Method and device for generating Tibetan scene-based audio corpus based on AIGC
By automatically generating and translating Tibetan contextual audio corpora using AIGC technology, the problems of difficulty and low efficiency in collecting Tibetan contextual audio corpora are solved. A standardized corpus dataset is provided to support the rapid training and recognition of Tibetan speech recognition technology.
Patent Information
- Application Number
- CN202311520471.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-14
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2043-11-14
AI Technical Summary
Collecting Tibetan language contextual audio corpora is difficult and inefficient. Existing technologies are unable to perform text arrangement and recording in different scenarios, which raises concerns about information sensitivity and privacy.
Based on AIGC technology, the system automatically generates and classifies Tibetan contextual audio corpora by establishing a knowledge base, extracting keywords, generating Chinese contextualized text, translating and recording Tibetan contextualized audio corpora.
It significantly improves the generation efficiency and standardized management of Tibetan contextual audio corpora, and provides standard and regulated corpus datasets, which are suitable for rapid training and recognition of Tibetan speech recognition technology modules.
Smart Images

Figure CN117520479B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of corpus generation, and in particular to a method and device for generating Tibetan language scenario audio corpus based on AIGC (Artificial Intelligence Generated Content). Background Art
[0002] Since the Tibetan language population is relatively sparse and the application scenarios of speech are relatively scarce, Tibetan data collection projects are mainly concentrated on recording by students in universities or recruiting part-time staff in the society to record audio corpus under the demand of scientific research projects. Some of them use real-life speech recordings and then have the audio cleaned and annotated by specialized Tibetan annotators to produce Tibetan audio corpus datasets.
[0003] Since the main method of collecting Tibetan language data is to have a group of people read and record it, this method requires recording in specific sub-fields when collecting data in specialized fields (such as the medical field). However, it is very difficult to arrange texts in different scenarios. Going to the field to record and obtain original data has privacy issues during the information-sensitive approval and collection process, making it difficult to record in a centralized manner.
[0004] In addition, the current voice data collection efficiency is low, and the speed is greatly limited by personnel experience and efficiency. Summary of the invention
[0005] In order to at least overcome the problems of difficulty in collecting Tibetan scenario audio corpus and low collection efficiency in the related art to a certain extent, the present application provides a method and device for generating Tibetan scenario audio corpus based on AIGC.
[0006] The scheme of this application is as follows:
[0007] According to a first aspect of an embodiment of the present application, a method for generating Tibetan scenario audio corpus based on AIGC is provided, comprising:
[0008] Establish a corresponding knowledge base based on the demand areas of Tibetan audio corpus;
[0009] Extracting keywords from the knowledge base;
[0010] Based on the AIGC model, generate Chinese scenario-based text corpus corresponding to the demand field according to the extracted keywords;
[0011] Based on the translation engine, the generated Chinese scenario text corpus is translated into Tibetan scenario text corpus;
[0012] Acquire Tibetan scenario-based audio corpus recorded according to the Tibetan scenario-based text corpus, and bind the Tibetan scenario-based audio corpus and the Tibetan scenario-based text corpus;
[0013] The bound Tibetan scenario-based audio corpus and Tibetan scenario-based text corpus are classified and stored according to the required fields and scenarios.
[0014] Preferably, a corresponding knowledge base is established according to the demand field of Tibetan audio corpus, including:
[0015] Divide the process links of the demand areas;
[0016] Relevant textual knowledge is obtained for different process links, and a knowledge base corresponding to the required field is established;
[0017] The established knowledge base is summarized and labeled according to the dictionary content index.
[0018] Preferably, extracting keywords from the knowledge base includes:
[0019] Based on NLU (Natural Language Understanding) technology, domain detection, intent recognition and slot filling are performed on the dictionary text in the knowledge base to extract keywords;
[0020] Classify and tag the extracted keywords.
[0021] Preferably, based on the AIGC model, Chinese scenario-based text corpus corresponding to the demand field is generated according to the extracted keywords, including:
[0022] Writing a prompt text of the AIGC model according to the extracted keywords; the prompt text is a short text that interacts with the AIGC model; the prompt text is used to provide instructions to the AIGC model to guide the AIGC model to generate text answer content related to the keywords;
[0023] The prompt text is input into the AIGC model, so that the AIGC model generates Chinese scenario text corpus corresponding to the demand field.
