A method and system for automatically generating listening audio based on the original text of listening test questions

Abstract

The present invention discloses a method, system and device for automatically generating listening audio based on the original text of a listening test question. The method comprises: taking the original text of the listening test question as input and dividing its logical structure, specifically including chapter division and test question / prompt text division; based on the divided logical structure, organizing the original text of the listening test question into various chapters, and organizing the text of each chapter into two categories: "exam prompts" and "test question content"; for the "exam prompts" text, extracting the first category of text processing information on a paragraph basis and generating audio processing instructions; for the "test question content" text, extracting the second category of text processing information on a paragraph basis and generating audio processing instructions. The user only needs to divide the logical structure of the original text of the listening test question, without having to specify the reading method, pause time and other contents for each sentence as in the prior art. In addition to significantly reducing the workload, it effectively lowers the learning threshold for users.

Description

Technical Field

[0001] The present invention relates to a method and system for automatically generating listening audio based on original listening test texts, and belongs to the technical field of information processing. Background Art

[0002] Text-to-speech (TTS) technology is now mature and widely used in all aspects of life, work, and study. However, for various listening tests, such as English listening tests, audio recording often involves dialogue. Traditional manual recording methods would require at least one male and one female teacher to read the test questions aloud, placing high demands on production conditions and costs. Therefore, the production of English listening test audio is rapidly evolving from manual recording to the use of TTS technology, significantly reducing labor costs and ensuring consistent production quality.

[0003] When using "text-to-speech" technology to generate listening test audio, it is necessary to specify or set the AI announcer used for each text, the pause time to be inserted after each test question, the number of times some test questions need to be repeated, and other production information (such as Figure 1 Therefore, even though the application of “text-to-speech” technology reduces production costs overall, it still requires a lot of production work. Summary of the Invention

[0004] Purpose of the invention: In response to the problems existing in the prior art, the present invention provides a method and system for automatically generating listening audio based on the original text of listening test questions. It only needs to divide the original text of the listening test questions into logical structures, specifically including chapter division and test question / prompt text division, to automatically identify / extract various key information in the original text of the listening test questions, and parse and process these key information according to the logical structure to generate corresponding audio processing instructions, thereby automatically generating listening audio.

[0005] Technical Solution: To achieve the above objectives, the present invention provides a method for automatically generating listening audio based on the original text of a listening test question, comprising the following steps:

[0006] The listening test text is used as input, and the logical structure of the listening test text is divided. The logical structure division includes the division into chapters and the division of test questions / prompt texts (the division form is not limited);

[0007] Based on the logical structure, the listening test text is organized into chapters, and the text in each chapter is organized into two categories: "Test Tips" and "Test Content";

[0008] For the "Exam Tips" text, extract the first type of text processing information in paragraph units and generate audio or audio processing instructions. If the extraction is unsuccessful, generate audio or audio processing instructions according to the set first type of text processing information, wherein the first type of text processing information includes the number of repetitions and pause time, etc.;

[0009] For the "Question Content" text, extract the second type of text processing information on a paragraph basis and generate audio or audio processing instructions. If the extraction is unsuccessful, generate audio or audio processing instructions based on the set second type of text processing information. The second type of text processing information includes question number, dialogue character, and pause time.

[0010] The pause time is processed as a separate audio or audio processing instruction in the form of silence; the question number is processed as a separate audio or audio processing instruction and is used to distinguish between different test questions; the dialogue role is used to change the broadcast role of the current paragraph and does not generate a separate audio or audio processing instruction; the number of repetitions is used to repeat each test question in the chapter a specified number of times;

[0011] When all chapters are processed, all generated audio or audio processing instructions are linked in the order of chapters and topic numbers to generate complete audio or audio processing instructions and return them.

[0012] Furthermore, for the pause time in the "Exam Tips" text, if the extraction is unsuccessful, the position of the current paragraph is further determined: if the current paragraph is in the middle of the "Exam Tips" text, the set pause time H1 is inserted at the end of the paragraph; if the current paragraph is before the "Test Question Content" text, the set pause time H2 is inserted at the end of the paragraph; if the current paragraph is at the end of the chapter, the set pause time H3 is inserted at the end of the paragraph.

