A data preprocessing method and system of an audio book of a speech recognition model

Abstract

The application provides a data preprocessing method and system for an audio book of a speech recognition model, and belongs to the technical field of speech recognition, and specifically comprises the following steps: determining book quality reliable users in uploaded users based on distribution data of quality risk books in different albums of the uploaded users and historical browsing data; determining distribution data of quality risk books of albums corresponding to uploaded audio books of the image quality reliable users; and combining historical quality inspection data in similar albums of the albums corresponding to the uploaded audio books, determining a pre-quality inspection processing method of a speech recognition model before the uploaded audio books are put on the market, and improving the book quality of the audio books.

Description

Technical Field

[0001] This invention belongs to the field of speech recognition technology, and in particular relates to a data preprocessing method and system for audiobooks based on a speech recognition model. Background Technology

[0002] Audiobooks have become increasingly popular due to their convenience, but at the same time, the quality of audiobooks varies greatly, which has a significant impact on the user experience.

[0003] Existing invention patents often focus on quality control during the production process of audiobooks, such as invention patent applications CN202410203822.3 "A method, system, device and storage medium for producing multi-emotional audiobooks" and CN202310894064.X "An automatic audiobook generation method based on a multimodal large language model". However, none of the above technical solutions consider how to determine the pre-quality inspection strategy for audiobooks that have not yet been released. The albums in which specific audiobooks are located vary, and the quality risk factors such as the presence of harmful information also vary to a certain extent. If differentiated pre-quality inspection methods cannot be generated based on the differences in the albums in which the audiobooks are located, it will affect the user experience.

[0004] To address the aforementioned technical issues, this application provides a data preprocessing method and system for audiobooks based on a speech recognition model. Summary of the Invention

[0005] To achieve the objectives of this invention, the following technical solution is adopted:

[0006] Specifically, in the first aspect, this application provides a data preprocessing method for audiobooks based on a speech recognition model, which specifically includes:

[0007] S1 obtains the audiobook listing demand data from the audiobook platform. If it is determined based on the listing demand data that the audiobook data cannot be processed by manual quality inspection, proceed to the next step.

[0008] S2 uses the analysis results of user review data of the user's audiobooks to determine the matching of review keywords and preset risk keywords in audiobooks within different albums, and determines the quality risk books in the audiobooks based on the matching results;

[0009] S3 determines the book quality reliable user in the upload user based on the distribution data of the quality risk books in different albums of the upload user and the historical browsing data, determines the distribution data of the quality risk books of the album corresponding to the uploaded audio book of the book quality reliable user, and determines the pre-quality inspection processing method of the uploaded audio book before the upload by using the voice recognition model in combination with the historical quality inspection data in the similar albums of the album corresponding to the uploaded audio book.

[0010] The application has the following advantages:

[0011] Based on the matching of the comment keywords in the audio book and the preset risk keywords, the quality risk books in the audio book are determined, so that the quality risk books in the audio book are screened from the matching of the evaluation keywords and the preset risk keywords, and a foundation is laid for further generating a differentiated quality inspection processing strategy according to the upload data of the quality risk books of the user.

[0012] According to the distribution data of the quality risk books of the album corresponding to the uploaded audio book of the book quality reliable user, the historical quality inspection data in the similar albums of the album corresponding to the uploaded audio book, the pre-quality inspection processing method of the uploaded audio book before the upload by using the voice recognition model is determined, which considers the difference of the probability of the existence of bad information in the corresponding album type due to the distribution data of the quality risk books in the corresponding album, and also considers the difference of the quality risk caused by the difference of the quality inspection data of the bad information in the similar albums, so as to generate a differentiated quality inspection processing strategy from the perspective of quality risk, and improve the accuracy and efficiency of quality inspection processing.

[0013] The further technical scheme is that the upload demand data of the audio book includes the number of audio books to be uploaded and processed by the audio book platform in different dates.

[0014] The further technical scheme is that the quality inspection processing of the book data of the audio book cannot be performed by using artificial quality inspection, specifically including:

[0015] According to the upload demand data of the audio book, the audio books to be uploaded and processed in different dates are determined as upload demand books;

[0016] According to the number of upload demand books in different dates, the determination of the quality inspection processing busy date in the date is performed;

[0017] According to the proportion of the number of quality inspection processing busy dates, it is determined whether the quality inspection processing of the book data of the audio book can be performed by using artificial quality inspection.

