License plate number text correction method, device, equipment, medium and program product
By using reverse search and error correction to process the speech-to-text license plate number, the problem of low accuracy in license plate number recognition during car insurance telephone services has been solved, achieving efficient and accurate license plate number acquisition and information retrieval.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SUNSHINE DIGITAL INTELLIGENCE TECHNOLOGY CO LTD
- Filing Date
- 2026-03-06
- Publication Date
- 2026-06-09
AI Technical Summary
In the automotive industry's car insurance telephone service scenario, the accuracy of license plate recognition is low due to insufficient accuracy of speech-to-text conversion and the influence of customers' language habits, making it difficult to directly obtain accurate license plate numbers from speech-recognized text.
By reverse-searching the speech recognition text, starting from the last preset license plate number length, candidate text is gradually obtained and error correction is performed, including dividing the first and second segments into non-first and second segments, detecting and correcting anomalies such as homophones, homonyms, abbreviated quantity descriptions, irregular inputs and noisy inputs, until the license plate number length requirement is met, and finally review and correction are performed.
It improves the accuracy of license plate number recognition, ensuring that the corrected license plate numbers meet the standards and can be effectively used for information retrieval.
Smart Images
Figure CN122174830A_ABST
Abstract
Claims
1. A method for correcting errors in license plate number text, characterized in that, The method includes: The speech recognition text of the license plate number is obtained, and the search is performed backward from the preset length of the license plate number in the speech recognition text to the first character used to represent the region, so as to obtain the initial candidate text; The initial candidate text is corrected to obtain the initial candidate license plate number; In response to the candidate license plate number being less than the preset license plate number length and the reverse search not reaching the starting position of the speech recognition text, the speech recognition text is reverse searched to the next character used to represent the region to obtain updated candidate text, and the updated candidate text is corrected to obtain updated candidate license plate numbers. The reverse search stops when the candidate license plate number meets the preset license plate number length requirement or the reverse search reaches the starting position of the speech recognition text, and the latest candidate license plate number is used as the pending license plate number. The undetermined license plate number is reviewed and corrected to obtain the standard license plate number.
2. The method according to claim 1, characterized in that, The step of correcting the initial candidate text to obtain the initial candidate license plate numbers includes: The candidate text is divided into first-order segments and non-first-order segments; Anomaly detection is performed on the first segment and the non-first segment to obtain first-first anomaly information and non-first-first anomaly information. The first-first anomaly information includes at least one of homophone confusion and homophone letter confusion. The non-first-first anomaly information includes at least one of homophone letter confusion, abbreviation quantity description, irregular input, and noise input. The first-order abnormal information and the non-first-order abnormal information are processed in a targeted manner to update the first-order segment and the non-first-order segment. Based on the update result, the latest first-order segment and the non-first-order segment are spliced together to obtain the initial candidate license plate number.
3. The method according to claim 2, characterized in that, Anomaly detection is performed on the non-first-order segments to obtain non-first-order anomaly information, including: In response to the presence of pre-defined obfuscated characters in the non-first-order segment, non-first-order abnormal information representing the presence of homophone obfuscation is obtained; In response to the presence of a preset quantifier in the non-first segment, where the preceding character of the preset quantifier is a number and the following character is a number or a letter, non-first segment abnormal information representing the presence of abbreviated quantity description is obtained; The non-first-order segment is annotated with phonetic symbols to obtain the pinyin of each character. In response to the intersection between the pinyin of the non-first-order segment and the pinyin of the preset irregular pinyin mapping table, non-first-order abnormal information representing the existence of irregular input is obtained. Obtain characters in the non-first-order segment that do not belong to the preset non-first-order segment character set to obtain noise characters. In response to the noise characters including at least one character, obtain non-first-order anomaly information representing the presence of noise characters.
4. The method according to claim 3, characterized in that, In response to the presence of non-first-order anomaly information described in abbreviated quantities, the method further includes: Extract the quantifier and the character before and after it to obtain the abbreviated quantity description text; The abbreviated quantity description text is replaced with a target number of repeated target characters to update the non-first-end segment, wherein the target number is the starting character of the abbreviated quantity description text and the target character is the last character of the abbreviated quantity description text.
5. The method according to claim 3, characterized in that, In response to the presence of non-first-order abnormal information of irregular input, the intersection of the non-first-order pinyin and the pinyin of the preset irregular pinyin mapping table is obtained, and the original text corresponding to the intersection pinyin in the non-first-order segment is obtained. According to the preset irregular pinyin mapping table, the original text is mapped to the standard character corresponding to the intersection pinyin to update the non-first-order segment. In response to a non-first-order exception message indicating the presence of a noisy character, the noisy character is deleted to update the non-first-order segment.
6. The method according to claim 2, characterized in that, Anomaly detection is performed on the first and second segments to obtain the first and second anomaly information, including: In response to the first segment being a preset confused letter, first segment abnormal information representing the presence of homophone confusion is obtained; In response to the first segment being text and the text not belonging to a preset region set, first segment abnormal information indicating homophone confusion is obtained. The preset region set is obtained by extracting the text of each standard region from a preset pinyin region mapping table.
7. The method according to claim 6, characterized in that, In response to the first-order abnormal information indicating the presence of homophones, the preset confused letter is replaced with a preset confused character that is homophonous with the preset confused letter, thereby updating the first-order segment; In response to the first-character error message indicating homophonic confusion, the standard regional character corresponding to the pinyin of the first character is obtained by filtering from a preset pinyin regional mapping table, and the text of the first-character segment is replaced with the standard regional character to update the first-character segment.
8. The method according to claim 1, characterized in that, The initial candidate text obtained by reverse searching from the pre-defined length of the license plate number of the speech-recognized text to the first character representing the region includes: The speech-recognized text is annotated with pinyin to obtain the pinyin corresponding to each character; Based on the pinyin corresponding to each character, a reverse search is performed starting from the pre-preset length of the license plate number in the speech recognition text. The first character found is taken as the initial regional character, wherein the pinyin of the regional character belongs to a pre-preset set of regional pinyin or a pre-preset set of confused letters. The speech recognition text is extracted using the initial region character as the starting character to obtain the initial candidate text.
9. A vehicle license plate number text correction device, characterized in that, The device includes: The acquisition module is used to acquire the speech recognition text of the license plate number, and to search backwards from the preset length of the license plate number in the speech recognition text to the first character used to represent the region, so as to obtain the initial candidate text; The error correction processing module is used to perform error correction processing on the initial candidate text to obtain the initial candidate license plate number; The iterative search module is used to respond to the situation where the candidate license plate number is less than the preset license plate number length and the reverse search has not reached the starting position of the speech recognition text, to reverse search the speech recognition text to the next character used to represent the region, to obtain updated candidate text, and to perform error correction processing on the updated candidate text to obtain updated candidate license plate numbers. The search termination judgment module is used to stop the reverse search and use the latest candidate license plate number as the pending license plate number when the candidate license plate number meets the preset license plate number length requirement or the reverse search reaches the beginning position of the speech recognition text; and The verification module is used to verify and correct the undetermined license plate number to obtain a standard license plate number.
10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 8.