Powerpoint generation method and digital human video generation method

Abstract

The invention provides a PowerPoint generation method and a digital human video generation method, relates to the technical field of computers, and is characterized in that a PowerPoint explanation document with a paragraph structure is determined by using user input information, and then a target PowerPoint is generated page by page by using the PowerPoint explanation document with the paragraph structure. According to the method and the device, each paragraph in the presentation explanation document corresponds to one presentation page in the target presentation, so that the matching degree of typesetting and content of the target presentation can be improved, the target presentation meets the application requirements of a user, and the satisfaction degree of the user is improved. Moreover, according to the method, the user is prevented from manually selecting the PowerPoint template, so that the operation actions of the user are reduced, and the possibility that the problems of non-uniform styles, logic contradictions and the like of the target PowerPoint can be reduced.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method for generating presentations and a method for generating digital human videos. Background Technology

[0002] A presentation is a visual display file that combines multimedia elements such as text, images, charts, and animations. It is typically used in scenarios such as speeches, teaching, or reports to help convey information and can be called slides (Microsoft Office PowerPoint, PPT), transparencies, or other names.

[0003] With the rapid development of generative artificial intelligence technology, intelligent PPT generation technology can automatically generate corresponding PPT files based on user-imported Word files and user-selected PPT templates. Users can then modify the PPT files according to their actual needs to obtain the final PPT file. Using this generation technology, users no longer need to create PPT files from scratch; they only need to modify the generated PPT files, greatly improving the efficiency of PPT creation.

[0004] With the emergence of large language models, it has become possible to generate a PPT file outline based on user input commands, and then generate the PPT file based on the outline. This method eliminates the need for users to import Word files, making PPT file creation more convenient and faster.

[0005] However, PPT files generated based on user input often suffer from mismatches between layout and content. Even when users manually select a PPT template, inconsistencies in template layout and content can lead to inconsistent file styles and logical contradictions. Summary of the Invention

[0006] This invention provides a method for generating presentations and a method for generating digital human videos, in order to overcome the deficiencies existing in related technologies.

[0007] This invention provides a method for generating presentations, comprising: Obtain user input information; Based on the user input information, a presentation document with a paragraph structure is determined; each paragraph in the presentation document corresponds to a presentation page; Based on the presentation document, the target presentation is generated page by page.

[0008] The present invention also provides a method for generating digital human videos, comprising: Obtain the target presentation; the target presentation is generated based on the presentation generation method described above; Identify the presentation explanation document with paragraph structure corresponding to the target presentation, and generate a digital human video based on the presentation explanation document.

[0009] The present invention also provides a presentation generation device, comprising: The information acquisition module is used to acquire user input information; The presentation document determination module is used to determine a presentation document with a paragraph structure based on the user input information; each paragraph in the presentation document corresponds to a presentation page; The presentation generation module is used to generate the target presentation page by page based on the presentation explanation document.

[0010] The present invention also provides a digital human video generation apparatus, comprising: The document acquisition module is used to acquire the target presentation document; the target presentation document is generated based on the presentation document generation method described above. The digital human video generation module is used to determine the presentation document with paragraph structure corresponding to the target presentation document, and generate a digital human video based on the presentation document.

[0011] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the presentation generation method or the digital human video generation method described above.

[0012] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the presentation generation method or the digital human video generation method as described above.

[0013] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the presentation generation method or the digital human video generation method as described above.

[0014] The presentation generation method and digital human video generation method provided by this invention utilize user input information to determine a presentation document with a paragraph structure. Then, using this paragraph-structured presentation document, a target presentation is generated page by page. Each paragraph in the presentation document corresponds to a page in the target presentation, improving the matching degree between the layout and content of the target presentation, ensuring it meets the user's application needs and increasing user satisfaction. Furthermore, this method avoids users manually selecting presentation templates, reducing user actions and minimizing the possibility of inconsistent style or logical contradictions in the target presentation. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the presentation generation method provided by the present invention.

[0017] Figure 2 This is a flowchart illustrating the digital human video generation method provided by the present invention.

