identified in the speech recognition text to standard city names; wherein, the preset database includes at least a city name database in the transportation field and a dynamically updated misspelling comparison table, the city name database in the transportation field is used as a benchmark reference for city name correction, and the misspelling comparison table is used to store the mapping relationship between non-standard city names and corresponding standard city names. The misspelling comparison table is dynamically updated based on the matching relationships generated during the correction process to improve the accuracy of city name correction in subsequent speech recognition.

[0006] According to an embodiment of this application, a method for improving the accuracy of speech recognition includes performing multi-level matching processing on the city name string to be recognized based on a preset database, correcting non-standard city names that are incorrectly recognized in the speech recognition text to standard city names, including: The city name string to be identified is used as the first city name, and a first-level match is performed with the city name database in the transportation field to generate a first-level match result. If the first-level matching result indicates a successful match, the first city name is determined as the corrected standard city name.

[0007] According to an embodiment of this application, a method for improving the accuracy of speech recognition, after using the city name string to be recognized as a first city name and performing a first-level match with the city name database in the transportation field to generate a first-level match result, the method further includes: If the first-level matching result indicates that the match is unsuccessful, the first city name is matched with the misspelling comparison table to generate a second-level matching result.

[0008] According to an embodiment of this application, a method for improving the accuracy of speech recognition includes performing a secondary matching between the first city name and the misspelling lookup table to generate a secondary matching result, comprising: Query the misspelling comparison table to see if there are any non-standard city name records that match the name of the first city, and generate the query results; If the query result indicates that there is a non-standard city name record in the misspelling comparison table that matches the first city name, the second city name corresponding to the first city name is obtained from the misspelling comparison table, and the second city name is returned to the first-level matching and rematched with the city name database in the transportation field.

[0009] According to an embodiment of this application, a method for improving the accuracy of speech recognition, after performing a secondary matching between the first city name and the misspelling comparison table to generate a secondary matching result, the method further includes: If the second-level matching result indicates that the match is unsuccessful, proceed to the third-level matching process; The three-level matching process includes: processing the first city name with textual guidance to generate guidance text; inputting the guidance text into a large language model for recognition to obtain the third city name; and returning the third city name to the first-level matching to re-match it with the city name database in the transportation field.

[0010] According to an embodiment of this application, a method for improving the accuracy of speech recognition includes processing the first city name with text guidance to generate guidance text, comprising: Based on the geographical naming characteristics in the transportation field, add a guiding prefix before the first city name; The guiding prefix is ​​used to indicate that the large language model recognizes the string as a city name.

[0011] According to an embodiment of this application, a method for improving the accuracy of speech recognition includes dynamically updating the misspelling comparison table based on the matching relationship generated during the correction process, comprising: After generating the third city name in the three-level matching process, the third city name is returned to the first-level matching process for matching. If the name of the third city is inconsistent with the name of the first city, and the name of the third city is successfully matched to the city name database in the first-level matching, the mapping relationship between the name of the first city and the name of the third city is added to the misspelling comparison table.

[0012] According to an embodiment of this application, a method for improving the accuracy of speech recognition, before using the city name string to be recognized as the first city name, the method further includes: The speech-recognized text is input into the large language model for keyword recognition, and the city name in the speech-recognized text is output as the city name string to be recognized. The city name string to be identified is stored in a cache library for priority retrieval during subsequent matching processes.

[0013] This application embodiment also provides a speech recognition accuracy improvement device, including: An acquisition module is used to acquire speech recognition text; wherein, the speech recognition text includes a string of city names to be recognized; The matching module is used to perform multi-level matching processing on the city name string to be identified based on a preset database, and correct non-standard city names that are incorrectly identified in the speech recognition text to standard city names; wherein, the preset database includes at least a city name database in the transportation field and a dynamically updated misspelling comparison table, the city name database in the transportation field is used as a benchmark reference for city name correction, and the misspelling comparison table is used to store the mapping relationship between non-standard city names and corresponding standard city names; The enhancement module is used to dynamically update the misspelling comparison table based on the matching relationships generated during the correction process, so as to improve the accuracy of city name correction in subsequent speech recognition.

[0014] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-described methods for improving the accuracy of speech recognition.

[0015] This application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the speech recognition accuracy improvement method as described above.

[0016] This application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-described methods for improving the accuracy of speech recognition.

