Method for realizing automatic transcription of hot word extraction specification based on NLP

Abstract

The invention discloses an NLP-based method for realizing automatic transcription of hot word extraction specifications, which comprises the following steps of: presetting different scenes, defining hot word types in the scenes and constructing a common hot word library, and preprocessing and cleaning a document; automatically analyzing the cleaned context data, and automatically extracting hot words in the corresponding file content; the extracted hot words are intelligently classified, and a dynamic hot word library is constructed; meanwhile, a voice transfer result is transferred into a hot word verification module in real time; carrying out semantic analysis according to the context of the voice recognition result, judging the most coincident hot word, and carrying out replacement display on the extracted hot word result; according to the method, a normalized result of hot word replacement is displayed, a text result of speech recognition is automatically adjusted, weights of hot words in a dynamic hot word library are updated in real time, representative hot words are extracted from preprocessed texts based on a trained NLP semantic model by applying technologies such as a clustering algorithm and topic analysis, and the accuracy of transcription is improved.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and specifically provides a method for realizing automatic transcription of hot word extraction specifications based on NLP. Background Art

[0002] Traditional hot word extraction methods primarily include statistical methods, such as term frequency statistics (TF-IDF), and rule-based methods, such as regular expression matching. While these methods can achieve hot word extraction to a certain extent, they have numerous limitations. Traditional statistical and rule-based hot word extraction methods often focus solely on the frequency of occurrence or surface form of words, ignoring the semantic connections and contextual information between words. This results in insufficient accuracy in the extracted hot words and an inability to accurately reflect the core themes of the text. These traditional methods suffer from low computational efficiency when processing large-scale text data, making it difficult to meet the requirements of real-time applications. They also struggle to accurately identify representative hot words in semantically complex texts, and their processing capabilities for new and rare words are limited. When transcribing and normalizing extracted hot words, traditional methods typically rely on manually formulated rules and templates, lacking flexibility and adaptability, making it difficult to address the diverse needs of texts in different fields and styles. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for automatic transcription of hot word extraction specifications based on NLP to solve the problems raised in the above background technology.

[0004] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for realizing automatic transcription of hot word extraction specifications based on an NLP semantic model, comprising the following steps: Step S1: Preset different target scenarios, define the hot word types in the current scenario and build a common hot word library, and preprocess and clean the documents; Step S2: Automatically analyze the cleaned context data and automatically extract the hot words in the corresponding file content according to the content in the current text; Step S3: Intelligently classify the extracted hot words in the file and build a dynamic hot word library; Step S4: Build a real-time speech recognition engine, establish a double-buffered parallel processing module, perform segmented transcription of the speech stream and hotword matching asynchronously, and transfer the speech transcription results to the hotword verification module in real time; Step S5: Establish a context-aware replacement mechanism to perform semantic analysis based on the context of the speech recognition results, determine the most suitable hot words, and replace the extracted hot word results for display; Step S6: Display the normalized results of hot word replacement, automatically adjust the text results of speech recognition, and update the hot word weights in the dynamic hot word library in real time.

[0005] Preferably, the hot word types include company names, legal provisions, personal names, and chemical names; the main method of document preprocessing and cleaning is to extract text content and use semantic analysis to clean the content, remove garbled characters in the text, and unify the date format.

[0006] Preferably, the main method of intelligent classification of hot words in step S3 is to divide the hot words into multiple dimensions, and the multi-dimensional method includes hot word semantics and hot word pronunciation; if hot words are automatically classified according to semantics, the cosine similarity between word vectors is calculated, and the K-means clustering method is used to automatically cluster the hot word meanings, and a corresponding hot word library is constructed according to the classification results for transcription and call; if clustering is performed according to the hot word pronunciation, for hot words with similar or identical pronunciations, it is necessary to combine the context semantics, add the corresponding semantic understanding field when constructing the corresponding hot word library, and perform hot word matching in combination with the context.

[0007] Preferably, the speech recognition engine in step S4 is designed as a double-buffered parallel processing module, which performs segmented transcription of the speech stream and hot word matching asynchronously to ensure the real-time performance of the speech transcription, and transfers the result of the speech transcription to the hot word verification module in real time to perform hot word matching according to the context.