[0024] Preferably, the method further comprises:
[0025] After the Chinese scenario text corpus is generated, synonyms and near-synonyms are extracted and converted for the prompt text, and the replaced prompt text is re-input into the AIGC model to generate multiple groups of Chinese scenario text corpora.
[0026] Preferably, the method further comprises:
[0027] The prompt text is optimized in terms of prompt structure and content.
[0028] Preferably, before translating the generated Chinese scenario-based text corpus into Tibetan scenario-based text corpus, the method further includes:
[0029] Review the logic, compliance, usability, accuracy and repeatability of the Chinese scenario text corpus generated by the AIGC model;
[0030] The approved Chinese scenario text corpus is segmented into sentences.
[0031] Preferably, the Tibetan scenario audio corpus is recorded based on preset scenario simulation parameters;
[0032] The scenario simulation parameters include at least: recording sampling rate, environmental noise during recording, and amplitude during recording.
[0033] Preferably, the method further comprises:
[0034] The keywords to be extracted
[0035] Based on the translation engine, the extracted keywords are translated into Tibetan form;
[0036] The extracted keywords and the translated Tibetan keywords are cleaned and then entered into the knowledge base.
[0037] According to a second aspect of an embodiment of the present application, there is provided a Tibetan scenario audio corpus generation device based on AIGC, comprising:
[0038] Processor and memory;
[0039] The processor and the memory are connected via a communication bus:
[0040] Wherein, the processor is used to call and execute the program stored in the memory;
[0041] The memory is used to store a program, and the program is at least used to execute a method for generating Tibetan scene-based audio corpus based on AIGC as described in any one of the above items.
[0042] The technical solution provided by the present application may include the following beneficial effects: The method for generating Tibetan scenario audio corpus based on AIGC in the present application includes: establishing a corresponding knowledge base according to the demand field of Tibetan audio corpus; extracting keywords from the knowledge base; generating Chinese scenario text corpus corresponding to the demand field according to the extracted keywords based on the AIGC model; translating the generated Chinese scenario text corpus into Tibetan scenario text corpus based on the translation engine; obtaining Tibetan scenario audio corpus recorded according to the Tibetan scenario text corpus, binding the Tibetan scenario audio corpus and the Tibetan scenario text corpus; classifying the bound Tibetan scenario audio corpus and the Tibetan scenario text corpus into the library according to the demand field and scenario. The technical solution in the present application, by extracting keywords in the required Tibetan audio corpus field, automatically generates, outputs and translates the corresponding Tibetan scenario text corpus according to the keywords, the technical solution in the present application can significantly improve the generation efficiency of Tibetan scenario audio corpus. The generated Tibetan scenario audio corpus standards and specifications are conducive to the standardized management of Tibetan scenario audio corpus.
[0043] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0045] Figure 1 It is a flowchart of a method for generating Tibetan language scenario audio corpus based on AIGC provided by an embodiment of the present application;
[0046] Figure 2 This is a schematic diagram of the content structure of a medical knowledge base provided by an embodiment of the present application;
[0047] Figure 3 This is a schematic diagram of the interaction between an AIGC model and a prompt text provided by an embodiment of the present application;
[0048] Figure 4 It is a structural diagram of a Tibetan language scenario audio corpus generation device based on AIGC provided by an embodiment of the present application.
[0049] Reference numerals: processor-21; memory-22. DETAILED DESCRIPTION
[0050] Here, exemplary embodiments will be described in detail, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Instead, they are only examples of methods consistent with some aspects of the present application as detailed in the attached claims.
[0051] Embodiment 1
[0052] Figure 1 This is a flowchart of a method for generating Tibetan language scenario audio corpus based on AIGC provided by an embodiment of the present application, with reference to Figure 1 , a method for generating Tibetan scene audio corpus based on AIGC, comprising:
[0053] S11: Establish a corresponding knowledge base based on the demand areas of Tibetan audio corpus;
[0054] S12: extract keywords from the knowledge base;
[0055] S13: Based on the AIGC model, generate Chinese scenario-based text corpus corresponding to the demand field according to the extracted keywords;
[0056] S14: Based on the translation engine, the generated Chinese scenario text corpus is translated into Tibetan scenario text corpus;
[0057] S15: Acquire Tibetan scenario-based audio corpus recorded according to the Tibetan scenario-based text corpus, and bind the Tibetan scenario-based audio corpus and the Tibetan scenario-based text corpus;
[0058] S16: Classify and store the bound Tibetan scenario-based audio corpus and Tibetan scenario-based text corpus according to the required fields and scenarios.