[0017] This application provides a method, apparatus, and electronic device for improving the accuracy of speech recognition. The method involves acquiring speech recognition text, which includes a string of city names to be recognized. Based on a preset database, the method performs multi-level matching processing on the string of city names to be recognized, correcting incorrectly recognized non-standard city names in the speech recognition text to standard city names. The preset database includes at least a city name database in the transportation sector and a dynamically updated misspelling table. The city name database in the transportation sector serves as a benchmark for city name correction, and the misspelling table stores the mapping relationship between non-standard city names and their corresponding standard city names. The misspelling table is dynamically updated based on the matching relationships generated during the correction process to improve the accuracy of city name correction in subsequent speech recognition. Therefore, this application embodiment, based on the city name database in the transportation field contained in the preset database, can provide a precise benchmark for city name correction and ensure the accuracy of the correction direction. At the same time, the dynamically updated misspelling comparison table records the mapping relationship between the non-standard city names and standard names that are identified incorrectly in real time. This allows each correction result to feed back into the system, directly hitting the corrected error type in subsequent recognitions. This not only effectively corrects misidentified words to correct place names, but also continuously enriches the misspelling comparison table with continuous use. As a result, the system's recognition accuracy of transportation-specific terms gradually improves with the frequency of use, thereby significantly improving the interactive experience of voice ticketing, querying, etc. in transportation scenarios. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the method for improving the accuracy of speech recognition provided in this application embodiment.

[0020] Figure 2This is a framework diagram of the speech recognition accuracy improvement method provided in the embodiments of this application.

[0021] Figure 3 This is a complete flowchart of the speech recognition accuracy improvement method provided in the embodiments of this application.

[0022] Figure 4 This is a schematic diagram of the structure of the speech recognition accuracy improvement device provided in the embodiments of this application.

[0023] Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the embodiments of this application.

[0025] The following description, in conjunction with the accompanying drawings, describes a method, apparatus, and electronic device for improving the accuracy of speech recognition according to embodiments of this application.

[0026] Figure 1 This is a flowchart illustrating the method for improving the accuracy of speech recognition provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes the following: Step 100: Obtain the speech recognition text; wherein, the speech recognition text includes the city name string to be recognized.

[0027] Specifically, acquiring speech recognition text refers to the initial text result obtained by converting user speech from a voice terminal using a general speech recognition engine. This result contains city name strings that need to be corrected later. For example, if a user says "buy tickets for Tonghua," the recognition text may misidentify "Tonghua" as "call" or "talk." The city name string to be recognized is the word that is suspected to be the city name extracted from this text, such as "call."

[0028] After step 100, the method further includes: The speech-recognized text is input into the large language model for keyword recognition, and the city name in the speech-recognized text is output as the city name string to be recognized. The city name string to be identified is stored in a cache library for priority retrieval during subsequent matching processes.

[0029] Specifically, the original speech recognition text is input into a large language model for keyword recognition. This large language model has been trained for domain adaptation and can accurately extract lexical units belonging to city names from sentences expressed in natural language. For example, it can recognize "Hua" as a suspected city name from "I want to buy a ticket to Hua", or recognize the corresponding text expression of "Beijing" from "The train to Beijing". These recognition results output by the large language model are used as the city name strings to be recognized, which are the core objects to be processed by multi-level matching in the following steps.

[0030] To improve the system response speed and matching efficiency, in this embodiment, the recognized city name strings to be recognized are stored in a cache library. This cache library serves as a temporary high-speed storage area and can preferentially call the recognized city names during the multi-level matching process, avoiding repeated calls to the large language model for secondary recognition of the same text. Thus, while ensuring the recognition accuracy, the overall processing performance is optimized, providing fast data access support for subsequent private domain library matching, typo对照表 query and other links.

[0031] Step 200: Perform multi-level matching processing on the city name strings to be recognized based on a preset database, and correct non-standard city names recognized incorrectly in the speech recognition text into standard city names. Among them, the preset database includes at least a city name library in the transportation field and a dynamically updated typo对照表. The city name library in the transportation field is used as a reference benchmark for city name correction, and the typo对照表 is used to store the mapping relationship between non-standard city names and corresponding standard city names.

[0032] Specifically, the preset database is the basic data support for correction in this embodiment and contains at least two types of core data: one is the city name library in the transportation field, which is a pre-constructed set of standard place names covering all standard city names involved in the transportation system, such as "Tonghua", serving as a reference benchmark for correction; the other is the dynamically updated typo对照表, which is used to record the mapping relationship between incorrect words and correct standard names that appear in the actual recognition process, such as "Tonghua → Tonghua", "Hua → Hua", etc.