[0008] Preferably, the main steps of establishing the context-aware replacement mechanism in step S5 are: setting a hot word replacement threshold based on the imported speech recognition results; if two or more replaceable hot words appear in the hot word library, performing semantic analysis based on the context of the hot word to be determined and the context of the current speech recognition result, determining the most suitable hot word, and replacing and displaying the extracted hot word results.

[0009] Preferably, in step S6, incremental learning is performed based on user feedback. After the speech transcription is completed, the normalized results after the hot words are replaced are highlighted, the automatic normalization results of the hot words are confirmed or modified, and the user behavior data is recorded to automatically adjust the converted hot words or subsequent speech recognition text results, and the hot word weights in the hot word library are updated in real time.

[0010] Compared with the prior art, the present invention has the following beneficial effects: The present invention prevents transcription errors of homophones by maintaining a fixed hot word library of commonly used and standardized words and using NLP to extract hot words from nouns or characteristic phrases in the discussion scope; based on the trained NLP semantic model, the present invention uses clustering algorithms, topic analysis and other technologies to extract representative hot words from the preprocessed text to improve the accuracy of transcription. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 is a flow chart of the present invention; DETAILED DESCRIPTION

[0012] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Example

[0013] like Figure 1 As shown, the present invention provides a method for realizing automatic transcription and normalization of hot word extraction specifications based on NLP, comprising the following steps: Step S1: Preset different target scenarios, define the hot word types in the current scenario and build a common hot word library, and pre-process and clean the documents. For example, in a court scenario, it is often necessary to first collect some historical documents such as meeting minutes, project files, and other materials, and then define hot word types such as company names, legal provisions, personal names, chemical names, parties, time, institution names, etc. according to the document standards or specifications of the court scenario. When pre-processing the court documents, first extract the text content in the data and use semantic analysis tools to clean it, remove garbled characters in the text, and unify the date format, etc. If there are documents in image format, perform OCR recognition, extract the text content, and then perform text content cleaning again; Step S2: Automatically analyze the cleaned context data and extract hot words from the corresponding file content according to the content of the current text for automated intelligent extraction; Step S3: Intelligently classify the extracted hot words in the file and build a dynamic hot word library. By dividing the hot words into multiple dimensions, intelligently classify them according to the hot word semantics and hot word pronunciation. If the hot words are automatically classified according to semantics, the cosine similarity between word vectors is calculated, and the K-means clustering method is used to automatically cluster the hot word meanings. According to the classification results, the corresponding hot word library is constructed for transcription call; if the hot words are clustered according to the pronunciation, the hot words with similar or identical pronunciations need to be combined with the context semantics. When building the corresponding hot word library, the corresponding semantic understanding field is added, and the hot words are matched in combination with the context, so as to avoid confusion between homophones or words with similar pronunciations. The hot words in the dynamic hot word library can be updated dynamically at any time. For example, the hot words in the court can be classified according to the type of case, the stage of procedure, the participants or the relevant policies. If there are homophones in the hot words, such as "cross-examination" and "zhizheng", "final trial" and "lifelong", etc., the relationship between the semantics of the hot words and the context can be combined to better classify the hot words. For example, hot words that are easily confused, such as "ruling" and "award", "mediation" and "coordination", "fine" and "penalty", can be semantically analyzed according to the context to facilitate the call of hot words. Step S4: Build a real-time speech recognition engine and establish a double-buffered parallel processing module, that is, when performing speech-to-text conversion, there are some homophones or words with similar syllables, which are corrected during speech-to-text conversion, and the speech stream segmentation transcription and hot word matching are performed asynchronously. The speech transcription result is transferred to the hot word verification module in real time to ensure the real-time performance of the speech transcription. The result of the speech transcription is transferred to the hot word verification module in real time, that is, the hot word verification module is used to check whether the transcribed word is in the hot word library during speech-to-text conversion, or the recognition result of the speech recognition engine is replaced with the word in the corresponding hot word library, so as to perform hot word matching in the context; Step S5: Establish a context-aware replacement mechanism, perform semantic analysis based on the context of the above-mentioned speech recognition results, determine the most suitable hot words, and replace and display the extracted hot word results; the main method of replacement display is to set a hot word replacement threshold based on the imported speech recognition results. The hot word replacement threshold refers to judging whether the speech-transcribed text can be replaced by the hot words in the hot word verification module based on the set standard value. For example, if the standard value is set to 0.8, the speech recognition result is Jiangsu Zhongwei Technology Software Co., Ltd., and the hot word library contains Jiangsu Zhongwei Technology Software Co., Ltd., and this hot word exceeds 0.8, then this word is replaced with Jiangsu Zhongwei Technology Software Co., Ltd. If there are two or more replaceable hot words in the hot word library, semantic analysis is performed in combination with the context of the hot word to be determined and the context of the current speech recognition result to determine the most suitable hot word, and replace and display the extracted hot word results; Step S6: Incremental learning is performed based on user feedback. After the speech transcription is completed, the normalized results after the hot words are replaced are highlighted, the automatic normalization results of the hot words are confirmed or modified, and the converted hot words or subsequent speech recognition text results are automatically adjusted by recording user behavior data, and the hot word weights in the hot word library are updated in real time.