[0013] Furthermore, for the pause time in the "test content" text, if the extraction is unsuccessful, the position of the current paragraph is further determined: if the current paragraph is in the middle of the test question, the set pause time Q1 is inserted at the end of the paragraph; if the current paragraph is before the next test question, the set pause time Q2 is inserted at the end of the paragraph; if the current paragraph is at the end of the chapter, the set pause time Q3 is inserted at the end of the paragraph.

[0014] The present invention also provides a system for automatically generating listening audio based on the original text of the listening test questions, comprising:

[0015] A control module is used to input the original text of the listening test questions and divide the logical structure of the original text of the listening test questions, wherein the logical structure division includes the division into chapters and the division of test questions / prompt texts;

[0016] The chapter management module is used to organize the listening test text into chapters according to the divided logical structure, and organize the text in each chapter into two categories: "Test Tips" and "Test Content";

[0017] The chapter processing module includes a distribution submodule, an exam prompt processing submodule, and an exam question content processing submodule. The distribution submodule is used to distribute the "exam prompt" and "exam question content" texts to the exam prompt processing submodule and the exam question content processing submodule according to the text type.

[0018] For the "Exam Tips" text, the exam tips processing submodule extracts the first type of text processing information in paragraph units and generates audio processing instructions. If the extraction is unsuccessful, the audio processing instructions are generated according to the set first type of text processing information, which includes the number of repetitions and pause time.

[0019] For the "Question Content" text, the Question Content Processing submodule extracts the second type of text processing information in paragraph units and generates audio processing instructions. If the extraction is unsuccessful, the audio processing instructions are generated according to the set second type of text processing information. The second type of text processing information includes question number, dialogue character, and pause time.

[0020] The pause time is processed as a separate audio processing instruction in the form of silence; the question number is processed as a separate audio processing instruction and is used to distinguish between different test questions; the dialogue role is used to change the broadcast role of the current paragraph and does not generate a separate audio processing instruction; the number of repetitions is used to repeat each test question in the chapter a specified number of times;

[0021] The audio generation module is used to record the audio processing instructions generated by the chapter processing module and call the text-to-speech system for processing to generate complete listening audio.

[0022] Furthermore, the listening audio generation system also includes a template management module, which includes a template library and a template matching engine. The test prompt processing submodule and the test question content processing submodule respectively call the template matching engine to match the template in the template library, thereby obtaining corresponding text processing information.

[0023] Furthermore, the template matching engine is implemented based on the matching and capture grouping functions of regular expressions.

[0024] Furthermore, the chapter processing module includes a context state memory submodule for storing the extracted and set first-category text processing information and second-category text processing information, wherein the set first-category text processing information includes pause time H1, pause time H2, and pause time H3, etc., and the set second-category text processing information includes pause time Q1, pause time Q2, and pause time Q3, etc.;

[0025] For the pause time in the "Exam Tips" text, if the extraction is unsuccessful, the exam prompt processing sub-module further determines the position of the current paragraph: if the current paragraph is in the middle of the "Exam Tips" text, the set pause time H1 is inserted at the end of the paragraph; if the current paragraph is before the "Exam Question Content" text, the set pause time H2 is inserted at the end of the paragraph; if the current paragraph is at the end of the chapter, the set pause time H3 is inserted at the end of the paragraph.

[0026] For the pause time in the "Question Content" text, if the extraction is unsuccessful, the question content processing submodule further determines the position of the current paragraph: if the current paragraph is in the middle of the question, the set pause time Q1 is inserted at the end of the paragraph; if the current paragraph is before the next question, the set pause time Q2 is inserted at the end of the paragraph; if the current paragraph is at the end of the chapter, the set pause time Q3 is inserted at the end of the paragraph.

[0027] Furthermore, the control module may also adjust the set first-category text processing information and second-category text processing information according to the user's input, and update the information to the context state memory submodule.

[0028] In addition, there is also a device for automatically generating listening audio based on the original text of the listening test questions, which includes a processor for running the above-mentioned automatic generation system of listening audio.

[0029] Beneficial Effects: The method and system for automatically generating listening audio based on the original text of listening test questions provided by the present invention have the following advantages over the prior art:

[0030] 1. Users only need to divide the logical structure of the original listening test text, without having to specify the reading method, pauses, etc. for each sentence as required by existing technologies. In addition to significantly reducing the workload, the logical structure of the listening test questions is also well known to the listening comprehension producer (usually the corresponding subject teacher), which lowers the learning threshold for users. Users do not need to understand the characteristics and usage of the listening audio generation system, but only need to focus on the listening test content itself.