[0033] Multi-level matching processing means sequentially performing correction attempts on the city name strings according to preset matching levels, gradually converting non-standard city names recognized incorrectly into standard city names. Among them, non-standard city names refer to various city name expressions that appear in the recognition text, do not directly match the city name library in the transportation field, and may be incorrect, including forms such as typos, homophones, and mispronunciations of polyphones.

[0034] Step 300: Dynamically update the misspelling comparison table based on the matching relationship generated during the correction process to improve the accuracy of city name correction in subsequent speech recognition.

[0035] Specifically, throughout the entire correction process, the correspondence between each successfully established erroneous word and the standard name is used to dynamically update the misspelling comparison table. This allows the comparison table to be continuously enriched and optimized as the system is used, enabling faster and more accurate correction when encountering the same or similar recognition errors in subsequent processing. Ultimately, this achieves a continuous improvement in the accuracy of city name correction in speech recognition in the transportation field.

[0036] Figure 2 This is a framework diagram of the speech recognition accuracy improvement method provided in the embodiments of this application, such as... Figure 2 As shown, the architecture provided in this embodiment includes at least four core modules: a traffic voice engine, an AI large model, a private domain library, and a cache library. The traffic voice engine receives user voice input and performs initial speech-to-text processing, generating speech-recognized text containing the string of the city name to be recognized. The private domain library is a specific implementation of a pre-defined database, internally deploying a city name database for the transportation field and a dynamically updated misspelling table, providing benchmark reference data and mapping relationships for city name correction. The AI ​​large model (i.e., a large language model) undertakes keyword recognition and secondary recognition tasks; it can extract the city name string to be recognized from the speech-recognized text and perform deep semantic understanding of the guiding text during matching processing to generate corrected city names. The cache library temporarily stores the city name strings recognized by the AI ​​large model, allowing for priority retrieval during multi-level matching and reducing the time spent on repeated recognition. Through the collaborative work of these four modules, this embodiment can achieve the acquisition, multi-level matching processing, correction, and dynamic updating of city name strings in speech-recognized text, thereby continuously improving the accuracy of city name recognition in voice interaction scenarios in the transportation field.

[0037] The above describes the steps of the speech recognition accuracy improvement method provided in the embodiments of this application. As can be seen from the above description, according to the speech recognition accuracy improvement method provided in the embodiments of this application, the following steps are taken: First, speech recognition text is obtained; wherein the speech recognition text includes a string of city names to be recognized; multi-level matching processing is performed on the string of city names to be recognized based on a preset database to correct non-standard city names with recognition errors in the speech recognition text to standard city names; wherein the preset database includes at least a city name database in the transportation field and a dynamically updated misspelling comparison table, the city name database in the transportation field is used as a benchmark reference for city name correction, and the misspelling comparison table is used to store the mapping relationship between non-standard city names and corresponding standard city names; the misspelling comparison table is dynamically updated according to the matching relationship generated during the correction process to improve the accuracy of city name correction in subsequent speech recognition. Therefore, this application embodiment, based on the city name database in the transportation field contained in the preset database, can provide a precise benchmark for city name correction and ensure the accuracy of the correction direction. At the same time, the dynamically updated misspelling comparison table records the mapping relationship between the non-standard city names and standard names that are identified incorrectly in real time. This allows each correction result to feed back into the system, directly hitting the corrected error type in subsequent recognitions. This not only effectively corrects misidentified words to correct place names, but also continuously enriches the misspelling comparison table with continuous use. As a result, the system's recognition accuracy of transportation-specific terms gradually improves with the frequency of use, thereby significantly improving the interactive experience of voice ticketing, querying, etc. in transportation scenarios.

[0038] Based on the above embodiments, in this embodiment, step 200 performs multi-level matching processing on the city name string to be identified based on a preset database, correcting the non-standard city names with recognition errors in the speech recognition text to standard city names, including: Step 210: Use the city name string to be identified as the first city name and perform a first-level match with the city name database in the transportation field to generate a first-level match result.

[0039] Step 220a: If the first-level matching result indicates a successful match, the first city name is determined as the corrected standard city name.

