Audio document generation method, system, device, medium and program product

Abstract

The application relates to a document generation method, system, device, medium and program product of audio, comprising: acquiring a text sequence corresponding to target audio conversion; acquiring a timestamp of each character in the text sequence in the target audio; acquiring a difference between the timestamp of each character and the timestamp of the next character, obtaining a silence interval corresponding to each character; generating a possible punctuation set corresponding to each character according to the relative quantile of each silence interval in all silence intervals; inputting the text sequence and the possible punctuation set corresponding to each character into a punctuation prediction model for punctuation prediction to obtain a target document including punctuation. Through effective identification of the silence interval in the audio and association of the silence interval with the character, the application can more intelligently predict and insert punctuation symbols, thereby generating a more coherent and readable target document, improving the readability of the text, reducing the workload of manual correction, and enabling users to more efficiently obtain and use information.

Description

Technical Field

[0001] This application relates to the field of audio processing, and more specifically, to an audio document generation method, system, device, medium, and program product. Background Technology

[0002] With the rapid development of information technology and the acceleration of digitalization, voice information, as an efficient and natural communication medium, has experienced explosive growth in its output. This has led to a demand for the efficient archiving, retrieval, analysis, and application of massive amounts of voice information. Audio-to-document technology has emerged to address this need. Its core objective is to automatically and accurately convert the semantic content contained in voice signals into structured text documents that can be read, edited, stored, and analyzed. Audio-to-document technology has significant value in many fields, including daily meeting minutes, classroom lecture notes, legal and medical records, media content production, voice assistant interaction log processing, and accessible information services. Automatically converting audio into documents not only greatly liberates manpower and improves work efficiency, overcoming the problems of high cost, low efficiency, and high error rates associated with manual dictation, but also enables rapid digital accumulation of information content. This facilitates subsequent advanced intelligent processing using natural language processing technologies such as knowledge mining, content search, and summary generation, making it a key link in artificial intelligence-enabled information processing.

[0003] Documents generated by audio-to-text technology often lack punctuation. While they can achieve basic speech-to-text conversion, they lack contextual understanding when dealing with complex sentences and long paragraphs, resulting in incoherent and unreadable text. This deficiency forces users to spend extra time editing and proofreading, reducing the document's usability and readability. Summary of the Invention

[0004] In response, this application provides a method, system, device, medium, and program product for generating audio documents, to at least partially solve the above-mentioned technical problems.

[0005] This application provides a method for generating audio documents, including the following steps: Obtain the text sequence corresponding to the target audio; Obtain the timestamp of each character in the text sequence in the target audio; By obtaining the difference between the timestamps of each character and the next character, the silence interval corresponding to each character can be obtained. Based on the relative position of each silence interval among all silence intervals, generate a set of possible punctuation marks for each character; The text sequence and the set of possible punctuation marks corresponding to each character are input into a punctuation prediction model trained on context to predict punctuation marks, thereby obtaining a target document that includes punctuation marks.

[0006] Furthermore, the step of obtaining the timestamp of each character in the text sequence in the target audio includes: The target audio and the text sequence are time-aligned, and the timestamp of each character in the text sequence in the target audio is extracted.

[0007] Furthermore, the step of generating the possible punctuation set for each character based on the relative quantiles of each silence interval across all silence intervals includes: Calculate the dynamic silence interval threshold based on all silence intervals; Mark all characters whose silence interval exceeds the dynamic silence interval threshold as characters to be added with punctuation. For each character to be punctuated, a possible set of punctuation marks is generated based on the relative position of its corresponding silence interval among all silence intervals.

[0008] Furthermore, the steps for calculating the dynamic silence interval threshold based on all silence intervals include: Calculate the interquartile range for all silence intervals; Based on the interquartile range, a set of normal silence intervals is obtained by filtering from all silence intervals; Calculate the mean and standard deviation of all silence intervals in the set of normal silence intervals; The dynamic silence interval threshold is calculated using the following formula. : ; in, This represents the average value. This represents the standard deviation. This represents the first adjustable coefficient.