[0031] 2. Regular expressions are used to abstract the concept of "template" and provide online updating of "template". This means that the entire system does not need to be frequently modified due to the diversity of listening test texts. Only the corresponding templates need to be provided. In addition, since the chapter content is strictly divided into two categories: "exam tips" and "test content", each paragraph of text only needs to match the corresponding template. This is also a good embodiment of determinism. Therefore, the solution of the present invention is highly applicable to various listening test questions, the results are reliable, and the process is controllable. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 An operating interface for generating listening test audio using "text-to-speech" technology in the prior art;

[0033] Figure 2 The original text of the English listening test after logical division in the embodiment of the present invention;

[0034] Figure 3 A diagram showing the text structure divided by the chapter management module in an embodiment of the present invention;

[0035] Figure 4 This is a structural diagram of a system for automatically generating hearing audio according to an embodiment of the present invention;

[0036] Figure 5 These are some commonly used templates implemented based on regular expressions in the embodiments of the present invention;

[0037] Figure 6 is an English listening audio processing instruction that can be generated in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The preferred embodiments of the present invention will be described below in conjunction with the accompanying drawings to more clearly and completely illustrate the technical solutions of the present invention.

[0039] Since the original text of the listening test questions is in natural language, it is difficult to accurately identify and process them effectively, and the forms are very diverse. The test types alone include various listening tests such as Chinese, English, and Japanese. Moreover, even for the most common English listening test, the "test tips" for each chapter in the lower grade test content are in Chinese, while the "test tips" for the higher grade English listening test and the CET-4 and CET-6 listening test are all in English. It can be seen that the specific format of the original text of each listening test question is not exactly the same.

[0040] Therefore, the listening test text is divided into logical structures, specifically chapters and question / prompt text. This aims to avoid directly processing the ever-changing content and format of the listening test text, instead focusing on the unchanging information within the text—the logical structure—and then performing targeted processing based on this information, which greatly improves certainty. Furthermore, the less information that needs to be annotated, the simpler the use and the higher the degree of automation. Therefore, the logical structure division must be concise yet effective. First, due to different ways of extracting and processing information, question / prompt text division is necessary. Second, it has been observed that each chapter / question contains one or more groups of questions, and the information presented in the prompts for a chapter / question often affects the entire chapter. For example, if a chapter prompt says "Read each question three times," if that chapter contains multiple groups of questions, then each group of questions should be read three times. In other words, the chapter represents the scope of application.

[0041] Therefore, practical experience has shown that dividing the listening test text into two logical sections, chapter by chapter and question / prompt text by chapter, is the most optimal approach. The listening test text is then organized into multiple chapters, each containing one or more sets of prompts and questions. This creates a relatively fixed format for the text, making automated processing possible.

[0042] Considering the aforementioned issues of diverse listening test content and non-fixed formats, the present invention further addresses this issue through the use of a template library. The template library is divided into "common templates," "question templates," and "prompt templates" according to the type of test text. Different types of text only require matching with the corresponding template, rather than randomly matching templates. This further enhances certainty and specificity. Therefore, through a meticulously designed, step-by-step approach, the present invention addresses numerous technical difficulties in automated processing, thus resolving the significant drawback of the prior art, which requires manual configuration of all instructions.

[0043] In the present invention, the logical structure of the listening test text is not limited to any specific form. For example, it can be expressed through content formatting (e.g., each chapter begins with a "title" format, using boldface and larger font, while the test questions use italics), or through blank lines (e.g., two blank lines between each chapter, with one blank line before and after each test question), or indentation (e.g., each chapter is separated by a blank line, with the test questions indented to the right), or through user annotation, such as inserting symbols such as <Chapter Start><Chapter End> before and after each chapter / question, and <Question Start><Question End> before and after each test question, or through recognition and division by an artificial intelligence program. Regardless of the specific situation, the present invention does not impose any restrictions. The present invention only requires dividing the logical structure of the listening test text and then generating the listening test audio or audio instructions based on the divided logical structure.