[0059] It should be noted that the corresponding knowledge base is established according to the demand field of Tibetan audio corpus, including:
[0060] Divide the process links into demand areas;
[0061] Obtain relevant text knowledge for different process links and establish a knowledge base corresponding to the required field;
[0062] The established knowledge base is summarized and labeled according to the dictionary content index.
[0063] In practice, we first need to establish a corresponding knowledge base for the demand areas of Tibetan audio corpus. Take the medical field as an example:
[0064] The content structure of the collected medical knowledge base is: first-level department category, second-level department category, symptoms and symptom knowledge text, such as Figure 2 As shown, the internal medicine - gastroenterology - stomach pain - symptom knowledge text is a hierarchical structure of the medical knowledge base content.
[0065] It should be noted that keyword extraction from the knowledge base includes:
[0066] Based on NLU technology, the dictionary text in the knowledge base is subjected to domain detection, intent recognition and slot filling to extract keywords;
[0067] Classify and tag the extracted keywords.
[0068] In specific practice, NLU technology is used to analyze and summarize dictionary texts to extract keywords, which mainly includes three parts: domain detection (domain), intent recognition (intent), and slot filling (slot).
[0069] In this embodiment, these keywords are also preliminarily classified and tagged with parts of speech.
[0070] Part-of-speech tagging, also known as grammatical tagging or word class resolution, is a text data processing technology in corpus linguistics that marks the part of speech of words in a corpus according to their meaning and context. Part-of-speech tagging can be done manually or by a specific algorithm. Using machine learning methods to implement part-of-speech tagging is the research content of NLP. Common part-of-speech tagging algorithms include hidden Markov models and conditional random fields. Part-of-speech tagging is mainly used in the fields of text mining and NLP (Natural Language Generation), and is a preprocessing step for various text-based machine learning tasks, such as semantic analysis and coreference resolution.
[0071] For example:
[0072] Keywords were extracted from the text in the knowledge base [Stomachache is a non-specific symptom. Stomachache, also known as epigastric pain, refers to the pain in the epigastric region above the lower edge of the ribs to the xiphoid process, near the heart. The pain has different properties such as distending pain, stabbing pain, dull pain, and severe pain. It is often accompanied by upper gastrointestinal symptoms such as loss of appetite, nausea and vomiting, noisy heartburn, belching, and vomiting. Characteristics of the disease: Most of them are middle-aged and young people, and most of them have a history of recurrent attacks. Before the onset of the disease, there are obvious inducements, such as weather changes, anger, fatigue, overeating, hunger, eating raw, cold, dry and hard food, spicy food, tobacco and alcohol, or taking drugs that damage the spleen and stomach. Common diseases include acute gastritis, chronic gastritis, gastric ulcer, duodenal ulcer, functional dyspepsia, gastric mucosal prolapse, and gastric cancer. The main examination methods are gastroscopy and biopsy, gastric pathological biopsy, gastric juice analysis, Helicobacter pylori detection, etc.]
[0073] Results after keyword extraction:
[0074] Field detection: medical;
[0075] Intent identification: medical examination;
[0076] Slot Entity:
[0077] {
[0078] Disease injury: stomachache, epigastric pain;
[0079] Symptoms: distending pain, stabbing pain, dull pain, severe pain;
[0080] Diagnosis and treatment methods: gastroscopy and biopsy, gastric pathological biopsy, gastric juice analysis, Helicobacter pylori detection;
[0081] }
[0082] Preferably, on the basis of preliminary classification and part-of-speech tagging, further editing is performed manually to obtain more accurate keyword extraction results.
[0083] In specific practice, the extracted entity information will also be subjected to structured extraction to facilitate subsequent processing into prompt text.
[0084] It should be noted that based on the AIGC model, Chinese scenario-based text corpus corresponding to the demand field is generated according to the extracted keywords, including:
[0085] The prompt text of the AIGC model is written according to the extracted keywords; the prompt text is a short text that interacts with the AIGC model; the prompt text is used to provide instructions to the AIGC model to guide the AIGC model to generate text answer content related to the keywords;
[0086] The prompt text is input into the AIGC model, so that the AIGC model generates Chinese scenario text corpus corresponding to the required field.
