Speech recognition error correction processing method and system, storage medium and program product

Abstract

The invention provides a speech recognition error correction processing method and system, a storage medium and a program product. The method comprises the following steps: acquiring original voice data; converting the original voice data into a first text; presetting a terminology prompt based on the first text; performing error correction processing on the first text by using a large language model to obtain a second text; and performing formal conversion on the second text by using the large language model to obtain a final text. According to the method, the terminology recognition accuracy can be improved, the oral text is converted into the formal text, the manual correction cost and time are reduced, and the specialty and usability of the text are improved. In a word, the ASR system conversion efficiency, the accuracy of processing professional nouns and industry specific nouns and the text formalization degree can be remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and in particular to a speech recognition error correction processing method, system, storage medium and program product. Background Art

[0002] In modern information technology, automatic speech recognition (ASR) technology is a core technology that allows computer systems to execute commands or transcribe text by recognizing and understanding user voice input. Although ASR technology has made significant progress and can provide high-accuracy speech-to-text conversion in a variety of environments, due to the complexity and diversity of speech signals, speech recognition technology often makes mistakes in practical applications, such as recognition errors, misidentification, etc., and there are still major challenges when dealing with professional terminology or industry-specific terms.

[0003] Traditional ASR systems mainly rely on extensive language models and acoustic models to process speech data. These models perform well in daily language communication, but there are still the following problems: First, when recognizing professional terms, especially words that do not often appear in daily conversations, recognition errors often occur. Because professional terms often appear less frequently in the training data corpus of the speech recognition model, it is difficult for the model to effectively learn and correct them. Second, during the speech recognition process, some speech signals may be lost or distorted. For some sentences with complex contexts, the ASR model may misunderstand their meaning, which will also affect the subsequent error correction effect. Third, even if the recognition is correct, traditional ASR systems can only provide colloquial text output, which is not suitable for application scenarios that require highly formalized text, such as legal, medical, and scientific report writing. In these fields, the formality, accuracy, and professionalism of the text are crucial, and any minor errors or colloquial expressions may lead to misunderstandings or legal liability issues.

[0004] Existing solutions to the above problems usually include manual correction of recognition results and the use of domain-specific speech recognition models. However, manual correction is costly and inefficient, and domain-specific models require a large amount of professional corpus to train, which is impractical in many cases. These limitations highlight the shortcomings of existing technologies in accurately transcribing speech containing professional terms or specific industry terms and providing formal text output. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a speech recognition error correction processing method, system, storage medium and program product. Through pre-set professional terminology prompts, the large language model can not only identify and correct errors when processing the converted text, but also professionalize the text to meet the needs of professional texts in multiple fields such as law, medicine, and technology.

[0006] In a first aspect, the present invention provides a method for correcting speech recognition errors, comprising the following steps:

[0007] Get the original voice data;

[0008] Converting raw speech data into a first text;

[0009] Preset professional terminology prompts based on the first text;

[0010] Using a large language model to perform error correction processing on the first text to obtain a second text;

[0011] The second text is formally transformed using a large language model to obtain the final text.

[0012] As a further improvement of the present invention, the step of converting the original voice data into the first text includes:

[0013] Performing preprocessing operations of denoising, enhancing and normalizing the original speech data to obtain preprocessed speech data;

[0014] The ASR model is used to convert the preprocessed speech data into a first text.

[0015] As a further improvement of the present invention, the method of converting the pre-processed speech data into the first text using the ASR model includes:

[0016] Segmenting the preprocessed speech data into audio segments;

[0017] Each audio segment is fed into the ASR model for conversion to obtain the first text.

[0018] As a further improvement of the present invention, the preset professional terminology prompt based on the first text includes:

[0019] Compile a professional terminology database based on professional terms in professional fields;

[0020] Automatically detect the professional field involved in the first text, and select relevant professional terminology sets from the professional terminology library to preset professional terminology prompts.

[0021] As a further improvement of the present invention, the step of using a large language model to perform error correction processing on the first text to obtain a second text includes:

[0022] Inputting the first text and the professional term prompt into the large language model together;

[0023] Based on the large language model and professional terminology prompts, incorrect professional terminology, contextual coherence and semantic errors in the first text are identified and corrected;

[0024] A second text is generated in which the errors in the professional terminology have been corrected.

[0025] As a further improvement of the present invention, the step of using the large language model to formally transform the second text to obtain the final text includes:

[0026] Analyze the second text to identify all colloquial expressions and structures;

[0027] The large language model converts colloquial expressions and structures into formal expressions and structures;

[0028] The formalized expressions and structures are formatted and proofread to obtain the final text.