[0018] The further technical solution is that the quality inspection processing busy date is a date in which the number of books to be put on the shelf is not within a preset demand quantity interval.

[0019] The further technical solution is that when the proportion of the number of quality inspection processing busy dates is greater than a preset quality inspection busy date proportion, it is determined that the quality inspection processing of the book data of the audio book cannot be performed in the manual quality inspection manner.

[0020] The further technical solution is that the pre-quality inspection processing method using the speech recognition model is:

[0021] The album corresponding to the uploaded audio book of the book quality reliable user is determined as a matching album, and the proportion of the number of quality risk books in the matching album is determined;

[0022] The proportion of the number of audio books with quality inspection quality problems in different similar albums is determined based on historical quality inspection data in similar albums of the matching album;

[0023] The average value of the proportion of the number of quality risk books of the matching album and the proportion of the number of audio books with quality inspection quality problems in different similar albums is determined as a quality defect risk value of the uploaded audio book, and the pre-quality inspection processing method using the speech recognition model is used for the uploaded audio book before being put on the shelf based on the quality defect risk value.

[0024] The further technical solution is that the pre-quality inspection processing method using the speech recognition model is used for the uploaded audio book before being put on the shelf based on the quality defect risk value, and specifically includes:

[0025] When the quality defect risk value is greater than a preset defect risk value threshold, the quality inspection processing of the book data of the audio book is performed in the manual quality inspection manner;

[0026] When the quality defect risk value is not greater than the preset defect risk value threshold, when the quality defect risk value is within a preset risk value interval, the quality inspection processing of the book data of the audio book is performed using the speech recognition model;

[0027] When the quality defect risk value is not within the preset risk value interval, the quality inspection processing of the book data of the audio book is performed using a preset keyword spectrum matching condition.

[0028] The further technical solution is that the quality inspection processing of the book data of the audio book is performed using the speech recognition model, and specifically includes:

[0029] The analysis processing of the book data of the audio book is performed by using the voice recognition model to obtain text, and the quality inspection processing of the book data of the audio book is performed based on the text and a preset semantic recognition model to determine whether there is illegal content.

[0030] Further technical solutions are to use the sound spectrum matching condition of the preset keyword to perform the quality inspection processing of the book data of the audio book, specifically including:

[0031] The sound spectrum matching condition of the preset keyword is used to determine whether the preset keyword exists in the book data of the audio book, and the quality inspection processing result is determined according to whether the preset keyword exists.

[0032] In a second aspect, the present application provides a computer system, comprising a memory and a processor connected in communication, and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the computer program to perform the above-mentioned data preprocessing method of an audio book of a voice recognition model.

[0033] Other features and advantages will be set forth in the following description, and the objectives and other advantages of the present application will be achieved and obtained by the structure particularly pointed out in the description and the drawings.

[0034] In order to make the above-mentioned objectives, features and advantages of the present application more obvious and easy to understand, the following preferred embodiments are specifically described below, and the accompanying drawings are described in detail as follows. BRIEF DESCRIPTION OF DRAWINGS

[0035] The above-mentioned and other features and advantages of the present application will become more apparent by describing example embodiments thereof with reference to the accompanying drawings.

[0036] Figure 1 A flowchart of a data preprocessing method of an audio book of a voice recognition model;

[0037] Figure 2 A flowchart of determining that the quality inspection processing of the book data of the audio book cannot be performed by using artificial quality inspection;

[0038] Figure 3 A flowchart of a method for determining quality risk books in an audio book;

[0039] Figure 4 A flowchart of a method for determining reliable book quality users in uploaded users. DETAILED DESCRIPTION

[0040] In order for those skilled in the art to better understand the technical solutions in the specification, the technical solutions in the specification will be clearly and completely described below in combination with the drawings in the specification. Obviously, the described embodiments are only part of the embodiments of the specification, not all. Based on the embodiments of the specification, all other embodiments obtained by those of ordinary skill in the art without creative labor should be within the scope of protection of the specification.

[0041] In the present application, according to the proportion of books with bad information in the album where the user uploads the book in the audiobook platform, a differentiated quality inspection processing strategy is generated, and the efficiency of quality inspection processing is improved.