[0009] Furthermore, the step of selecting the set of normal silence intervals from all silence intervals based on the interquartile range includes: Calculate the upper limit of the silence interval using the following formula. : ; in, This represents the third quartile of all silence intervals. This refers to the interquartile range. This represents the second adjustable coefficient; The set of normal silence intervals is obtained by filtering out silence intervals that are smaller than the upper limit of the silence interval from all silence intervals.

[0010] Furthermore, for each character to be punctuated, the step of generating a possible set of punctuation marks corresponding to the character to be punctuated, based on the relative positions of its corresponding silence interval among all silence intervals, includes: The relative quantile of each punctuation character to be added is calculated using the following formula. : ; in, This indicates the silence interval corresponding to the punctuation character to be added. This represents the first quartile of all silence intervals. This represents the third quartile of all silence intervals; Based on the punctuation marks corresponding to the preset relative punctuation range, a set of possible punctuation marks corresponding to the character to be punctuated is generated.

[0011] In another aspect, this application also provides an audio document generation system, comprising: The first acquisition module is used to acquire the text sequence corresponding to the target audio; The second acquisition module is used to acquire the timestamp of each character in the text sequence in the target audio. The third acquisition module is used to obtain the difference between the timestamps of each character and the next character, and to obtain the silence interval corresponding to each character. The generation module is used to generate a set of possible punctuation marks for each character based on the relative position of each silence interval across all silence intervals. The prediction module is used to input the text sequence and the set of possible punctuation marks corresponding to each character into a punctuation prediction model trained on the context to predict the punctuation marks and obtain the target document including the punctuation marks.

[0012] This application also provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the audio document generation method as described above.

[0013] In another aspect, this application provides a computer-readable storage medium having stored thereon computer program instructions that can be executed by a processor to implement the audio document generation method described above.

[0014] Another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the audio document generation method described above.

[0015] This application enables more intelligent prediction and insertion of punctuation marks by effectively identifying silence intervals in audio and associating them with characters, thereby generating more coherent and readable target documents. This method not only improves text readability but also reduces the workload of manual proofreading, allowing users to access and use information more efficiently. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of an audio document generation method provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the structure of an audio document generation system provided in an embodiment of this application.

[0019] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0021] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, or product comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, apparatus, or products.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] Please see Figure 1 As shown, Figure 1 This is a flowchart illustrating an audio document generation method disclosed in an embodiment of this application. Figure 1 As shown, the method for generating the audio document may include the following operations: S101. Obtain the text sequence corresponding to the target audio; In this optional embodiment, Automatic Speech Recognition (ASR) technology can be used to convert the target audio into document. The basic principle of ASR is to analyze the audio signal and generate corresponding text through the combination of an acoustic model, a language model, and a decoder. The acoustic model is responsible for mapping audio features to phonemes or words, the language model provides contextual information to improve recognition accuracy, and the decoder combines the outputs of the acoustic and language models to ultimately generate text. In this optional embodiment, various open-source ASR tools can be used, and this application embodiment is not limited to any particular tool.

[0024] S102. Obtain the timestamp of each character in the text sequence in the target audio; In an optional embodiment, the step of obtaining the timestamp of each character in the text sequence in the target audio includes: The target audio and the text sequence are time-aligned, and the timestamp of each character in the text sequence in the target audio is extracted.

[0025] In this optional embodiment, the Montreal Forced Aligner can be used to perform timing alignment on the target audio and the text sequence. Montreal Forced Aligner (MFA) is an open-source speech-text forced alignment tool. Its core function is to accurately match the audio signal with the corresponding text in the time dimension through an acoustic model and a pronunciation dictionary, so as to output the start and end timestamps of each character in the audio. The Montreal Forced Aligner first preprocesses the audio and text, and splits the text into short sentence units; then extracts speech features; converts the text into a phoneme sequence according to the dictionary; then calculates the feature likelihood probability through the acoustic model, and models the phoneme timing using the Hidden Markov Model (HMM) state; finally performs forced alignment, iteratively optimizes the HMM state path, and determines the accurate time boundary of each character. Finally, the start and end time points of each character, which are two timestamps, are output.