[0044] like Figure 2 As shown, for the sake of convenience, it is assumed that the user logically divides the original text of the listening test by inserting annotations, that is, inserting symbols such as <Chapter Start> and <Chapter End> before and after each chapter / question, and inserting symbols such as <Question Start> and <Question End> before and after the question.

[0045] The embodiment of the present invention can automatically parse, identify and process the original text of the listening test questions. The specific process is as follows:

[0046] 1. Based on the divided logical structure, organize the listening test text into chapters, automatically mark the unmarked parts as chapters (mainly the title and end of the question), and divide the text in the middle chapters into two parts: "Test Tips" and "Test Content". Since "Test Content" has been marked, the unmarked parts in the chapters are regarded as "Test Tips", thus organizing the listening test text into Figure 3 The header chapter and the footer chapter are relatively special chapters because they do not contain the "question content" section. They are simplified "chapters", but this does not affect subsequent processing.

[0047] 2. For the "Exam Tips" text, extract the first-category text processing information in paragraph units and generate audio processing instructions. If the extraction is unsuccessful, generate audio processing instructions according to the set first-category text processing information. The first-category text processing information extracted and set here mainly includes the broadcasting role, number of repetitions, and pause time.

[0048] For the announcer role, in the first type of text processing information set, it is only necessary to make a unified configuration according to the language to generate audio processing instructions. For example: if it is Chinese, the paragraph is set to be read by Chinese female voice 1; if it is an English sentence, the paragraph is set to be read by English female voice 1, without having to specify the reading method for each sentence as in the existing technology.

[0049] Regarding pause times, if there is a pause at the end of a paragraph, it is extracted and a corresponding pause is inserted at the end of the paragraph (i.e., a separate audio processing instruction is generated), rather than using the set pause time. If extraction is unsuccessful, the current paragraph's location is further determined: if the current paragraph is in the middle of the "Exam Tips" text, the set pause time H1 is inserted at the end of the paragraph; if the current paragraph is before the "Exam Content" text, the set pause time H2 is inserted at the end of the paragraph; if the current paragraph is at the end of a chapter, the set pause time H3 is inserted at the end of the paragraph.

[0050] 3. For the "test content" text, extract the second-category text processing information in paragraph units and generate audio processing instructions. If the extraction is unsuccessful, generate audio processing instructions according to the set second-category text processing information. The second-category text processing information extracted and set here mainly includes the question number, dialogue role, and pause time.

[0051] The question number can be configured to be read aloud by a designated announcer (e.g., English female voice 3), or a pause Q4 can be inserted after the question number, generating two audio processing instructions. Furthermore, the question number can be used to further distinguish between test questions, making it easier to determine the current paragraph position.

[0052] For dialogue characters, if a dialogue character identifier is present in a paragraph, the subsequent text is assigned to the corresponding announcer until a new dialogue character identifier appears or the paragraph ends. For example, if the "M:" marker is identified, the announcer corresponding to that marker, "English Male Voice 4," is used. If the extraction fails (indicating that the text is not a dialogue), the designated announcer is assigned to the text.

[0053] For pause time, if there is a pause at the end of a paragraph, it is extracted and inserted at the end of the paragraph instead of the preset pause time. If the extraction is unsuccessful, the current paragraph's position is further determined: if the current paragraph is in the middle of the test question, the preset pause time Q1 is inserted at the end of the paragraph; if the current paragraph is before the next test question, the preset pause time Q2 is inserted at the end of the paragraph; if the current paragraph is at the end of a chapter, the preset pause time Q3 is inserted at the end of the paragraph.

[0054] For the number of repetitions, the instructions contained in each question under the chapter are set to repeat n times, where n is the number of repetitions extracted or set (extraction takes priority).

[0055] 4. After the above steps, each paragraph has been set up with which announcer to read it, the pause time required for each paragraph, etc., so it can be processed according to the generated audio processing instructions (such as Figure 6 As shown in the figure, the “text-to-speech” technology is directly called to generate the final English listening audio.

[0056] like Figure 4 The figure shows the automatic generation system of listening audio in the implementation scheme of the present invention, which is composed of a control module, a chapter management module, a chapter processing module, a template management module, a cloud template library, an audio generation module, and a text-to-speech service module. The text-to-speech service module is an external module and is consistent with the existing technology.