[0040] Specifically, the city name string extracted from the speech recognition text is defined as the first city name. For example, when a user inputs "buy a ticket for a phone call," the word "phone call" recognized by the large language model is the first city name. Then, this first city name is matched against a city name database in the transportation field. The first-level matching process compares the first city name one by one with the standard names in the database, generating a successful or unsuccessful first-level matching result. If the match is successful, for example, if the first city name is exactly "A" and completely matches "A" in the database, this first city name is directly identified as the corrected standard city name and output, without needing to proceed to the next level of matching. If the match fails, for example, if "phone call" does not exist in the database, it indicates that the first city name may be incorrectly identified, and the next level of matching process needs to be initiated for further correction. Through this first-level matching process, correctly identified city names can be quickly identified, avoiding unnecessary subsequent processing and thus improving overall correction efficiency.

[0041] Furthermore, after step 210, the method further includes: Step 220b: If the first-level matching result indicates that the matching is unsuccessful, perform a second-level matching between the first city name and the misspelling comparison table to generate a second-level matching result.

[0042] Step 220b specifically includes: Query the misspelling comparison table to see if there are any non-standard city name records that match the name of the first city, and generate the query results; If the query result indicates that there is a non-standard city name record in the misspelling comparison table that matches the first city name, the second city name corresponding to the first city name is obtained from the misspelling comparison table, and the second city name is returned to the first-level matching and rematched with the city name database in the transportation field.

[0043] It should be noted that if the first-level matching result indicates that the matching is unsuccessful, this embodiment further performs second-level matching processing to attempt to correct the first city name.

[0044] Specifically, the first city name is matched against a misspelling lookup table, a dynamically updated mapping database that pre-stores at least one pair of non-standard city names and their corresponding standard city names, such as "Tonghua → Tonghua" or "Hua → Hua". The process of querying the misspelling lookup table involves checking if a non-standard city name record matches the first city name, generating the corresponding query result. Taking the user's voice recognition of "Tonghua" as an example, if the first city name is "Tonghua", querying the misspelling lookup table reveals a mapping record of "Tonghua → Tonghua", then the corresponding second city name "Tonghua" is retrieved from the table. Subsequently, this second city name is returned to the first-level matching process and re-matched against the city name database in the transportation sector. At this point, "Tonghua", as a standard city name, should successfully match into the city name database, thus completing the correction. Through this two-level matching mechanism, this embodiment can leverage historically accumulated error correction experience to directly target recorded common recognition errors, avoiding repeated complex processing and significantly improving correction efficiency and accuracy.

[0045] Furthermore, after step 220b, the method further includes: If the second-level matching result indicates that the match is unsuccessful, proceed to the third-level matching process; The three-level matching process includes: processing the first city name with textual guidance to generate guidance text; inputting the guidance text into a large language model for recognition to obtain the third city name; and returning the third city name to the first-level matching to re-match it with the city name database in the transportation field.

[0046] The step of processing the first city name with text guidance to generate guiding text includes: Based on the geographical naming characteristics in the transportation field, add a guiding prefix before the first city name; The guiding prefix is ​​used to indicate that the large language model recognizes the string as a city name.

[0047] Specifically, if the second-level matching result indicates that the matching is unsuccessful, that is, the first city name neither directly matches the city name database in the transportation field nor finds a corresponding mapping relationship in the misspelling comparison table, this embodiment further enters the third-level matching process to attempt a deeper level of correction.

[0048] This three - level matching process first performs text guiding and processing on the first city name. The so - called text guiding and processing means adding a specific guiding prefix before the first city name according to the geographical name characteristics in the transportation field. For example, the recognized "Hua" is processed into "City Hua" or "Geographical Name Hua". The guiding prefixes such as "City", "Geographical Name", "Destination", etc. are used to clearly prompt the large - language model that the subsequent string should be understood as a city name rather than an ordinary word. After generating the guiding text, it is input into the large - language model for secondary recognition. Combining the prompt of the guiding prefix, the large - language model can more accurately recognize the words that might have been misjudged originally. For example, it correctly recognizes "City Hua" as "Tonghua" and outputs it as the third city name. Subsequently, this third city name is returned to the first - level matching and rematched with the city name library in the transportation field. If "Tonghua" successfully matches the city name library at this time, the correction process is completed.