[0087] It should be noted that the AIGC model in a narrow sense refers to a production method that automatically generates content through artificial intelligence technology. In a broad sense, AIGC can automatically generate new content and data in various forms such as text, images, music, video, 3D interactive content, etc. based on training data and generation algorithm models.
[0088] In specific practice, after extracting keywords, NLG technology is used to write the prompt text of the AIGC model based on the extracted keywords, so as to generate a large amount of Chinese scenario text corpus corresponding to the demand field.
[0089] It should be noted that the method also includes:
[0090] Optimize the prompt structure and content of the prompt text.
[0091] In this embodiment, prompt engineers need to participate in optimizing the prompt structure and content with prompt texts, thereby limiting the input of the AIGC model to a specific range, thereby better controlling the output of the AIGC model.
[0092] It should be noted that the method also includes: after the Chinese scenario text corpus is generated, the prompt text is extracted and converted into synonyms and near-synonyms, and the replaced prompt text is re-input into the AIGC model to generate multiple groups of Chinese scenario text corpora.
[0093] In this embodiment, by extracting and converting synonyms and near-synonyms of a set of prompt texts and then repeatedly inputting them into the AIGC model, the AIGC model can generate multiple sets of Chinese scenario text corpora, thereby enriching and extending the text collection.
[0094] It should be noted that the prompt text is the short text that interacts with the AIGC model. The AIGC model will decompose the words and phrases mentioned in it into smaller tokens (the smallest unit of sentence splitting in the text). The prompt words in the prompt text can be compared with the huge text data during model training, and then relevant text feedback can be generated.
[0095] The role of prompt text is to guide the AIGC model to generate highly relevant, accurate and high-quality text dialogue content by providing clear and specific instructions. It is an important breakthrough in the field of natural language processing. In actual NLG tasks, better and more accurate prompts can help the AIGC model generate more coherent and accurate text.
[0096] For example:
[0097] The prompt text input into the AIGC model and the Chinese scenario text corpus output by the AIGC model are as follows: Figure 3 shown.
[0098] Prompt text: Please simulate and generate a dialogue text between a doctor and a patient in the [Medical] theme in the [Medical Examination] scenario. The patient's registered department is [Internal Medicine - Gastroenterology], the patient's disease injury is [stomach pain], the main symptoms are [bloating, nausea and vomiting], and the diagnosis and treatment methods are [gastroscopy, gastric juice analysis, Helicobacter pylori detection].
[0099] Synonymous replacement:
[0100] The "symptoms" words are replaced with synonyms and supplemented to produce other dialogue texts with the same symptoms.
[0101] Chinese scenario text corpus output by the AIGC model:
[0102] Doctor: Hello, are you here to see the gastroenterologist?
[0103] Patient: Yes, I have had severe stomach pain recently, with bloating, nausea and vomiting.
[0104] Doctor: OK, please fill in the medical history form first, and then we will do some necessary examinations. First, I suggest you do a gastroscopy so that we can have a clearer understanding of your stomach condition.
[0105] Patient: Will gastroscopy be painful?
[0106] Doctor: Don't worry, we will spray you with some anesthetic so that you won't feel any pain during the examination. At the same time, we will also analyze your gastric juice to understand your gastric acid secretion.
[0107] Patient: What if the test results show a problem?
[0108] Doctor: If the test results show that you have stomach inflammation or other problems, we will develop a treatment plan based on the specific situation. In addition, we will test you for Helicobacter pylori, because this bacterium can cause gastritis and stomach ulcers.
[0109] Patient: OK, so when can I have these tests?
[0110] Doctor: You can go for a gastroscopy first. This examination needs to be booked in advance. After the gastroscopy is completed, we will arrange a time for gastric juice analysis and Helicobacter pylori testing based on your examination results. Please rest assured that we will arrange treatment for you as soon as possible.
[0111] The following are examples of synonym extraction:
[0112] Taking stomachache as an example, synonyms of stomachache include: stomach discomfort, gastrointestinal discomfort, chronic gastroenteritis, chronic gastritis, abdominal distension, frequent stomach pain, constipation, indigestion, etc.
[0113] In specific practice, the AIGC model uses the interface of iFlytek Spark Engine.
[0114] Synonym data is calculated using Tencent’s word vector.