[0057] The present invention actually has multiple implementations / deployment methods. This implementation assumes that most modules are deployed on the device / terminal / client side. In this case, the template library needs to provide online update means. If the entire system is deployed on the cloud, the device side can simply provide a web page display. In this case, the template management module is merged with the cloud template library, and the template management module no longer needs an online update submodule. The following uses a relatively complex device-side deployment method to introduce:

[0058] 1) After the system is turned on, the control module first notifies the template management module to update the template. The template management module then calls the online update submodule to query the cloud template library for new templates and save them locally. Simultaneously, the user can adjust the first and second text processing information settings through the control module and update them in the context state memory submodule.

[0059] 2) The user inputs the original text of the listening test questions into the control module, and divides the logical structure of the original text of the listening test questions. After the division is completed, the control module first notifies the chapter management module to read the above listening test question text by chapter, and regards the content before the first <Chapter Start> as an independent chapter (header), and the content after the last <Chapter End> as an independent chapter (tail). In this way, all text data is organized in the form of chapters. The particularity of the two chapters "header" and "tail" is reflected in the fact that these two chapters do not contain the "test question content" part, but this does not affect subsequent processing. Secondly, since the chapter contains two parts of content, one is the "test question content" text that has been marked, and the remaining parts are divided into "exam tips" text, thereby organizing the original text of the listening test questions into Figure 3The entire listening test text is organized into multiple chapters, and the text content in each chapter is strictly divided into two categories: "Exam Tips" and "Exam Content." This organizes the difficult-to-process natural language into relatively unified structured data.

[0060] 3) The chapter management module assigns each chapter to the chapter processing module. The chapter processing module consists of a distribution submodule, a test prompt processing submodule, a test question content processing submodule, and a context state memory submodule. The distribution submodule distributes the "test prompt" and "test question content" text to the test prompt processing submodule and the test question processing submodule for processing based on the text type.

[0061] For the "Exam Tips" text, the Exam Tips processing submodule needs to identify whether it contains the first type of text processing information; for the "Exam Question Content" text, the Exam Question Content processing submodule needs to identify whether it contains the second type of text processing information. This information extraction process actually involves text matching, so using regular expressions is more appropriate. Regular expressions describe a string matching pattern that can be used to check whether a string contains a certain substring, replace matching substrings, or extract substrings from a string that meet certain conditions.

[0062] The present invention refers to each regular expression as a "matching template", or "template" for short. Figure 5 As shown, the present invention divides templates into three categories: "public templates," "exam reminder templates," and "question content templates." If the content to be matched may exist in both the "exam reminder" and "question content" texts, this type of template is classified as a "public template." If the content to be matched only exists in the "exam reminder" text, this type of template is classified as an "exam reminder template." If the content to be matched only exists in the "question content" text, this type of template is classified as a "question content template."

[0063] Each template performs its own function, but because actual listening test texts may contain a variety of formats, multiple templates may need to be matched sequentially. For example, when parsing the dialogue character markers in the "Test Content" text, some listening passages may use M and W to represent male and female, while others use direct names, such as Mike and Mary. This means that when using templates for matching, the templates need to account for multiple scenarios. The solution here is to add multiple templates to a single template (with the same name) to cover these various possibilities. During the actual matching process, each template is matched sequentially until a match is found. No further matching is required, or if no matches occur, indicating that the content does not exist in the text.

[0064] The present invention uses regular expressions / templates to make the solution implementation more concise and clear, avoiding a large amount of logical judgment. At the same time, it can easily deal with various forms and contents that may appear in the listening test text. If new forms and contents appear, only the corresponding templates need to be added without modifying the entire terminal / system.

[0065] Therefore, the present invention adds a template management module to store all templates and provide template matching functionality. If the template management module is deployed on the device, a cloud-based template library can be provided. Then, an online update submodule can be added to the template management module. This allows the device to handle new forms of listening test text without requiring any fundamental changes; it only needs to keep up to date with the cloud-based template library.

[0066] 4) For the "Exam Tips" text, the exam tips processing submodule reads the text segment by segment and calls the template matching engine to match all public templates and exam tips templates. If a corresponding template is matched, the extracted information is saved to the context state memory submodule. Finally, instructions are generated based on whether the content is Chinese or English (each instruction includes a corresponding serial number, such as Figure 6 As shown), and sent to the audio generation module. The instruction generation method refers to step 2 in the above-mentioned hearing audio generation method.