[0029] As a further improvement of the present invention, the large language model converts colloquial expressions and structures into formal expressions and structures, and the conversion method includes replacing colloquial terms, adjusting grammatical structures, standardizing words and unifying styles.

[0030] In a second aspect, the present invention provides a computer system comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in the first aspect.

[0031] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0032] In a fourth aspect, the present invention provides a computer program product, comprising a computer program, which implements the steps of the method described in the first aspect when executed by a processor.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] First, the present invention can improve the accuracy of professional terminology recognition. Traditional ASR systems often make mistakes when recognizing terms related to specific professional fields. The present invention effectively improves the recognition accuracy of professional terms by integrating a large language model and combining professional term preset prompts.

[0035] Second, the present invention can convert colloquial text into formal text. The text output by the standard ASR system is usually colloquial and not suitable for professional fields that require formal text. The present invention uses the natural language understanding ability of the large language model to convert the colloquial text into formal text that meets professional standards.

[0036] Third, the present invention can reduce the cost and time of manual correction. By automating the process of identifying errors and text conversion, the present invention reduces the need for manual intervention, thereby significantly reducing costs and increasing processing speed.

[0037] Fourth, the present invention can improve the professionalism and usability of the text. By ensuring the accuracy of the recognized content and improving the formality of the text, the present invention improves the usability of the text, especially in the fields of law, medicine and scientific research.

[0038] In summary, the present invention can significantly improve the conversion efficiency of the ASR system, the accuracy of processing professional terms and industry-specific terms, and the degree of text formalization. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The present invention discloses a flowchart of a method for correcting speech recognition errors. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the present invention will combine the accompanying drawings to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, wherein steps S1, S2... in the embodiments described in the present invention do not limit the only execution steps of the present invention; the various models described in the present invention are not the only limiting methods of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0041] In the present invention, computer devices / equipment / systems refer to related entities applied to computers, such as hardware, a combination of hardware and software, software or software in execution, etc. In detail, for example, software includes but is not limited to a process running on a processor, a processor, an object, executable software, an execution thread, a program and / or a computer. In addition, an application or script program running on a server, a server can also be software. One or more software can be in an execution process and / or thread, and the software can be localized on one computer and / or distributed between two or more computers, and can be run by various computer-readable media.

[0042] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application may be combined with each other.

[0043] Example 1

[0044] In a first aspect, an embodiment of the present invention provides a method for correcting speech recognition errors. Figure 1 As shown, the specific process can be as follows:

[0045] S1: Get original voice data;

[0046] S2: Converting the original voice data into a first text;

[0047] Preferably, converting the original voice data into the first text comprises:

[0048] S21: performing preprocessing operations of denoising, enhancing and normalizing the original speech data to obtain preprocessed speech data;

[0049] In the embodiment of the present invention, the captured raw voice data undergoes a preprocessing stage, including denoising, enhancement and normalization operations, which may specifically be signal amplification, filtering and digitization of sound signals. The purpose of this step is to optimize the clarity and recognition efficiency of the voice signal and provide more accurate input data for subsequent automatic speech recognition (ASR) processing. The preprocessed digital voice data is converted into a format suitable for ASR model processing, such as a wave file or a specific audio coding format.

[0050] S22: Convert the pre-processed speech data into a first text using an ASR model.

[0051] In the embodiment of the present invention, the application of the ASR model is a key step in converting the pre-processed speech data into text. The ASR model in the embodiment of the present invention is based on the Transformer structure to improve the understanding and transcription accuracy of the speech signal.

[0052] Preferably, the converting the pre-processed speech data into the first text by using the ASR model comprises:

[0053] S221: Configuring the ASR model to accept pre-processed speech data as input;

[0054] In the embodiment of the present invention, the ASR model is specially trained to handle speech inputs containing various accents, dialects, and industry-specific terms. In order to adapt to different languages ​​and professional terms, the ASR model training includes a large number of speech samples and corresponding text transcriptions to ensure that the model can cover a wide range of speech expressions.

[0055] S222: Segment the pre-processed voice data into audio segments;

[0056] In an embodiment of the present invention, the pre-processed speech data is first divided into smaller audio segments. These segments are fed into the ASR model for processing one by one. The model uses the language pattern recognition ability acquired during its training to convert the audio segments into corresponding text outputs. This step may include complex algorithms to ensure transcription accuracy, such as sound feature extraction, sound to text mapping, and context understanding. These algorithms are not the only way to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0057] S223: Send each audio segment to the ASR model for conversion to obtain a first text.

[0058] Through the above detailed steps, the ASR model in the present invention can provide efficient and accurate speech-to-text transcription services to meet specific application requirements and changing environmental conditions.