[0115] It should be noted that before translating the generated Chinese scenario text corpus into Tibetan scenario text corpus, the method further includes:
[0116] Review the logic, compliance, usability, accuracy and repeatability of the Chinese scenario text corpus generated by the AIGC model;
[0117] The approved Chinese scenario text corpus is segmented into sentences.
[0118] It is understandable that the produced text needs to be reviewed to ensure its logic and content compliance, its usability in actual field use, and to ensure that it does not repeat or contain errors.
[0119] The approved Chinese scenario text corpus is segmented and organized into single sentences to facilitate subsequent recording and data storage.
[0120] In specific practice, Chinese scenario-based text corpus is translated into Tibetan scenario-based text corpus through Chinese-Tibetan machine translation, the machine translation terminology database is automatically matched during the translation process, and translation memory is automatically generated for the proofreading during the proofreading process by Tibetan professionals to optimize the translation content of subsequent related content.
[0121] In this embodiment, the Chinese-Tibetan machine translation interface is called to translate text, and after correction for word translation errors, corpus can be added to optimize the machine translation engine.
[0122] It should be noted that the Tibetan scenario audio corpus was recorded based on the preset scenario simulation parameters;
[0123] The scenario simulation parameters at least include: recording sampling rate, environmental noise during recording, and amplitude during recording.
[0124] In practice, the translated Tibetan scenario audio corpus can be recorded according to scenario simulation parameters, which at least include: recording sampling rate, environmental noise during recording, and amplitude during recording, etc. The simulation can be performed according to the deployment environment required by the project.
[0125] It should be noted that the environmental noise during recording needs to be acquired in advance in the actual scene through a recording device and stored as an audio file. The audio file is played through the audio device during recording, and the environmental noise in the actual scene is used as the recorded background audio. The environmental noise in the actual scene can be the chanting sound, the sound of people flowing, the ringing of bells, etc. in the temple, or the sound of calling numbers and the groaning of patients in the hospital. In this technical solution, the recorded audio corpus is made more authentic by adding the environmental noise in the actual scene as the recorded background audio during the recording process.
[0126] It should be noted that environmental noise can also be generated through computer simulation, but the simulated environmental noise is slightly less realistic than the directly recorded environmental noise, and is generally suitable for actual scenarios where environmental noise cannot be recorded directly.
[0127] The audio data after recording is used by developers for model training and testing.
[0128] It should be noted that the Tibetan scenario-based audio corpus and the Tibetan scenario-based text corpus are bound to obtain a standard audio corpus dataset.
[0129] The bound Tibetan scenario-based audio corpus and Tibetan scenario-based text corpus are classified and stored according to the required fields and scenarios for model developers to optimize the acoustic model or language model.
[0130] It should be noted that the method also includes:
[0131] The keywords to be extracted
[0132] Based on the translation engine, the extracted keywords are translated into Tibetan form;
[0133] The extracted keywords and the translated Tibetan keywords are cleaned and entered into the knowledge base.
[0134] In this embodiment, the extracted keywords and the translated Tibetan keywords are cleaned and entered into the knowledge base to enrich the vocabulary.
[0135] It can be understood that the method for generating Tibetan scene-based audio corpus based on AIGC in this embodiment includes: establishing a corresponding knowledge base according to the demand field of Tibetan audio corpus; extracting keywords from the knowledge base; based on the AIGC model, generating Chinese scene-based text corpus corresponding to the demand field according to the extracted keywords; based on the translation engine, translating the generated Chinese scene-based text corpus into Tibetan scene-based text corpus; obtaining Tibetan scene-based audio corpus recorded according to Tibetan scene-based text corpus, binding Tibetan scene-based audio corpus and Tibetan scene-based text corpus; classifying and storing the bound Tibetan scene-based audio corpus and Tibetan scene-based text corpus according to the demand field and scene. The technical solution in this application can significantly improve the generation efficiency of Tibetan scene-based audio corpus by extracting keywords in the required Tibetan audio corpus field, automatically generating, outputting and translating the corresponding Tibetan scene-based text corpus according to the keywords. The technical solution in this application can significantly improve the generation efficiency of Tibetan scene-based audio corpus. The generated Tibetan scene-based audio corpus standards and specifications are conducive to the standardized management of Tibetan scene-based audio corpus.