[0026] In this optional embodiment, a character refers to the smallest text unit that is recognized and recorded during the process of converting audio to document, including but not limited to letters, numbers, and Chinese characters. For example, if a Chinese character syllable "我" appears in an audio and is converted into the character "我" in the document, the timestamp of this character in the audio includes the start timestamp and end timestamp of the complete pronunciation of the character "我" in the audio.

[0027] It can be seen that through timing alignment processing, this optional embodiment accurately matches the target audio with the text sequence, so that the timestamp of each character can accurately reflect its position in the audio. This processing method ensures that the basic data for subsequent punctuation prediction has high accuracy, thereby improving the overall quality and readability of the finally generated document.

[0028] S103: obtaining the difference between the timestamps corresponding to each character and the next character after it, to obtain the silence interval corresponding to each character; In this optional embodiment, the next character after one character refers to the next character of the character in the text sequence. For example, if there is text "大语言模型" in the text sequence, the next character after the character "大" is "语".

[0029] The difference between the timestamps corresponding to one character and the next character after it refers to the time difference between the end timestamp of the character and the start timestamp of the next character after it. When a character is the last character in the text sequence, the corresponding silence interval is preset as an interval constant. In this embodiment, the interval constant is 5 seconds.

[0030] S104: generating a possible punctuation set corresponding to each character according to the relative quantile of each silence interval among all silence intervals; In an optional embodiment, the step of generating a possible set of punctuation marks for each character based on the relative quantiles of each silence interval across all silence intervals includes: Calculate the dynamic silence interval threshold based on all silence intervals; Mark all characters whose silence interval exceeds the dynamic silence interval threshold as characters to be added with punctuation. For each character to be punctuated, a possible set of punctuation marks is generated based on the relative position of its corresponding silence interval among all silence intervals.

[0031] In this optional embodiment, relative quantile refers to the distribution position of a specific value relative to the set of data. Marking a character as a punctuation mark can be done by adding a specific string after the character, or by recording the position of the character to be marked in the text sequence; this embodiment does not impose any limitations.

[0032] As can be seen, this optional embodiment uses a dynamic threshold to filter characters to be added for punctuation, avoiding the problem of insufficient adaptability of a fixed threshold to different speech speed scenarios. The design of generating punctuation sets based on the relative quantile of silence intervals enables the punctuation candidate set to adapt to the distribution of speech pause features, providing structured constraints for subsequent language model prediction and effectively improving the targeting of punctuation prediction.

[0033] In an optional embodiment, the step of calculating the dynamic silence interval threshold based on all silence intervals includes: Calculate the interquartile range for all silence intervals; Based on the interquartile range, a set of normal silence intervals is obtained by filtering from all silence intervals; Calculate the mean and standard deviation of all silence intervals in the set of normal silence intervals; The dynamic silence interval threshold is calculated using the following formula. : ; in, This represents the average value. This represents the standard deviation. This represents the first adjustable coefficient.

[0034] In this optional embodiment, the interquartile range (IQM) is a statistical metric used to measure the dispersion of data. It is the difference between the third quartile and the first quartile of a dataset, representing the range of the middle 50% of the data. The IQM effectively reflects the centrality and dispersion of data, and is particularly representative when outliers are present. The first quartile is the value located at the 25th percentile of the data after sorting the dataset in ascending order; that is, 25% of the data points are below this value. The third quartile is the value located at the 75th percentile of the data after sorting the dataset in ascending order; that is, 75% of the data points are below this value. For example, in a dataset [3, 7, 8, 12, 13, 14, 18], the first quartile is 7 and the third quartile is 13. Therefore, the IQM is 13 - 7 = 6, indicating that the middle 50% of the data range is 6 units.