[0059] S3: preset professional terminology prompts based on the first text;

[0060] In this step, compiling and presetting professional terminology prompts can improve the ASR model's recognition accuracy for specific professional terms and industry-specific terms. This step creates a professional terminology library and uses these terms as prompts during the speech-to-text process to guide the large language model to process and convert relevant content more accurately and professionally. Especially in application scenarios containing highly specialized terms, the quality and applicability of the final output are ensured.

[0061] Preferably, the preset professional terminology prompt based on the first text includes:

[0062] S31: Compile a professional terminology database based on professional terms in professional fields;

[0063] In the embodiment of the present invention, corresponding term lists are collected and compiled for different professional fields (such as medical, legal, technology, etc.). These term lists are reviewed and confirmed by experts to ensure that the terms included are accurate and up-to-date. The establishment of the term library not only includes the words themselves, but also covers their variants, synonyms and related expressions to cover the diversity and complexity of language use. Each term is marked with a clear definition and usage context to help the language model accurately apply these words in different scenarios.

[0064] S32: Automatically detect the professional field involved in the first text, and select a set of related professional terms from a professional term library to preset professional term prompts.

[0065] In an embodiment of the present invention, based on the first text content output by the ASR model, the system automatically detects the professional field involved, and selects a relevant set of terms from the professional terminology library to form professional terminology prompts. These professional terminology prompts are presented in a specific format, for example: "Note: The text may contain the following professional terms: [list of terms]. Please pay special attention when processing." The professional terminology prompts will be directly input into the subsequent large language model processing steps as a reference for text processing and correction.

[0066] S4: using the large language model to perform error correction processing on the first text to obtain a second text;

[0067] In this step, the text output by the automatic speech recognition (ASR) model is deeply corrected using a large language model (LLM). This step involves integrating the appropriate language model, performing text correction, and finally outputting the corrected text. This process not only improves the accuracy of the text, but also enhances its professionalism and formality, making it suitable for a variety of professional applications, especially in situations where high-standard professional texts are required, such as legal documents, medical reports, and other fields.

[0068] Preferably, the using a large language model to perform error correction processing on the first text to obtain the second text includes:

[0069] S41: Selecting and configuring large language models;

[0070] In this step, select a suitable large language model, such as a model based on the Transformer architecture, which can be GPT, Llama and other models. These models are selected for their excellent language understanding and generation capabilities.

[0071] In the embodiment of the present invention, the Llama model is selected and customized to meet specific application requirements, such as enhancing the understanding and application of professional terms. This step includes detailed parameter adjustment of text data introduced in a large amount of professional fields to optimize the model's processing of specific terms.

[0072] S42: inputting the first text and the professional term prompt into the large language model together;

[0073] S43: Based on the large language model and professional terminology prompts, the incorrect professional terminology, contextual coherence and semantic errors in the first text are identified and corrected;

[0074] In the embodiment of the present invention, the model will provide correct alternative text for the errors detected to ensure the professionalism and logical consistency of the output. During the text error correction process, the model can also prioritize the correct use of professional terms based on the term list in the professional term prompt, thereby improving the professional accuracy of the overall text.

[0075] The text that has been processed through the model will be thoroughly checked to confirm that all proposed corrections have been accurately implemented and to ensure the smoothness and professionalism of the text.

[0076] S44: Generate a second text in which errors in professional terminology have been corrected.

[0077] The corrected second text will be output to the next processing stage or directly provided to the user. This output not only includes updating the text content, but also involves ensuring that the format and structure meet the requirements of professional documents.

[0078] S5: Use the large language model to formalize the second text to obtain the final text;

[0079] This step further converts and optimizes the second text after error correction by the large language model, transforming it from spoken expression into formal written text, while ensuring that the format and structure of the text meet professional standards.

[0080] Preferably, the step of using the large language model to formally transform the second text to obtain the final text includes:

[0081] S51: Analyze the second text and identify all colloquial expressions and structures;

[0082] S52: The large language model converts spoken expressions and structures into formal expressions and structures;

[0083] Preferably, the large language model converts colloquial expressions and structures into formal expressions and structures, and the conversion method includes replacing colloquial terms, adjusting grammatical structures, standardizing words and unifying styles.

[0084] In an embodiment of the present invention, this step may include replacing informal expressions, adjusting grammatical structures to conform to the rigor of written language, and using words and phrases that are more suitable for written expression. The conversion process utilizes the generative power of a large language model to ensure that the converted text is not only grammatically correct, but also semantically accurate and coherent.

[0085] S53: Format and proofread the formalized expression and structure to obtain the final text.

[0086] This step adjusts the layout of the text, such as paragraph division, title setting, list layout, etc., to ensure that it meets the requirements of the specific document format. In addition, in an embodiment of the present invention, some automatic proofreading tools are used to carefully check the text to correct spelling errors, grammatical errors and any subtle formatting problems. Further quality control is performed to ensure that all modifications meet high standards of professionalism. Ultimately, the formatted and proofread text will be confirmed as the final version, ready for publication or submission to the end user.