[0136] The technical solution in this embodiment solves the problem of difficulty and low efficiency in collecting Tibetan scenario audio corpus, so that most AI systems that need to train Tibetan speech recognition technology modules can quickly, stably and risk-free obtain the complete and well-annotated speech data sets in a standardized format that can be adapted to the actual scenarios required for training the recognition engine.
[0137] Embodiment 2
[0138] Figure 4is a schematic diagram of a Tibetan language scenario audio corpus generation device based on AIGC provided by an embodiment of the present application, with reference to Figure 4 , a Tibetan scene audio corpus generation device based on AIGC, comprising:
[0139] Processor 21 and memory 22;
[0140] The processor 21 and the memory 22 are connected via a communication bus:
[0141] The processor 21 is used to call and execute the program stored in the memory 22;
[0142] The memory 22 is used to store a program, and the program is at least used for a method for generating Tibetan scene-based audio corpus based on AIGC as described in the above embodiment.
[0143] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.
[0144] It should be noted that, in the description of this application, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" refers to at least two.
[0145] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.
[0146] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0147] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.
[0148] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.
[0149] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.
[0150] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0151] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.
Claims
1. A method for generating Tibetan scene-based audio corpus based on AIGC, characterized in that: include: Establish a corresponding knowledge base based on the demand areas of Tibetan audio corpus; Extracting keywords from the knowledge base; Based on the AIGC model, generate Chinese scenario-based text corpus corresponding to the demand field according to the extracted keywords; Based on the translation engine, the generated Chinese scenario text corpus is translated into Tibetan scenario text corpus; Acquire Tibetan scenario-based audio corpus recorded according to the Tibetan scenario-based text corpus, and bind the Tibetan scenario-based audio corpus and the Tibetan scenario-based text corpus; Classify and store the bound Tibetan scenario-based audio corpus and Tibetan scenario-based text corpus according to the required fields and scenarios; Establish corresponding knowledge base according to the demand areas of Tibetan audio corpus, including: Divide the process links of the demand areas; Relevant textual knowledge is obtained for different process links, and a knowledge base corresponding to the required field is established; Summarize and mark the established knowledge base according to the dictionary content index; Based on the AIGC model, Chinese scenario-based text corpus corresponding to the demand field is generated according to the extracted keywords, including: Writing a prompt text of the AIGC model according to the extracted keywords; the prompt text is a short text that interacts with the AIGC model; the prompt text is used to provide instructions to the AIGC model to guide the AIGC model to generate text answer content related to the keywords; The prompt text is input into the AIGC model, so that the AIGC model generates Chinese scenario text corpus corresponding to the demand field.
2. The method according to claim 1, characterized in that: Extracting keywords from the knowledge base includes: Based on NLU technology, the dictionary text in the knowledge base is subjected to domain detection, intent recognition and slot filling to extract keywords; Classify and tag the extracted keywords.
3. The method according to claim 1, characterized in that The method further comprises: After the Chinese scenario text corpus is generated, synonyms and near-synonyms are extracted and converted for the prompt text, and the replaced prompt text is re-input into the AIGC model to generate multiple groups of Chinese scenario text corpora.
4. The method according to claim 1, characterized in that: The method further comprises: The prompt text is optimized in terms of prompt structure and content.
5. The method according to claim 1, characterized in that Before translating the generated Chinese scenario text corpus into Tibetan scenario text corpus, the method further includes: Review the logic, compliance, usability, accuracy and repeatability of the Chinese scenario text corpus generated by the AIGC model; The approved Chinese scenario text corpus is segmented into sentences.
6. The method according to claim 1, characterized in that The Tibetan scenario audio corpus is recorded based on preset scenario simulation parameters; The scenario simulation parameters include at least: recording sampling rate, environmental noise during recording, and amplitude during recording.
7. The method according to claim 1, characterized in that The method further comprises: The keywords to be extracted Based on the translation engine, the extracted keywords are translated into Tibetan form; The extracted keywords and the translated Tibetan keywords are cleaned and then entered into the knowledge base.
8. A Tibetan scenario audio corpus generation device based on AIGC, characterized in that: include: Processor and memory; The processor and the memory are connected via a communication bus: Wherein, the processor is used to call and execute the program stored in the memory; The memory is used to store a program, and the program is at least used to execute the method for generating Tibetan scene-based audio corpus based on AIGC as described in any one of claims 1-7.