[0042] Embodiment 1

[0043] As Figure 1 shown, the present application provides a data preprocessing method for audio books of a speech recognition model, which specifically includes:

[0044] S1 obtains the shelving demand data of audio books of an audiobook platform, and when it is determined that the quality inspection processing of the book data of the audio books cannot be performed by manual quality inspection based on the shelving demand data, the next step is entered;

[0045] Further, the shelving demand data of the audio books includes the number of audio books to be shelved in different dates of the audiobook platform.

[0046] Specifically, as Figure 2 shown, it is determined that the quality inspection processing of the book data of the audio books cannot be performed by manual quality inspection, which specifically includes:

[0047] Based on the shelving demand data of the audio books, the audio books to be shelved in different dates are determined, and they are used as shelving demand books;

[0048] According to the number of shelving demand books in different dates, the determination of the busy date of quality inspection processing in the date is performed;

[0049] According to the number proportion of the busy date of quality inspection processing, it is determined whether the quality inspection processing of the book data of the audio books can be performed by manual quality inspection.

[0050] Further, the busy date of quality inspection processing is a date in which the number of shelving demand books is not within a preset demand quantity interval.

[0051] It should be noted that when the number proportion of the busy date of quality inspection processing is greater than a preset busy date number proportion of quality inspection, it is determined that the quality inspection processing of the book data of the audio books cannot be performed by manual quality inspection.

[0052] Optionally, it is determined that the quality inspection processing of the book data of the audio book cannot be performed in the manner of manual quality inspection, specifically including:

[0053] With the shelving demand data of the audio book, the audio books to be shelved in the nearest preset time period are determined as the shelving demand books.

[0054] According to the number of shelving demand books, it is determined whether the quality inspection processing of the book data of the audio book can be performed in the manner of manual quality inspection.

[0055] Further, when the number of shelving demand books is not in the preset demand quantity interval, it is determined that the quality inspection processing of the book data of the audio book can be performed in the manner of manual quality inspection.

[0056] In another possible embodiment, it is determined that the quality inspection processing of the book data of the audio book cannot be performed in the manner of manual quality inspection, specifically including:

[0057] S11 With the shelving demand data of the audio book, the audio books to be shelved in the nearest preset time period are determined as the shelving demand books.

[0058] It can be understood that before proceeding to the next step, it is also necessary to determine that when the number of shelving demand books is not in the preset demand quantity interval, i.e. when the number of shelving demand books is too large, it is determined that the quality inspection processing of the book data of the audio book cannot be performed in the manner of manual quality inspection, and only when the number of shelving demand books is not too large, the determination of the book distribution aggregation value is required.

[0059] S12 Based on the interval time length between different shelving demand books and combined with the number of shelving demand books in the nearest preset time period, the book distribution aggregation value in the nearest preset time period is determined.

[0060] In addition, it should be noted that the book distribution aggregation value in the above steps is determined by the output value of the mathematical model with the interval time length between different shelving demand books and the number of shelving demand books in the nearest preset time period as input values, wherein the mathematical model can be constructed by using existing analytic hierarchy process, neural network model.

[0061] Specifically, when the book distribution aggregation value in the nearest preset time period does not meet the requirements, i.e. when the book distribution aggregation value is greater than a certain threshold value, it is determined that the quality inspection processing of the book data of the audio book cannot be performed in the manner of manual quality inspection, and in other cases, the determination of the busy date for quality inspection processing is performed.

[0062] S13 determines the quality inspection processing busy date in the date according to the number of shelving demand books in the date, and determines the quality inspection processing busy value of different quality inspection processing busy dates based on the number of shelving demand books in different quality inspection processing busy dates and the time length of shelving demand books;

[0063] It should be noted that the quality inspection processing busy date is a date in which the number of shelving demand books is within the target range, and the quality inspection processing busy value of the quality inspection processing busy date can be determined according to the product of the total time length of shelving demand books and the preset proportion factor based on the number of shelving demand books in different quality inspection processing busy dates and the time length of shelving demand books.

[0064] Optionally, before entering the next step, it is also necessary to determine whether the quality inspection processing busy value meets the requirements, that is, the number of quality inspection processing busy dates with larger quality inspection processing busy values is not within the preset busy date number interval, then it is determined that the artificial quality inspection mode cannot be used for the quality inspection processing of the audio book data of the audio book, and specifically, the threshold value can be used to determine whether the quality inspection processing busy value meets the requirements, and only when it is within the preset busy date number interval, the next step is entered.