[0049] The method for improving the accuracy of speech recognition provided in this embodiment, through a three - level matching mechanism, can, in the case where both conventional matching and table - look - up fail,借助引导提示和大语言模型的深度理解能力,实现对新型或复杂错误类型的有效校正,进一步扩展了系统的错误覆盖范围。

[0050] Based on the above - mentioned embodiment, in this embodiment, dynamically updating the typo对照表 according to the matching relationship generated during the correction process in step 300 includes: Step 310: After generating the third city name in the three - level matching process, return the third city name to the first - level matching for matching.

[0051] Step 320: When the third city name is inconsistent with the first city name and the third city name successfully matches the city name library in the first - level matching, add the mapping relationship between the first city name and the third city name to the typo对照表.

[0052] It should be noted that in the original text of step 3 in the English translation, there is an incomplete part "借助引导提示和大语言模型的深度理解能力,实现对新型或复杂错误类型的有效校正,进一步扩展了系统的错误覆盖范围。" which seems to be an incomplete expression in Chinese. It is recommended to check and correct the original Chinese text for a more accurate translation.Specifically, when the third-level matching process generates the third city name, it first returns the third city name to the first-level matching process to re-match and verify it with the city name library in the transportation field. Taking a specific scenario as an example, assume that the first city name is "Hua". After being processed by text guidance and input into the large language model for recognition, the third city name "Tonghua" is obtained. At this time, "Tonghua" is returned to the first-level matching and compared with the city name library in the transportation field. Since "Tonghua" is a standard city name, the matching result should be successful. On this basis, this embodiment further judges two key conditions: one is that the third city name "Tonghua" is inconsistent with the first city name "Hua", indicating that correction has indeed occurred; the other is that the third city name "Tonghua" is successfully matched to the city name library in the transportation field during the first-level matching, confirming the accuracy of the correction result. When both of these conditions are met, it means that the third-level matching process has successfully discovered a new set of incorrect mapping relationships this time, that is, the incorrect "Hua" has been successfully corrected to the correct "Tonghua". This embodiment immediately adds this set of mapping relationships, that is, the corresponding relationship between the first city name "Hua" and the third city name "Tonghua", to the typo对照表.

[0053] The method for improving the accuracy of speech recognition provided by this embodiment can continuously enrich and expand the typo对照表 with the actual use of the system through a dynamic update mechanism, without repeatedly executing the third-level matching process, thereby realizing the continuous evolution of the system's correction ability and the gradual improvement of the recognition accuracy.

[0054] Figure 3 is the complete flowchart of the method for improving the accuracy of speech recognition provided by the embodiments of this application. The following结合 Figure 3 , the method for improving the accuracy of speech recognition provided by the embodiments of this application will be specifically described.

[0055] Such as Figure 3 It should be noted that the term "typo对照表" in the original text seems to be an incorrect or unclear expression. It might need to be further clarified in the original context for a more accurate translation.As shown, the system obtains the user's voice input string through a traffic voice engine and converts it into speech recognition text, which contains the city name string to be recognized. Then, the speech recognition text is input into a large language model for keyword recognition. The large language model outputs the city name as the first city name to be recognized and stores it in a cache for later priority retrieval. Next, a multi-level matching process is initiated. First, the first city name is matched against the traffic domain city name database in the private domain library. If a match is successful, it is directly output as the standard city name. If a match fails, the first city name is matched against a dynamically updated misspelling table to check for a corresponding mapping relationship. If a mapping relationship exists, the corresponding second city name is obtained and the first-level matching result is returned. If the second-level matching fails, the system proceeds to the third-level matching process. The first city name undergoes textual guidance processing, with a guiding prefix added to generate guiding text. This text is then input into the large language model for recognition, yielding the third city name, which is then returned to the first-level matching process. If the third-level matching successfully matches the third city name to the transportation domain city name database but does not match the first city name, the system adds the mapping relationship between the first and third city names to the misspelling comparison table, completing a dynamic update. Finally, the system controls the processing level by determining the number of optimization attempts, ensuring that the third-level matching is executed only once to avoid performance loss. After the entire process is completed, the system outputs the corrected standard city name. As the misspelling comparison table becomes increasingly rich, the correction of the same errors in subsequent recognitions will become faster and more accurate.