[0035] First adjustable coefficient This is an adjustable hyperparameter. When the set value is small, the dynamic silence interval threshold is lower, and almost any pause exceeding the average level will be suspected of needing punctuation. When the set value is large, only exceptionally long pauses will be marked as needing punctuation, suitable for applications that ensure every predicted punctuation point has a very high confidence level. In this embodiment, The value range is [1.5, 3.0].

[0036] As can be seen, this optional embodiment can effectively filter out the set of normal silence intervals by calculating the interquartile range of the silence interval, which enhances the accuracy of the analysis of silence intervals and makes the calculation of the dynamic silence interval threshold more targeted, thereby improving the performance of the punctuation prediction model and the accuracy of the generated text.

[0037] In an optional embodiment, the step of filtering the set of normal silence intervals from all silence intervals based on the interquartile range includes: Calculate the upper limit of the silence interval using the following formula. : ; in, This represents the third quartile of all silence intervals. This refers to the interquartile range. This represents the second adjustable coefficient; The set of normal silence intervals is obtained by filtering out silence intervals that are smaller than the upper limit of the silence interval from all silence intervals.

[0038] In this optional embodiment, the second adjustable coefficient The value can be adjusted according to the actual application scenario and data distribution characteristics. It is recommended to take a value in the range of 1.0 to 3.0 to achieve the best balance between precision and recall in punctuation prediction. (Example from this application) The value is 1.5.

[0039] As can be seen, this optional embodiment, by excluding abnormally long pauses, makes the set of normal silence intervals more accurately reflect the characteristics of natural speech rate, provides a clean data basis for dynamic threshold calculation, avoids threshold offset caused by outliers, and enhances robustness to extreme speech rate changes.

[0040] In an optional embodiment, the step of generating a possible set of punctuation marks corresponding to each character to be punctuated, based on the relative positions of its corresponding silence interval among all silence intervals, includes: The relative quantile of each punctuation character to be added is calculated using the following formula. : ; in, This indicates the silence interval corresponding to the punctuation character to be added. This represents the first quartile of all silence intervals. This represents the third quartile of all silence intervals; Based on the punctuation marks corresponding to the preset relative punctuation range, a set of possible punctuation marks corresponding to the character to be punctuated is generated.

[0041] In natural language, pauses after smaller grammatical units, such as within phrases or between parallel components, are typically short and often use connecting punctuation marks like commas and pauses. Pauses after larger grammatical units, such as between clauses or at the end of sentences, are typically longer and often use terminating punctuation marks like periods and question marks. This application will use relative quantile values... Dividing the punctuation from smallest to largest corresponds to a continuous change from short to long pauses, thus mapping to a change in syntactic function from connectors to terminators. One possible punctuation setting corresponding to a relative range of punctuation marks is: when At that time, the possible punctuation set is {","}; when At that time, the possible punctuation set is {"," "、"}; when At that time, the possible punctuation set is {";" "."}; when At that time, the possible punctuation set is {“.” “?” “!”}.

[0042] All of these are configurable parameters, satisfying... For example, the values ​​can be 1, 2, and 3 respectively.

[0043] In an optional embodiment, when multiple languages ​​appear in the target audio, the relative quantile values ​​are... Adaptive adjustments are made for different languages.

[0044] As can be seen, this optional embodiment converts the absolute silence duration into a distributed location parameter, eliminating the influence of differences in speaker speed. By mapping the preset quantile range to the punctuation type, the candidate punctuation set is generated in accordance with the statistical laws of speech pauses, thus improving the accuracy of the generated punctuation.

[0045] S105. Input the text sequence and the set of possible punctuation marks corresponding to each character into the punctuation prediction model trained by the context to predict the punctuation marks, and obtain the target document including the punctuation marks.

[0046] In this optional embodiment, the punctuation prediction model is a large language model that has undergone context learning. Context learning refers to the large language model using contextual information from the input text to understand and generate more relevant and appropriate responses without updating its parameters. By analyzing the context, the large language model can capture the semantics and intent in sentences, paragraphs, or even larger blocks of text, thereby improving the accuracy and coherence of its responses.