[0014] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments are to be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be encompassed therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A method for automatically transcribing hot word extraction specifications based on NLP, characterized in that: The following steps are involved: Step S1: Preset different target scenarios, define the hot word types in the current scenario and build a common hot word library, and preprocess and clean the documents; Step S2: Automatically analyze the cleaned context data and automatically extract the hot words in the corresponding file content according to the content in the current text; Step S3: Intelligently classify the extracted hot words in the file and build a dynamic hot word library; Step S4: Build a real-time speech recognition engine, establish a double-buffered parallel processing module, perform segmented transcription of the speech stream and hotword matching asynchronously, and transfer the speech transcription results to the hotword verification module in real time; Step S5: Establish a context-aware replacement mechanism to perform semantic analysis based on the context of the speech recognition results, determine the most suitable hot words, and replace the extracted hot word results for display; Step S6: Display the normalized results of hot word replacement, automatically adjust the text results of speech recognition, and update the hot word weights in the dynamic hot word library in real time.

2. The method for realizing automatic transcription of hot word extraction specifications based on NLP according to claim 1 is characterized in that: The hot word types include company names, legal provisions, personal names, and chemical names; the main method of document preprocessing and cleaning is to extract text content and use semantic analysis to clean the content, remove garbled characters in the text, and unify the date format.

3. The method for realizing automatic transcription of hot word extraction specifications based on NLP according to claim 1 is characterized in that: The main method of intelligent classification of hot words in step S3 is to divide hot words into multiple dimensions, including hot word semantics and hot word pronunciation; if hot words are automatically classified according to semantics, the cosine similarity between word vectors is calculated, and the K-means clustering method is used to automatically cluster the hot word meanings, and a corresponding hot word library is constructed according to the classification results for transcription call; If clustering is performed based on the pronunciation of hot words, hot words with similar or identical pronunciations need to be combined with contextual semantics. When building the corresponding hot word library, corresponding semantic understanding fields should be added to match hot words based on the context.

4. The method for realizing automatic transcription of hot word extraction specifications based on NLP according to claim 1 is characterized in that: In step S4, the speech recognition engine is designed as a double-buffered parallel processing module, which performs segmented transcription of the speech stream and hot word matching asynchronously to ensure the real-time performance of the speech transcription. The result of the speech transcription is transferred to the hot word verification module in real time, and hot word matching is performed in the context.

5. The method for realizing automatic transcription of hot word extraction specifications based on NLP according to claim 1 is characterized in that: The main steps of establishing the context-aware replacement mechanism in step S5 are: setting a hot word replacement threshold based on the imported speech recognition results; if two or more replaceable hot words appear in the hot word library, performing semantic analysis based on the context of the hot word to be determined and the context of the current speech recognition result, determining the most suitable hot word, and replacing and displaying the extracted hot word results.

6. The method for realizing automatic transcription of hot word extraction specifications based on NLP according to claim 1 is characterized in that: In step S6, incremental learning is performed based on user feedback. After the speech transcription is completed, the normalized results after the hot words are replaced are highlighted, the automatic normalization results of the hot words are confirmed or modified, and the user behavior data is recorded to automatically adjust the converted hot words or subsequent speech recognition text results, and the hot word weights in the hot word library are updated in real time.

Images