[0067] 5) For the “Question Content” text, the Question Content Processing submodule reads the text paragraph by paragraph and calls the template matching engine to match all the common templates and question content templates. If the corresponding template is matched, the extracted information is saved to the context state memory submodule. Finally, the corresponding instructions are generated according to step 3 of the above-mentioned listening audio generation method (such as Figure 6 and sent to the audio generation module.

[0068] After receiving the instruction, the audio generation module first records the command and then executes it according to the command. That is: if it is a text-to-speech command, it is handed over to the text-to-speech service module for processing, and the processed audio file is attached to the command record; if it is a pause command, silent audio of the corresponding number of seconds is directly generated and attached to the command record. In this way, each command generates its corresponding audio.

[0069] 6) After all chapters are processed, the audio processing module connects the audio recorded by each command in sequence. If there is a "repeat" command, the audio within the repeat range is repeated a specified number of times. Finally, all are connected to generate a complete audio and returned to the control module.

[0070] In addition, there is also a device for automatically generating listening audio based on the original text of the listening test questions, which includes a memory and a processor, which are respectively used to store and run the above-mentioned automatic generation system of listening audio.

[0071] First, the present invention uses the original listening test text as input and divides the text into logical sections, specifically into chapters and test questions / prompt text. This aims to organize difficult-to-process natural language into structured data with a relatively uniform format, thereby facilitating subsequent automated processing. Furthermore, each section of text within a chapter is strictly divided into "test prompt" text and "test question content" text, significantly enhancing certainty and ensuring a clearer and more organized approach to subsequent processing.

[0072] Secondly, the present invention uses regular expressions to abstract the concept of "template" and provides online updating of "templates." This eliminates the need for frequent system modifications due to the diverse text of listening comprehension test questions. Instead, only the corresponding templates need to be provided and updated, fully ensuring the system's universality and accuracy. Furthermore, because each section of text in a chapter is strictly divided into "Exam Tips" and "Question Content," each section of text only needs to match the corresponding template, which also demonstrates determinism.

[0073] Therefore, the present invention solves many technical difficulties in automated processing through careful design, thereby resolving the obvious defect in the prior art that all instructions need to be manually set.

[0074] The above specific embodiments merely describe preferred embodiments of the present invention and do not limit the scope of protection of the present invention. Without departing from the design concept and spirit of the present invention, various modifications, substitutions, and improvements made by those skilled in the art to the technical solution of the present invention based on the text description and drawings provided herein shall fall within the scope of protection of the present invention.

Claims

1. A method for automatically generating listening audio based on the original text of a listening test question, characterized in that: include: Taking the original listening test text as input, and dividing the original listening test text into logical structures, wherein the logical structure division includes the division into chapters and the division into test questions / prompt texts; Based on the logical structure, the listening test text is organized into chapters, and the text in each chapter is organized into two categories: "Test Tips" and "Test Content"; For the "Exam Tips" text, extract the first type of text processing information in paragraph units and generate audio or audio processing instructions. If the extraction is unsuccessful, generate audio or audio processing instructions according to the set first type of text processing information, where the first type of text processing information includes the number of repetitions and pause time. For the "Question Content" text, extract the second type of text processing information on a paragraph-by-paragraph basis and generate audio or audio processing instructions. If extraction is unsuccessful, generate audio or audio processing instructions based on the set second type of text processing information. The second type of text processing information includes question number, dialogue character, and pause duration. The pause time is processed as a separate audio or audio processing instruction in the form of silence; the question number is processed as a separate audio or audio processing instruction and is used to distinguish between different test questions; the dialogue role is used to change the broadcast role of the current paragraph and does not generate a separate audio or audio processing instruction; the number of repetitions is used to repeat each test question in the chapter a specified number of times; When all chapters are processed, all generated audio or audio processing instructions are linked together in text order to generate a complete audio or audio processing instruction and return it.