[0056] Taking a specific transportation ticketing scenario as an example, suppose a user says "I want to buy a ticket to Hua" in a voice ticketing system. The transportation voice engine converts this speech into the text "I want to buy a ticket to Hua". The large language model identifies the city name string "Hua" as the first city name from this text. A first-level match is performed between this first city name "Hua" and the city name database in the transportation domain. It is found that "Hua" is not a standard city name, and the match fails. A second-level match is then performed against a misspelling lookup table to check if there is a mapping record for "Hua". If the misspelling lookup table has not yet included the error, the second-level match also fails. The system then enters a third-level matching process, where the text of "Hua" is modified by adding the guiding prefix "city". The system generates a guiding text in "city dialect," which is then input into a large language model for secondary recognition. The model, combined with the guiding prompts, correctly identifies the standard city name "Tonghua" as the third city name. "Tonghua" is then returned to the first-level matching database, where it successfully matches with the city name database in the transportation sector. Since "Tonghua" is inconsistent with the original first city name "hua," the system adds the "hua → Tonghua" mapping to the misspelling table. When the user or another user says "ticket to Hua" again, the system recognizes "hua" and can directly retrieve the corresponding "Tonghua" from the misspelling table through second-level matching, eliminating the need for third-level matching. This achieves continuous improvement in the accuracy of city name recognition in the transportation sector and continuous optimization of system response speed.

[0057] The speech recognition accuracy improvement device provided in the embodiments of this application will be described below. The speech recognition accuracy improvement device described below can be referred to in correspondence with the speech recognition accuracy improvement method described above.

[0058] Figure 4 This is a schematic diagram of the structure of the speech recognition accuracy improvement device provided in the embodiments of this application, as shown below. Figure 4 As shown in the embodiment of this application, the speech recognition accuracy improvement device includes: The acquisition module 401 is used to acquire speech recognition text; wherein, the speech recognition text includes a string of city names to be recognized; The matching module 402 is used to perform multi-level matching processing on the city name string to be identified based on a preset database, and correct the non-standard city names that are incorrectly identified in the speech recognition text to standard city names; wherein, the preset database includes at least a city name database in the transportation field and a dynamically updated misspelling comparison table, the city name database in the transportation field is used as a benchmark reference for city name correction, and the misspelling comparison table is used to store the mapping relationship between non-standard city names and corresponding standard city names; The enhancement module 403 is used to dynamically update the misspelling comparison table based on the matching relationship generated during the correction process, so as to improve the accuracy of city name correction in subsequent speech recognition.

[0059] The speech recognition accuracy improvement device provided in this application embodiment acquires speech recognition text, wherein the speech recognition text includes a string of city names to be recognized; multi-level matching processing is performed on the string of city names to be recognized based on a preset database to correct non-standard city names that are incorrectly recognized in the speech recognition text to standard city names; wherein the preset database includes at least a city name database in the transportation field and a dynamically updated misspelling comparison table, the city name database in the transportation field is used as a benchmark reference for city name correction, and the misspelling comparison table is used to store the mapping relationship between non-standard city names and corresponding standard city names; the misspelling comparison table is dynamically updated according to the matching relationship generated during the correction process to improve the correction accuracy of city names in subsequent speech recognition. Therefore, this application embodiment, based on the city name database in the transportation field contained in the preset database, can provide a precise benchmark for city name correction and ensure the accuracy of the correction direction. At the same time, the dynamically updated misspelling comparison table records the mapping relationship between the non-standard city names and standard names that are identified incorrectly in real time. This allows each correction result to feed back into the system, directly hitting the corrected error type in subsequent recognitions. This not only effectively corrects misidentified words to correct place names, but also continuously enriches the misspelling comparison table with continuous use. As a result, the system's recognition accuracy of transportation-specific terms gradually improves with the frequency of use, thereby significantly improving the interactive experience of voice ticketing, querying, etc. in transportation scenarios.

[0060] Based on the above embodiments, in this embodiment, the matching module 402 is specifically used for: The city name string to be identified is used as the first city name, and a first-level match is performed with the city name database in the transportation field to generate a first-level match result. If the first-level matching result indicates a successful match, the first city name is determined as the corrected standard city name.

[0061] Based on the above embodiments, in this embodiment, the device further includes a second matching module, specifically used for: The process involves using the string of the city name to be identified as the first city name and performing a first-level match with the city name database in the transportation field to generate a first-level match result. If the first-level matching result indicates that the match is unsuccessful, the first city name is matched with the misspelling comparison table to generate a second-level matching result.