[0018] Figure 3 This is a schematic diagram of the presentation generation device provided by the present invention.

[0019] Figure 4 This is a schematic diagram of the structure of the digital human video generation device provided by the present invention.

[0020] Figure 5 A schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] In existing technologies, when generating PPT files using user commands, the process typically begins by using the user commands to generate the title of each slide through a large language model. Then, the title of each slide is used to generate a summary or specific content. However, since the text generation capability of a large language model is dependent on the training data, to enable it to generate professional text in a specific domain, a large amount of text data from that domain is needed to fine-tune the general-purpose large language model. Because PPT files are used in various fields, it's impossible to achieve domain-specific professional text generation capabilities through fine-tuning alone. This results in titles and content generated by the large language model based on user input being rather vague and lacking in consistency between content and title, failing to meet user needs and requiring further refinement by the user. Furthermore, the generated PPT files also suffer from layout and content mismatch issues.

[0023] Users can also manually select a PPT template to fill in the content summary or specific content of each page of the generated PPT. However, this method still has the problem of inconsistent PPT file style and logical contradictions due to the mismatch between the PPT template layout and the PPT file content.

[0024] Based on this, this embodiment of the invention provides a method for generating presentation slides. For example... Figure 1 As shown, the method includes: S11, Obtain user input information; S12, Based on user input information, determine the presentation document with paragraph structure; each paragraph in the presentation document corresponds to a presentation page; S13, Generate the target presentation page by page based on the presentation document.

[0025] Specifically, the presentation generation method provided in this embodiment of the invention is executed by a presentation generation device, which can be configured in a computer. The computer can be a local computer or a cloud computer. The local computer can be a computer, tablet, etc., and no specific limitation is made here.

[0026] First, step S11 is executed to obtain user input information. Here, user input information refers to the relevant information that the user can provide to obtain the target presentation. The user can provide user input information to the executing entity through the human-computer interaction interface.

[0027] User input may include user instructions that specify the key information the target presentation must meet, such as the theme information. In addition, user input may include a knowledge base containing background knowledge related to the theme information of the target presentation. This knowledge base may be in the form of a text file, Word document, or PDF file containing the background knowledge.

[0028] User input may also include an initial explanatory document, which is a presentation script provided by the user corresponding to the target presentation.

[0029] Next, step S12 is executed, using user input information to determine a presentation document with a paragraph structure. This presentation document includes multiple paragraphs, each corresponding to a presentation page; that is, each paragraph can serve as the content for one presentation page.

[0030] Understandably, the number of paragraphs in the presentation's explanatory document can differ from the number of presentation pages. The number of paragraphs in the presentation's explanatory document can be less than the number of presentation pages, so that the target presentation can be optimized after it is generated. For example, an optimized presentation page containing video or audio can be inserted into the target presentation. This optimized presentation page can be played directly without explanation, and therefore does not have the corresponding paragraphs in the presentation's explanatory document.

[0031] In this step, different methods can be used to determine the presentation document based on the different content contained in the user input information. For example, if the user input information includes user instructions, the presentation document can be generated by generating a content outline using the user instructions; if the user input information includes both user instructions and a knowledge base, the presentation document can be generated by retrieving the knowledge base using the user instructions and generating a content outline using the search results; if the user input information includes an initial presentation document, the initial presentation document can be segmented by extracting a content outline from the initial presentation document, and then the presentation document can be generated.

[0032] Finally, step S13 is executed. Since the presentation document contains a paragraph structure, the target presentation can be generated page by page according to the content of each paragraph in the presentation document and the order of the paragraphs in the presentation document.

[0033] Here, generating the target presentation page by page means generating each presentation page in the target presentation page by page. The process of generating presentation pages is the process of generating the page content of each presentation page. During the process of generating the target presentation page by page, the page content of each already generated presentation page and the paragraph content corresponding to the current presentation page can also be combined to generate the current presentation page.