[0047] Here is an example of input to a large language model when performing context learning: Please choose one of the following punctuation marks and fill it in the punctuation position in the sentence: Sentence: Then let's look at question 4. 5.4a minus 3.8a. We observe that both 4.5 and 3.8 are multiplied by a, and then subtracted.

[0048] Optional punctuation list (only one can be selected, and it must be output as is): ["", ",", ".", "?", "!", "other"] Please return only the punctuation mark you selected (or an empty string ""), without outputting any explanation.

[0049] In an optional embodiment, when the target audio contains a multi-person dialogue, the target audio is split into multiple corresponding sub-audio segments according to the speakers, and a document is generated for each sub-audio segment using the audio document generation method provided by this invention. Please refer to [link to relevant documentation]. Figure 2 As shown, Figure 2 This is a schematic diagram of the structure of an audio document generation system disclosed in an embodiment of this application. For example... Figure 2 As shown, the system includes: The first acquisition module 201 is used to acquire the text sequence corresponding to the target audio; The second acquisition module 202 is used to acquire the timestamp of each character in the text sequence in the target audio. The third acquisition module 203 is used to acquire the difference between the timestamps corresponding to each character and the next character, and to obtain the silence interval corresponding to each character. The generation module 204 is used to generate a set of possible punctuation marks for each character based on the relative position of each silence interval across all silence intervals. The prediction module 205 is used to input the text sequence and the set of possible punctuation marks corresponding to each character into the punctuation prediction model trained by the context to predict the punctuation marks and obtain the target document including the punctuation marks.

[0050] Specific limitations regarding the audio document generation system can be found in the limitations of the audio document generation method described above, and will not be repeated here. Each module in the aforementioned audio document generation system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware format within or independently of the processor in the electronic device, or stored in software format in the memory of the electronic device, so that the processor can call the corresponding operations of each module.

[0051] It should be noted that, in order to highlight the innovative aspects of this application, this embodiment does not include modules that are not closely related to solving the technical problems proposed in this application, but this does not mean that there are no other modules in this embodiment.

[0052] like Figure 3 As shown, the electronic device 1 provided in this application may include a memory 12, a processor 13 and a bus, and may also include a computer program stored in the memory 12 and executable on the processor 13, such as an audio document generation program.

[0053] The memory 12 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 12 can be an internal storage unit of the electronic device 1, such as a portable hard drive. In other embodiments, the memory 12 can be an external storage device of the electronic device 1, such as a plug-in portable hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 1. Furthermore, the memory 12 can include both internal and external storage units of the electronic device 1. The memory 12 can be used not only to store application software and various types of data installed on the electronic device 1, such as the code for generating audio documents, but also to temporarily store data that has been output or will be output.

[0054] In some embodiments, the processor 13 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 13 is the control unit of the electronic device 1, connecting various components of the electronic device 1 through various interfaces and lines. It executes programs or modules (such as audio document generation programs) stored in the memory 12, and calls data stored in the memory 12 to perform various functions and process data of the electronic device 1.

[0055] The processor 13 executes the operating system of the electronic device 1 and various installed applications. The processor 13 executes the applications to implement the steps in the above-described audio document generation method.

[0056] For example, the computer program may be divided into one or more modules, which are stored in the memory 12 and executed by the processor 13 to complete this application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device 1. For example, the computer program may be divided into a first acquisition module 201, a second acquisition module 202, a third acquisition module 203, a generation module 204, and a prediction module 205.

[0057] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium, which can be non-volatile or volatile. The software functional module stored in the storage medium includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute some functions of the audio document generation method described in the various embodiments of this application.

[0058] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described audio document generation method embodiments.