[0062] Based on the above embodiments, in this embodiment, the second matching module is further configured to: Query the misspelling comparison table to see if there are any non-standard city name records that match the name of the first city, and generate the query results; If the query result indicates that there is a non-standard city name record in the misspelling comparison table that matches the first city name, the second city name corresponding to the first city name is obtained from the misspelling comparison table, and the second city name is returned to the first-level matching and rematched with the city name database in the transportation field.

[0063] Based on the above embodiments, in this embodiment, the device further includes a third matching module, specifically used for: After performing a secondary match between the first city name and the misspelling comparison table to generate the secondary match result... If the second-level matching result indicates that the match is unsuccessful, proceed to the third-level matching process; The three-level matching process includes: processing the first city name with textual guidance to generate guidance text; inputting the guidance text into a large language model for recognition to obtain the third city name; and returning the third city name to the first-level matching to re-match it with the city name database in the transportation field.

[0064] Based on the above embodiments, in this embodiment, the device further includes a guiding module, specifically used for: Based on the geographical naming characteristics in the transportation field, add a guiding prefix before the first city name; The guiding prefix is ​​used to indicate that the large language model recognizes the string as a city name.

[0065] Based on the above embodiments, in this embodiment, the lifting module 403 is specifically used for: After generating the third city name in the three-level matching process, the third city name is returned to the first-level matching process for matching. If the name of the third city is inconsistent with the name of the first city, and the name of the third city is successfully matched to the city name database in the first-level matching, the mapping relationship between the name of the first city and the name of the third city is added to the misspelling comparison table.

[0066] Based on the above embodiments, in this embodiment, the device further includes a storage module, specifically used for: Before using the city name string to be identified as the first city name. The speech-recognized text is input into the large language model for keyword recognition, and the city name in the speech-recognized text is output as the city name string to be recognized. The city name string to be identified is stored in a cache library for priority retrieval during subsequent matching processes.

[0067] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device can be a robot or other electronic device. This electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communication bus 550. The processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions from the memory 530 to execute methods for improving the accuracy of speech recognition, including: Obtain the speech recognition text; wherein, the speech recognition text includes a string of the city name to be recognized; Based on a preset database, the city name string to be identified is subjected to multi-level matching processing to correct non-standard city names that are incorrectly identified in the speech recognition text to standard city names; wherein, the preset database includes at least a city name database in the transportation field and a dynamically updated misspelling comparison table, the city name database in the transportation field is used as a benchmark reference for city name correction, and the misspelling comparison table is used to store the mapping relationship between non-standard city names and corresponding standard city names. The misspelling comparison table is dynamically updated based on the matching relationships generated during the correction process to improve the accuracy of city name correction in subsequent speech recognition.

[0068] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application embodiment, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in at least one embodiment of this application embodiment. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0069] On the other hand, embodiments of this application also provide a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the speech recognition accuracy improvement methods provided by the above methods, including: Obtain the speech recognition text; wherein, the speech recognition text includes a string of the city name to be recognized; Based on a preset database, the city name string to be identified is subjected to multi-level matching processing to correct non-standard city names that are incorrectly identified in the speech recognition text to standard city names; wherein, the preset database includes at least a city name database in the transportation field and a dynamically updated misspelling comparison table, the city name database in the transportation field is used as a benchmark reference for city name correction, and the misspelling comparison table is used to store the mapping relationship between non-standard city names and corresponding standard city names. The misspelling comparison table is dynamically updated based on the matching relationships generated during the correction process to improve the accuracy of city name correction in subsequent speech recognition.

[0070] In another aspect, embodiments of this application also provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the speech recognition accuracy improvement methods provided by the above methods, including: Obtain the speech recognition text; wherein, the speech recognition text includes a string of the city name to be recognized; Based on a preset database, the city name string to be identified is subjected to multi-level matching processing to correct non-standard city names that are incorrectly identified in the speech recognition text to standard city names; wherein, the preset database includes at least a city name database in the transportation field and a dynamically updated misspelling comparison table, the city name database in the transportation field is used as a benchmark reference for city name correction, and the misspelling comparison table is used to store the mapping relationship between non-standard city names and corresponding standard city names. The misspelling comparison table is dynamically updated based on the matching relationships generated during the correction process to improve the accuracy of city name correction in subsequent speech recognition.