[0034] Understandably, the content of each presentation page can include the content of text boxes and the content of image boxes. The content of text boxes and the images of image boxes in each presentation page can be generated by using the paragraph content corresponding to each presentation page or by combining the page content of each already generated presentation page. Then, the text boxes and image boxes are filled to obtain the page content of each presentation page.

[0035] The presentation generation method provided in this embodiment of the invention utilizes user input information to determine a presentation document with a paragraph structure. Then, using this paragraph-structured document, a target presentation is generated page by page. This ensures that each paragraph in the presentation document corresponds to a page in the target presentation, improving the matching degree between the layout and content of the target presentation. This allows the target presentation to meet the user's application needs and enhances user satisfaction. Furthermore, this method avoids the need for users to manually select presentation templates, reducing user actions and minimizing the possibility of inconsistent styles or logical contradictions in the target presentation.

[0036] Based on the above embodiments, the user input information includes user instructions; based on the user input information, a presentation document with a paragraph structure is determined, including: Generate a content outline based on user instructions; Based on user instructions and content outline, generate presentation slides and explanation documents.

[0037] Specifically, when user input includes user commands, to determine the presentation document, the user commands can be used to generate a content outline for the presentation document using a large language model. For example, user commands can be input into the large language model, and the powerful text generation capabilities of the large language model can be used to generate a content outline.

[0038] Understandably, a content outline can include at least one heading at a specified level, and further include tagging results obtained by marking each heading according to the presentation page type in the target presentation. Each heading at the specified level can serve as a heading in subsequent presentation pages of the target presentation, and the tagging results can be used to distinguish different parts of the content outline.

[0039] If the outline includes at least one heading at a specified level, an example of the outline could be the content in the box below: If a content outline includes at least one heading at a specified level, along with the tagging of each heading according to the presentation page type in the target presentation, then an example of a content outline could be the content shown in the box below: The specified heading level can include first-level headings and second-level headings, etc. Presentation page types in the target presentation can include cover pages, table of contents pages, title pages, detail pages, and ending pages. Each cover page, table of contents page, and ending page can each contain one presentation page, while title pages and detail pages can each contain multiple presentation pages. Each title page must correspond to at least one detail page. Different presentation page types can have different tagging results; for example, the tagging result for a cover page could be an opening statement, for a table of contents page could be a table of contents introduction, and for an ending page could be a closing statement.

[0040] Subsequently, using user instructions and content outlines, the large language model is applied again to generate the presentation document. That is, user instructions and content outlines are simultaneously input into the large language model, and the large language model generates and outputs the presentation document.

[0041] In the presentation documentation, each paragraph can correspond to a specific page type in the target presentation.

[0042] As an example, the presentation documentation could be the content shown in the box below: It is understandable that in the above example of presentation slides, each slide corresponds to a paragraph in the presentation slide document, and the content of each slide corresponds to the content of a paragraph in the presentation slide document.

[0043] In this embodiment of the invention, a content outline is generated through user instructions, and a presentation document is generated by combining the user instructions and the content outline. This not only improves the matching degree between the presentation document and the user input information, but also avoids the problems of empty or unreasonable content and mismatch between content and title caused by directly using the title of each PPT page to generate the content summary or specific content of that PPT page in the prior art.

[0044] Based on the above embodiments, the user input information also includes a knowledge base; Based on user instructions, a content outline is generated, which previously included: The knowledge base is divided into knowledge blocks to obtain multiple knowledge blocks and the corresponding knowledge block index; Based on the index of each knowledge block, determine the target knowledge block that matches the user's command; Based on user instructions, a content outline is generated, including: Generate a content outline based on user instructions and target knowledge blocks; And / or, based on user instructions and a content outline, generate presentation documentation including: Based on user instructions, content outlines, and target knowledge blocks, generate presentation slides and explanation documents.

[0045] Specifically, when the user input information includes user instructions and a knowledge base, the knowledge base can be preprocessed first. Preprocessing operations may include, but are not limited to, parsing the hierarchical logical structure of the knowledge base, dividing it into knowledge blocks, and building an index to obtain the knowledge block index corresponding to each knowledge block.