[0065] S14 determines the quality inspection processing delay probability of the audio book platform based on the proportion of the number of quality inspection processing busy dates and the quality inspection processing busy value of different quality inspection processing busy dates, and combines the average number of shelving demand books in different dates, and determines whether the artificial quality inspection mode can be used for the quality inspection processing of the audio book data of the audio book based on the quality inspection processing delay probability.

[0066] In one embodiment, the quality inspection processing delay probability of the audio book platform can be determined by using the average of the product of the proportion of the number of quality inspection processing busy dates, the average of the quality inspection processing busy value of different quality inspection processing busy dates, the average number of shelving demand books in different dates, and the preset proportion factor.

[0067] Further, when the quality inspection processing delay probability is greater than the preset delay probability threshold, it is determined that the artificial quality inspection mode can be used for the quality inspection processing of the audio book data of the audio book.

[0068] S2 determines the matching situation of the comment keywords in the audio books in different albums with the preset risk keywords based on the analysis result of the user comment data of the audio books of the user, and determines the quality risk books in the audio books based on the matching situation;

[0069] Specifically, the user comment data includes the number of comments and comment sentences under different audio books.

[0070] Further, the preset risk keyword is built by using a keyword library containing adverse information.

[0071] Specifically, as shown in the method for determining the quality risk book in the audio book is: Figure 3

[0072] Based on the matching of the comment keywords in the audio book and the preset risk keywords, it is determined that there are comments matching the preset risk keywords, and they are regarded as abnormal comments;

[0073] Users containing abnormal comments are regarded as abnormal comment users;

[0074] According to the number of abnormal comment users, it is determined whether the audio book is a quality risk book.

[0075] Further, when the number of abnormal comment users of the audio book is greater than the preset abnormal comment user number threshold, it is determined that the audio book is a quality risk book.

[0076] Optionally, the method for determining the quality risk book in the audio book is:

[0077] Based on the matching of the comment keywords in the audio book and the preset risk keywords, it is determined that there are comments matching the preset risk keywords, and they are regarded as abnormal comments;

[0078] Users containing abnormal comments are regarded as abnormal comment users;

[0079] According to the ratio of the number of abnormal comment users to the number of listening users of the audio book, it is determined whether the audio book is a quality risk book.

[0080] Further, when the ratio of the number of abnormal comment users to the number of listening users of the audio book is greater than the preset abnormal user number threshold, it is determined that the audio book is a quality risk book.

[0081] S3 determines the book quality reliable user in the upload user based on the distribution data of the quality risk book in different albums and the historical browsing data of the upload user, determines the distribution data of the quality risk book of the album corresponding to the upload audio book of the book quality reliable user, and combines the historical quality inspection data in the similar albums of the album corresponding to the upload audio book to determine the pre-quality inspection processing method of the upload audio book before listing using the voice recognition model.

[0082] Further, the album of the quality risk book is determined according to the album of the title of the quality risk book. ​

[0083] Specifically, as shown in Figure 4 The method for determining the book quality reliable user in the uploading user is:

[0084] The number proportion of the quality risk books in different albums is determined based on the distribution data of the quality risk books in different albums, and the quality risk value of different albums is determined based on the number proportion of the quality risk books.

[0085] The historical browsing times of different albums are determined based on the historical browsing data of different albums, and the determination of the hot album in the album is performed based on the historical browsing times.

[0086] The uploading user is determined to be a book quality reliable user according to the average value of the quality risk value of different hot albums.

[0087] Further, the hot album in the album is an album with a historical browsing time greater than a preset browsing time threshold.

[0088] It should be noted that when the average value of the quality risk value of different hot albums is greater than a preset quality risk value threshold, it is determined that the uploading user is not a book quality reliable user.

[0089] It can be understood that when the uploading user is not a book quality reliable user, all uploaded audio books of the uploading user are subjected to quality inspection of the book data of the audio book in the manner of artificial quality inspection.

[0090] Optionally, the method for determining the book quality reliable user in the uploading user is:

[0091] S31 The number proportion of the quality risk books in different albums is determined based on the distribution data of the quality risk books in different albums, and the quality risk value of different albums is determined based on the number proportion of the quality risk books.