[0071] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0072] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of this application, and are not intended to limit them; although the embodiments of this application have been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for improving the accuracy of speech recognition, characterized in that, include: Obtain the speech recognition text; wherein, the speech recognition text includes a string of the city name to be recognized; Based on a preset database, the city name string to be identified is subjected to multi-level matching processing to correct non-standard city names that are incorrectly identified in the speech recognition text to standard city names; wherein, the preset database includes at least a city name database in the transportation field and a dynamically updated misspelling comparison table, the city name database in the transportation field is used as a benchmark reference for city name correction, and the misspelling comparison table is used to store the mapping relationship between non-standard city names and corresponding standard city names. The misspelling comparison table is dynamically updated based on the matching relationships generated during the correction process to improve the accuracy of city name correction in subsequent speech recognition.

2. The method for improving the accuracy of speech recognition according to claim 1, characterized in that, The step of performing multi-level matching processing on the city name string to be identified based on a preset database, correcting non-standard city names that are incorrectly identified in the speech recognition text to standard city names, includes: The city name string to be identified is used as the first city name, and a first-level match is performed with the city name database in the transportation field to generate a first-level match result. If the first-level matching result indicates a successful match, the first city name is determined as the corrected standard city name.

3. The method for improving the accuracy of speech recognition according to claim 2, characterized in that, After using the city name string to be identified as the first city name and performing a first-level match with the city name database in the transportation field to generate a first-level match result, the method further includes: If the first-level matching result indicates that the match is unsuccessful, the first city name is matched with the misspelling comparison table to generate a second-level matching result.

4. The method for improving the accuracy of speech recognition according to claim 3, characterized in that, The step of performing a secondary match between the first city name and the misspelling comparison table to generate a secondary match result includes: Query the misspelling comparison table to see if there are any non-standard city name records that match the name of the first city, and generate the query results; If the query result indicates that there is a non-standard city name record in the misspelling comparison table that matches the first city name, the second city name corresponding to the first city name is obtained from the misspelling comparison table, and the second city name is returned to the first-level matching and rematched with the city name database in the transportation field.

5. The method for improving the accuracy of speech recognition according to claim 3, characterized in that, After performing a secondary match between the first city name and the misspelling comparison table to generate a secondary match result, the method further includes: If the second-level matching result indicates that the match is unsuccessful, proceed to the third-level matching process; The three-level matching process includes: processing the first city name with textual guidance to generate guidance text; inputting the guidance text into a large language model for recognition to obtain the third city name; and returning the third city name to the first-level matching to re-match it with the city name database in the transportation field.

6. The method for improving the accuracy of speech recognition according to claim 5, characterized in that, The step of processing the first city name with text guidance to generate guiding text includes: Based on the geographical naming characteristics in the transportation field, add a guiding prefix before the first city name; The guiding prefix is ​​used to indicate that the large language model recognizes the string as a city name.

7. The method for improving the accuracy of speech recognition according to claim 5, characterized in that, The step of dynamically updating the misspelling comparison table based on the matching relationships generated during the correction process includes: After generating the third city name in the three-level matching process, the third city name is returned to the first-level matching process for matching. If the name of the third city is inconsistent with the name of the first city, and the name of the third city is successfully matched to the city name database in the first-level matching, the mapping relationship between the name of the first city and the name of the third city is added to the misspelling comparison table.

8. The method for improving the accuracy of speech recognition according to claim 5, characterized in that, Before using the city name string to be identified as the first city name, the method further includes: The speech-recognized text is input into the large language model for keyword recognition, and the city name in the speech-recognized text is output as the city name string to be recognized. The city name string to be identified is stored in a cache library for priority retrieval during subsequent matching processes.

9. A device for improving the accuracy of speech recognition, characterized in that, include: An acquisition module is used to acquire speech recognition text; wherein, the speech recognition text includes a string of city names to be recognized; The matching module is used to perform multi-level matching processing on the city name string to be identified based on a preset database, and correct non-standard city names that are incorrectly identified in the speech recognition text to standard city names; wherein, the preset database includes at least a city name database in the transportation field and a dynamically updated misspelling comparison table, the city name database in the transportation field is used as a benchmark reference for city name correction, and the misspelling comparison table is used to store the mapping relationship between non-standard city names and corresponding standard city names; The enhancement module is used to dynamically update the misspelling comparison table based on the matching relationships generated during the correction process, so as to improve the accuracy of city name correction in subsequent speech recognition.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the speech recognition accuracy improvement method as described in any one of claims 1 to 8.

Images