[0046] When parsing the hierarchical logical structure of the knowledge base, the natural paragraphs in the knowledge base can be labeled first. Then, the background knowledge in the knowledge base is input into the large language model. The prompt word instructs the large language model to output the hierarchical heading outline of the knowledge base. The hierarchical heading outline includes the name, level and label of the natural paragraphs it covers for each heading.

[0047] When segmenting knowledge in a knowledge base, the hierarchical heading outline can be used to segment the background knowledge. The knowledge covered by the finest-grained headings can serve as an initial knowledge block. If the word count of this initial knowledge block does not exceed a word threshold, it is considered a final knowledge block; otherwise, a topic shift point extraction algorithm is used to detect the positions of sentences where topic shifts occur within the initial knowledge block. These sentence positions are then used as demarcation points to further segment the initial knowledge block, resulting in multiple final knowledge blocks.

[0048] It is understandable that each knowledge block obtained by dividing the knowledge base into knowledge blocks can be regarded as a whole knowledge point. The content theme within a knowledge block is relatively concentrated, while the content theme between different knowledge blocks is different.

[0049] During the knowledge segmentation process of the knowledge base, a knowledge block index can be assigned to each final knowledge block to identify it and facilitate subsequent searching. This knowledge block index can include a term index and a semantic vector index. The term index can be obtained by segmenting each knowledge block using a word segmentation tool and creating an inverted index for each term. The semantic vector index can be obtained by extracting semantic vectors from each knowledge block using a semantic encoding model. Here, the semantic encoding model can be built based on a large language model.

[0050] Furthermore, keywords of user commands can be extracted, and based on the term index corresponding to each knowledge block, several knowledge blocks in each knowledge block that contain the keywords of user commands most frequently are selected as candidate knowledge blocks. Then, the semantic vector of user commands is extracted, and the similarity between the semantic vector of user commands and the semantic vector index corresponding to each candidate knowledge block is calculated. The candidate knowledge block with the highest similarity is selected from each candidate knowledge block as the target knowledge block that matches the user command.

[0051] This target knowledge block can be used to generate content outlines, presentation slides, or both simultaneously.

[0052] When the target knowledge block is only used to generate a content outline, the user command and the target knowledge block can be input together into the large language model, and the target knowledge block can assist the large language model in generating the content outline.

[0053] When the target knowledge block is only used to generate presentation slides, user instructions, content outlines, and target knowledge blocks can be input into the large language model simultaneously. The target knowledge block then assists the large language model in generating the presentation slides.

[0054] When the target knowledge block is used to generate the content outline and presentation document, the content outline is generated first using user instructions and the target knowledge block, and then the presentation document is generated using user instructions, the content outline, and the target knowledge block.

[0055] In this embodiment of the invention, by introducing a knowledge base and retrieving target knowledge blocks from the knowledge base that match the user's instructions to generate at least one of a content outline and a presentation document, the accuracy of the generated content outline and presentation document can be improved.

[0056] Based on the above embodiments, the user input information includes an initial explanation document; based on the user input information, a presentation explanation document with a paragraph structure is determined, including: Extract the outline of the initial explanation document; Based on the content outline, the initial explanation document is divided into segments to obtain the presentation explanation document.

[0057] Specifically, based on user input including an initial explanatory document, when determining the presentation explanatory document, the user-provided initial explanatory document can first be input into a large language model, which then extracts the content outline of the initial explanatory document. This initial explanatory document's content outline can be the same as the presentation explanatory document's content outline and can be directly used as the presentation explanatory document's content outline.

[0058] Since the initial explanation document may not have a clear paragraph structure, its content outline can be used to segment the document, resulting in segmented segments. For example, the initial explanation document and its content outline can be input into a segmentation model, which will then generate the segmented results. Here, the segmentation model can be built based on a large language model.

[0059] The segmentation results can include multiple paragraphs from the initial explanatory document, as well as the content and presentation page type for each paragraph.

[0060] Furthermore, the initial explanation document, which contains known paragraphs, the content of each paragraph, and the presentation page type, can be used as a presentation explanation document with a paragraph structure.