[0092] Optionally, it is also necessary to determine whether the number of quality risk books of the uploading user meets the requirement in step S31. Specifically, when the number of quality risk books of the uploading user is too large, it can be determined that the uploading user is not a book quality reliable user.

[0093] In addition, it should be noted that when the number of quality risk books of the uploading user is not too large, it is also necessary to determine that when the quality risk value of different albums is within a preset risk value interval, i.e., is too small, if the number of albums with quality risk books meets the requirement, i.e., is too small, at this time, it is determined that the uploading user is a book quality reliable user.

[0094] When the number of albums with quality risk values not in the preset risk value interval does not meet the requirement, when the basic user risk value of the uploading user does not meet the requirement, it is determined that the uploading user does not belong to the book quality reliable user, and when the number of albums with quality risk values not in the preset risk value interval meets the requirement, the basic user risk value of the uploading user is determined based on the quality risk values of different albums and the number of audio books in different albums.

[0095] In one possible embodiment, the basic user risk value of the uploading user is determined according to the sum of the products of the quality risk values of different albums and the number of audio books in different albums, and whether the basic user risk value meets the requirement can be determined by a threshold value.

[0096] S32 determines the historical browsing times of different albums based on the historical browsing data of different albums, and determines the comment reliability value of different albums based on the historical browsing times and the number of audio books in different albums.

[0097] It should be noted that the determination of the hot album in the album based on the historical browsing times is that the album with a historical browsing time greater than a preset browsing time threshold value is determined as a hot album, and when the average value of the quality risk values of the hot album is large, i.e., does not meet the requirement, which can be determined by a threshold value, it is determined that the uploading user does not belong to the book quality reliable user.

[0098] When the average value of the quality risk values of the hot album meets the requirement, the comment reliability value of different albums is determined based on the ratio of the historical browsing times to the number of audio books, when there is an album with a comment reliability value greater than a preset reliability value threshold value, when the average value of the quality risk values of the album with the comment reliability value greater than the preset reliability value threshold value does not meet the requirement, it is determined that the uploading user does not belong to the book quality reliable user, when there is no album with a comment reliability value greater than a preset reliability value threshold value, it is determined that the uploading user does not belong to the book quality reliable user, and only when the average value of the quality risk values of the album with the comment reliability value greater than the preset reliability value threshold value meets the requirement, i.e., is small, the step S33 is entered, and whether the requirement is met is determined by a threshold value.

[0099] S33 determines a user quality risk value of the uploading user according to the quality risk values of different albums and the comment reliability values, and determines whether the uploading user is a book quality reliable user based on the user quality risk value.

[0100] It should be noted that the user quality risk value is determined according to the sum of the products of the quality risk values of different albums and the comment reliability values, wherein when the user quality risk value is greater than a threshold value, for example, greater than 0.5, it is determined that the uploading user does not belong to the book quality reliable user.

[0101] Further, the similar album is another album corresponding to a similar number of introduction keywords of the album, which is greater than a preset keyword number.

[0102] Specifically, the pre-quality inspection processing method using the speech recognition model is:

[0103] The album corresponding to the uploading audio book of the book quality reliable user is taken as a matching album, and the number proportion of quality risk books in the matching album is determined;

[0104] The historical quality inspection data in the similar album of the matching album is used to determine the number proportion of audio books with quality inspection quality problems in different similar albums;

[0105] According to the number proportion of quality risk books in the matching album and the average value of the number proportion of audio books with quality inspection quality problems in different similar albums, a quality defect risk value of the uploading audio book is determined, and the pre-quality inspection processing method using the speech recognition model is used to determine the uploading audio book of the book quality reliable user before being put on the shelf.

[0106] Further, based on the quality defect risk value, the pre-quality inspection processing method using the speech recognition model is used to determine the uploading audio book before being put on the shelf, specifically including:

[0107] When the quality defect risk value is greater than a preset defect risk value threshold, the quality inspection processing of the book data of the audio book is performed by using artificial quality inspection;

[0108] When the quality defect risk value is not greater than the preset defect risk value threshold, when the quality defect risk value is within a preset risk value interval, the quality inspection processing of the book data of the audio book is performed by using the speech recognition model;

[0109] When the quality defect risk value is not within the preset risk value interval, the quality inspection processing of the book data of the audio book is performed by using the preset keyword spectrum matching condition.