[0087] S6: Design system architecture and user interface.

[0088] This step involves the overall system architecture design, user interface design and implementation, and comprehensive testing and optimization of the system to ensure the efficiency, usability, and stability of the system.

[0089] Preferably, a modular system architecture is designed to ensure that the various processing steps from step S1 to S5 (such as ASR processing, text error correction, text formalization, etc.) can work together efficiently.

[0090] In the embodiment of the present invention, the system architecture supports multi-layer design, including data layer, logic layer and presentation layer, so as to facilitate management and expansion. In the data layer, the data storage and management mechanism is configured to ensure that all input voice data and output text data are stored safely and easily accessible. In the logic layer, the core functional modules such as voice data processing, text conversion and error detection are implemented. Efficient algorithms and models are used to ensure processing speed and accuracy. In the presentation layer, a clear user interface is developed to allow users to easily manage and monitor the entire text conversion process.

[0091] Preferably, design an intuitive and easy-to-use user interface that enables users to simply upload voice files, view the conversion progress, and receive the final text output. The interface supports multiple operating systems and devices, including desktop and mobile devices. Provide user configuration options to allow users to adjust settings such as voice recognition preferences, the scope of application of professional terminology, etc. as needed. Introduce interactive features such as real-time feedback and modification options to enable users to provide input and corrections during processing, enhancing the flexibility of the system and user satisfaction.

[0092] In this step, the system not only provides efficient and accurate services, but also ensures the user-friendly interface and long-term maintainability of the system, thus meeting the high standards required in commercial environments and professional applications.

[0093] Example 2

[0094] In a second aspect, an embodiment of the present invention provides a computer system, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in Embodiment 1 of the present invention.

[0095] Example 3

[0096] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored, characterized in that when the computer program is executed by a processor, the steps of the method described in Embodiment 1 of the present invention are implemented.

[0097] Example 4

[0098] In a fourth aspect, an embodiment of the present invention provides a computer program product, including a computer program, characterized in that when the computer program is executed by a processor, the steps of the method described in Embodiment 1 of the present invention are implemented.

[0099] The speech recognition error correction processing method, system, storage medium and program product provided by the present invention can improve the accuracy of professional terminology recognition, convert colloquial text into formal text, reduce the cost and time of manual correction, and improve the professionalism and usability of the text. In short, the present invention can significantly improve the conversion efficiency of the ASR system, the accuracy of processing professional terms and industry-specific terms, and the degree of text formalization.

Claims

1. A speech recognition error correction processing method, characterized in that: The following steps are involved: Get the original voice data; Converting raw speech data into a first text; Preset professional terminology prompts based on the first text; Using a large language model to perform error correction processing on the first text to obtain a second text; The second text is formally transformed using a large language model to obtain the final text.

2. The speech recognition error correction processing method according to claim 1, characterized in that: The converting the original voice data into the first text comprises: Performing preprocessing operations of denoising, enhancing and normalizing the original speech data to obtain preprocessed speech data; The ASR model is used to convert the preprocessed speech data into a first text.

3. The speech recognition error correction processing method according to claim 2, characterized in that: The method of converting the pre-processed speech data into a first text by using an ASR model includes: Segmenting the preprocessed speech data into audio segments; Each audio segment is fed into the ASR model for conversion to obtain the first text.

4. The speech recognition error correction processing method according to claim 1, characterized in that: The preset professional terminology prompt based on the first text includes: Compile a professional terminology database based on professional terms in professional fields; Automatically detect the professional field involved in the first text, and select relevant professional terminology sets from the professional terminology library to preset professional terminology prompts.

5. The speech recognition error correction processing method according to claim 1, characterized in that: The step of using the large language model to perform error correction processing on the first text to obtain the second text includes: Inputting the first text and the professional term prompt into the large language model together; Based on the large language model and professional terminology prompts, incorrect professional terminology, contextual coherence and semantic errors in the first text are identified and corrected; A second text is generated in which terminology errors are corrected.

6. The speech recognition error correction processing method according to claim 1, characterized in that: The step of using the large language model to formally transform the second text to obtain a final text includes: Analyze the second text to identify all colloquial expressions and structures; The large language model converts colloquial expressions and structures into formal expressions and structures; The formalized expressions and structures are formatted and proofread to obtain the final text.

7. The speech recognition error correction processing method according to claim 6, characterized in that: The large language model converts colloquial expressions and structures into formal expressions and structures, and the conversion method includes replacing colloquial terms, adjusting grammatical structures, standardizing words and unifying styles.

8. A computer system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images