[0061] In this embodiment of the invention, by extracting the content outline of the initial explanation document and segmenting the initial explanation document using the content outline, the presentation explanation document can be determined. This can give the presentation explanation document a paragraph structure, which helps to generate a target presentation document with a clear structure and logic.

[0062] Based on the above embodiments, the outline of the initial explanatory document is extracted, including: The initial explanatory document is segmented into sentences to obtain multiple statements; Perform semantic analysis on each statement to obtain the semantic analysis results; Based on the semantic analysis results, extract headings of a specified level from the initial explanatory document; Based on each heading, determine the content outline.

[0063] Specifically, in the process of extracting the outline of the initial explanation document, the document can be divided into sentences using the delimiters in the initial explanation document, resulting in multiple sentences. These delimiters can include punctuation marks with longer pauses, such as periods, colons, semicolons, question marks, and exclamation marks. The content between two adjacent delimiters, along with the content of the next delimiter, together form a complete sentence.

[0064] Subsequently, semantic analysis can be performed on each statement to obtain the semantic analysis results. These results can then be used to extract at least one heading of a specified level from the initial explanatory document. The semantic analysis results, i.e., the semantic vectors of each statement, can be input into the heading extraction model to obtain at least one heading of a specified level from the initial explanatory document. This heading extraction model can be constructed using a large language model.

[0065] Finally, by extracting at least one heading at a specified level, a content outline containing only at least one heading at a specified level can be directly constructed.

[0066] In this embodiment of the invention, by performing actions such as sentence segmentation, semantic analysis, and extraction of headings at specified levels from the initial explanatory document to determine the content outline, the precision of the content outline can be guaranteed and the accuracy of the content outline can be improved.

[0067] Based on the above embodiments, the content outline includes the titles corresponding to each presentation page type in the target presentation; Based on the headings, determine the content outline, including: Mark each title according to the corresponding presentation page type to obtain the content outline.

[0068] Specifically, each heading in the content outline corresponds to a page type in the target presentation. In other words, the content outline also includes the tagging results of each heading according to the page type in the target presentation. Therefore, when generating the content outline, it is necessary to tag each heading according to its corresponding page type to obtain a content outline containing each heading and its tagging results.

[0069] In this embodiment of the invention, by introducing the marking results of each title in the content outline according to the corresponding presentation page type, it is easier to distinguish different parts of the content outline, thereby improving the efficiency of generating presentation documents using the content outline.

[0070] Based on the above embodiments, the initial explanatory document is segmented into sentences to obtain multiple statements, which then include: Number each statement. Extract the outline of the initial explanation document, which then includes: Based on the sequence number of each statement, establish the correspondence between each statement and each title in the content outline.

[0071] Specifically, after the initial explanatory document is divided into sentences, each sentence can be numbered, meaning each sentence has a corresponding number.

[0072] The examples following the numbered statements in the initial explanation document can be found in the box below: As can be seen from the example above, the initial explanatory document was divided into 33 sentences.

[0073] Furthermore, after extracting the content outline of the initial explanation document, the correspondence between each statement and each title in the content outline can be established using the statement numbers. For example, each first-level title in the content outline can be labeled with the corresponding starting statement number in the initial explanation document.

[0074] For example, the outline can be updated as follows: In this embodiment of the invention, by establishing the correspondence between each statement and each title in the content outline through the sequence number of each statement in the initial explanation document, accurate explanation content can be provided for the speaker to explain the target presentation after the target presentation is generated.

[0075] Based on the above embodiments, the specified level of headings includes first-level headings, and the presentation page type includes at least one title page; Based on the content outline, the initial explanation document is segmented to obtain the presentation explanation document, including: Based on the content outline, the initial explanation document is segmented to obtain the segmentation results; Based on the starting sentences of each first-level heading and the starting sentences of each heading page, the segmentation results are calibrated to obtain the presentation document.

[0076] Specifically, after numbering each statement in the initial explanation document and marking each title according to the corresponding presentation page type, the segmentation result obtained by segmenting the initial explanation document using the content outline can include multiple paragraphs in the initial explanation document, as well as the statement number and presentation page type corresponding to each paragraph.