[0110] It should be noted that the quality inspection processing of the audio book data of the audio book by using the speech recognition model specifically includes:

[0111] The text is obtained by performing the analysis processing of the audio book data of the audio book by using the speech recognition model, and the quality inspection processing of the audio book data of the audio book is performed based on the text and a preset semantic recognition model to determine whether there is illegal content.

[0112] It can be understood that the quality inspection processing of the audio book data of the audio book by using the sound spectrum matching of the preset keyword specifically includes:

[0113] The sound spectrum matching of the preset keyword is used to determine whether there is a preset keyword in the audio book data of the audio book, and the quality inspection processing result is determined according to whether there is a preset keyword.

[0114] Embodiment 2

[0115] In a second aspect, the present application provides a computer system, comprising a memory and a processor connected in communication, and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the computer program to perform the above-mentioned data preprocessing method of the audio book of the speech recognition model.

[0116] Each of the embodiments in the specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0117] The above describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different than the order in which they are recited, and still achieve desirable results. In addition, the processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In some implementations, multitasking and parallel processing can be advantageous.

[0118] The above only describes one or more embodiments of the present application and does not limit the present application. One or more embodiments of the present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of the present application should be included in the scope of claims of the present application.

Claims

1. A data preprocessing method for audiobooks using a speech recognition model, characterized in that, Specifically, it includes: Obtain the audiobook listing demand data from the audiobook platform. If, based on the listing demand data, it is determined that manual quality inspection cannot be used to process the audiobook data, proceed to the next step. Based on the analysis results of user review data of audiobooks uploaded by users, the matching situation of review keywords and preset risk keywords in audiobooks within different albums is determined, and the quality risk books in the audiobooks are determined based on the matching situation; Based on the distribution data of quality risk books in different albums of the uploaded users and historical browsing data, reliable users of books among the uploaded users are identified, the distribution data of quality risk books in the albums corresponding to the uploaded audiobooks of the reliable users are identified, and combined with the historical quality inspection data in similar albums corresponding to the uploaded audiobooks, a pre-quality inspection processing method using a speech recognition model is determined before the uploaded audiobooks are put on the shelves. The pre-quality control method using the speech recognition model is as follows: The albums corresponding to the audiobooks uploaded by users with reliable book quality are used as matching albums to determine the proportion of books with quality risks in the matching albums. Based on historical quality inspection data from similar albums of the matched album, determine the percentage of audiobooks with quality inspection problems in different similar albums; Based on the percentage of quality-risk books in the matched album and the average percentage of audiobooks with quality inspection problems in different similar albums, the quality defect risk value of the uploaded audiobook is determined. Based on the quality defect risk value, the uploaded audiobooks of reliable users are pre-processed using a speech recognition model before being put on the shelves.

2. The data preprocessing method for audiobooks using the speech recognition model as described in claim 1, characterized in that, The audiobook listing demand data includes the number of audiobooks to be listed on the audiobook platform on different dates.

3. The data preprocessing method for audiobooks using the speech recognition model as described in claim 1, characterized in that, For audiobooks whose data quality control cannot be performed manually, the following are examples: Based on the audiobook listing demand data, determine the audiobooks that need to be listed on different dates and use them as listing demand books; Based on the number of books required for shelving on different dates, determine the busiest days for quality inspection processing on those dates; Based on the percentage of busy days for quality inspection, it can be determined whether manual quality inspection can be used for the quality inspection of audiobook data.

4. The data preprocessing method for audiobooks using the speech recognition model as described in claim 3, characterized in that, The busy days for quality inspection processing are those days when the number of books to be put on the shelves is not within the preset demand range.

5. The data preprocessing method for audiobooks using the speech recognition model as described in claim 1, characterized in that, The user review data includes the number of reviews and the types of comments for different audiobooks.

6. The data preprocessing method for audiobooks using the speech recognition model as described in claim 1, characterized in that, The preset risk keywords are constructed using a keyword library containing negative information.

7. The data preprocessing method for audiobooks using the speech recognition model as described in claim 1, characterized in that, The similar albums are other albums whose description keywords are more similar to the album than the preset number of keywords.

8. A computer system, comprising: A memory and processor connected by communication, and a computer program stored in the memory and capable of running on the processor, characterized in that, when the processor runs the computer program, it performs a data preprocessing method for an audiobook with a speech recognition model as described in any one of claims 1-7.

Images