2. The method for automatically generating hearing audio according to claim 1, wherein: For the pause time in the "Exam Tips" text, if the extraction is unsuccessful, the position of the current paragraph is further determined: if the current paragraph is in the middle of the "Exam Tips" text, the set pause time H1 is inserted at the end of the paragraph; if the current paragraph is before the "Test Content" text, the set pause time H2 is inserted at the end of the paragraph; if the current paragraph is at the end of the chapter, the set pause time H3 is inserted at the end of the paragraph.

3. The method for automatically generating hearing audio according to claim 1, wherein: For the pause time in the "Question Content" text, if the extraction is unsuccessful, the position of the current paragraph is further determined: if the current paragraph is in the middle of the question, the set pause time Q1 is inserted at the end of the paragraph; if the current paragraph is before the next question, the set pause time Q2 is inserted at the end of the paragraph; if the current paragraph is at the end of the chapter, the set pause time Q3 is inserted at the end of the paragraph.

4. A listening audio automatic generation system based on the original text of the listening test questions, characterized by: include: A control module is used to input the original text of the listening test questions and divide the logical structure of the original text of the listening test questions, wherein the logical structure division includes the division into chapters and the division of test questions / prompt texts; The chapter management module is used to organize the listening test text into chapters according to the divided logical structure, and organize the text in each chapter into two categories: "Test Tips" and "Test Content"; The chapter processing module includes a distribution submodule, an exam prompt processing submodule, and an exam question content processing submodule. The distribution submodule is used to distribute the "exam prompt" and "exam question content" texts to the exam prompt processing submodule and the exam question content processing submodule according to the text type. For the "Exam Tips" text, the Exam Tips Processing submodule extracts the first type of text processing information in paragraph units and generates audio processing instructions. If the extraction is unsuccessful, the audio processing instructions are generated according to the set first type of text processing information, which includes the number of repetitions and pause time. For the "Question Content" text, the Question Content Processing submodule extracts the second type of text processing information on a paragraph basis and generates audio processing instructions. If the extraction is unsuccessful, the audio processing instructions are generated based on the set second type of text processing information. The second type of text processing information includes the question number, dialogue character, and pause time. The pause time is processed as a separate audio processing instruction in the form of silence; the question number is processed as a separate audio processing instruction and is used to distinguish between different test questions; the dialogue role is used to change the broadcast role of the current paragraph and does not generate a separate audio processing instruction; The number of repetitions is used to repeat each question in the chapter for a specified number of times; The audio generation module is used to record the audio processing instructions generated by the chapter processing module and process them using text-to-speech technology to generate complete listening audio.

5. The hearing audio automatic generation system according to claim 4, characterized in that: It also includes a template management module, which includes a template library and a template matching engine. The test prompt processing submodule and the test question content processing submodule respectively call the template matching engine to match the template in the template library, thereby obtaining corresponding text processing information.

6. The hearing audio automatic generation system according to claim 5, characterized in that: The template matching engine is implemented based on the matching of regular expressions and the capture grouping function.

7. The hearing audio automatic generation system according to claim 4, characterized in that: The chapter processing module includes a context state memory submodule for storing the extracted and set first-category text processing information and second-category text processing information, wherein the set first-category text processing information includes pause time H1, pause time H2, and pause time H3, and the set second-category text processing information includes pause time Q1, pause time Q2, and pause time Q3; For the pause time in the "Exam Tips" text, if the extraction is unsuccessful, the exam tip processing submodule further determines the location of the current paragraph: if the current paragraph is in the middle of the "Exam Tips" text, the set pause time H1 is inserted at the end of the paragraph; if the current paragraph is before the "Exam Question Content" text, the set pause time H2 is inserted at the end of the paragraph; if the current paragraph is at the end of the chapter, the set pause time H3 is inserted at the end of the paragraph; For the pause time in the "test question content" text, if the extraction is unsuccessful, the test question content processing submodule further determines the position of the current paragraph: if the current paragraph is in the middle of the test question, the set pause time Q1 is inserted at the end of the paragraph; if the current paragraph is before the next test question, the set pause time Q2 is inserted at the end of the paragraph; if the current paragraph is at the end of the chapter, the set pause time Q3 is inserted at the end of the paragraph.

8. The hearing audio automatic generation system according to claim 7, characterized in that: The control module adjusts the set first-category text processing information and second-category text processing information according to the user's input, and updates the information to the context state memory submodule.

Images