[0059] In summary, the audio document generation method, system, device, and medium disclosed in this application, by effectively identifying silence intervals in audio and associating them with characters, can more intelligently predict and insert punctuation marks, thereby generating more coherent and readable target documents. This not only improves text readability but also reduces the workload of manual correction, enabling users to obtain and use information more efficiently. Therefore, this application effectively overcomes the various shortcomings of the prior art and has high industrial application value.

[0060] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for generating audio documents, characterized in that, Includes the following steps: Obtain the text sequence corresponding to the target audio; Obtain the timestamp of each character in the text sequence in the target audio; By obtaining the difference between the timestamps of each character and the next character, the silence interval corresponding to each character can be obtained. Based on the relative position of each silence interval among all silence intervals, generate a set of possible punctuation marks for each character; The text sequence and the set of possible punctuation marks corresponding to each character are input into a punctuation prediction model trained on context to predict punctuation marks, thereby obtaining a target document that includes punctuation marks.

2. The method for generating audio documents according to claim 1, characterized in that, The steps for obtaining the timestamp of each character in the text sequence in the target audio include: The target audio and the text sequence are time-aligned, and the timestamp of each character in the text sequence in the target audio is extracted.

3. The method for generating audio documents according to claim 1, characterized in that, The steps for generating the possible punctuation set for each character based on the relative position of each silence interval across all silence intervals include: Calculate the dynamic silence interval threshold based on all silence intervals; Mark all characters whose silence interval exceeds the dynamic silence interval threshold as characters to be added with punctuation. For each character to be punctuated, a possible set of punctuation marks is generated based on the relative position of its corresponding silence interval among all silence intervals.

4. The method for generating audio documents according to claim 3, characterized in that, The steps for calculating the dynamic silence interval threshold based on all silence intervals include: Calculate the interquartile range for all silence intervals; Based on the interquartile range, a set of normal silence intervals is obtained by filtering from all silence intervals; Calculate the mean and standard deviation of all silence intervals in the set of normal silence intervals; The dynamic silence interval threshold is calculated using the following formula. : ； in, This represents the average value. This represents the standard deviation. This represents the first adjustable coefficient.

5. The method for generating audio documents according to claim 4, characterized in that, The steps for selecting a set of normal silence intervals from all silence intervals based on the interquartile range include: Calculate the upper limit of the silence interval using the following formula. : ； in, This represents the third quartile of all silence intervals. This refers to the interquartile range. This represents the second adjustable coefficient; The set of normal silence intervals is obtained by filtering out silence intervals that are smaller than the upper limit of the silence interval from all silence intervals.

6. The method for generating audio documents according to claim 3, characterized in that, For each character to be punctuated, the step of generating a possible set of punctuation marks corresponding to the character to be punctuated, based on the relative position of its corresponding silence interval among all silence intervals, includes: The relative quantile of each punctuation character to be added is calculated using the following formula. : ； in, This indicates the silence interval corresponding to the punctuation character to be added. This represents the first quartile of all silence intervals. This represents the third quartile of all silence intervals; Based on the punctuation marks corresponding to the preset relative punctuation range, a set of possible punctuation marks corresponding to the character to be punctuated is generated.

7. An audio document generation system, characterized in that, include: The first acquisition module is used to acquire the text sequence corresponding to the target audio; The second acquisition module is used to acquire the timestamp of each character in the text sequence in the target audio. The third acquisition module is used to obtain the difference between the timestamps of each character and the next character, and to obtain the silence interval corresponding to each character. The generation module is used to generate a set of possible punctuation marks for each character based on the relative position of each silence interval across all silence intervals. The prediction module is used to input the text sequence and the set of possible punctuation marks corresponding to each character into a punctuation prediction model trained on the context to predict the punctuation marks and obtain the target document including the punctuation marks.

8. An electronic device, the electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the audio document generation method according to any one of claims 1-6.

9. A computer-readable medium having stored thereon computer program instructions that can be executed by a processor to implement the audio document generation method as described in any one of claims 1-6.

10. A computer program product comprising a computer program that, when executed by a processor, implements the audio document generation method as described in any one of claims 1-6.

Images