[0077] An example of the segmented result could be the content in the box below: In the example above, the 33 statements in the initial explanatory document are divided into 11 paragraphs, each corresponding to at least one statement and a presentation page type.

[0078] When the specified heading level includes first-level headings, and the presentation page type includes at least one title page, the segmentation model can calibrate the segmentation results using the starting statements of each first-level heading and each title page to avoid inaccurate segmentation. This calibrates the segmentation results and yields the presentation explanation document. For example, the starting statements of each title page can be aligned with the starting statements of the first-level headings in the initial explanation document, and the starting statements of each first-level heading can be aligned with the starting statements of the corresponding paragraphs in the initial explanation document. Specifically, it checks whether the starting statements of each title page match the starting statements of the first-level headings in the initial explanation document, and whether the starting statements of each first-level heading match the starting statements of the corresponding paragraphs in the initial explanation document. If all checks are consistent, the segmentation result is considered accurate, and the presentation explanation document is directly determined using the segmentation result. If the checks are inconsistent, the segmentation result is considered inaccurate, and the statement numbers within the relevant paragraphs in the segmentation result need to be adjusted until all checks are consistent. The adjusted segmentation result is then used to determine the presentation explanation document.

[0079] In this embodiment of the invention, by calibrating the segmentation results, the accuracy of the paragraph structure in the presentation document can be ensured, thereby making the content of each presentation page in the final generated target presentation more reasonable.

[0080] Based on the above embodiments, the target presentation is generated page by page based on the presentation document, including: Based on the presentation document, a matching target template is determined from a pre-set template database; Based on the presentation document and the target template, generate the target presentation page by page.

[0081] Specifically, in the process of generating the target presentation page by page, the presentation explanation document and the target template can be combined to generate the target presentation page by page. The target template can be a presentation template that matches the presentation explanation document in a pre-set template database.

[0082] Understandably, a template database may include multiple presentation templates, each with its own template description text. This description text describes the style, visual elements, applicable industries, and application scenarios of the corresponding presentation template. For example, "This is a high-end business-style PPT template. The pages include gold building outlines, smooth lines, and gradient backgrounds. It is suitable for business reports and strategic planning in the financial industry, and project presentations and market analysis in the real estate industry."

[0083] The template description text for each presentation template can be generated through a multimodal large model, or it can be manually annotated and filled in. In this embodiment of the invention, no specific restrictions are imposed on this.

[0084] By matching the presentation explanation document with the template description text of each presentation template, the presentation template with the highest matching degree can be selected as the target template.

[0085] After selecting a target template, the presentation document and the target template can be input into the presentation page generation model. The model then determines and fills in the text boxes and image boxes for each page of the target presentation, according to the format of the target template, thereby generating each page of the target presentation. Here, the presentation page generation model can be constructed based on a large language model.

[0086] In this embodiment of the invention, a template database is introduced, and the target template is automatically matched with the presentation document. This avoids the problem of mismatch between the template layout and the content of the presentation document caused by manually selecting the presentation template, making the style of the target presentation more consistent and the logic stronger.

[0087] Based on the above embodiments, the template database includes multiple presentation templates, each corresponding to template description text and multiple types of template tags; based on the presentation explanation document, a matching target template is determined from the pre-set template database, including: Extract the document description text and primary semantic tags from the presentation's explanation document; Based on the first semantic tag and the template tags corresponding to each presentation template, at least one candidate template is retrieved from the template database. Semantically match the document description text with the template description text of each candidate template to obtain the first semantic matching score, and determine the target template from each candidate template based on the first semantic matching score.

[0088] Specifically, each presentation template can have not only template description text, but also multiple types of template tags. Each template tag can be in JSON format, and the type of each template tag can include, but is not limited to, industry, application scenario, style, applicability, language, color, etc. Each presentation template can have one or more template tags of the same type, without specific limitations. Different types of template tags are shown in Table 1.

[0089] Table 1. Template Tags of Different Types The template tags for each presentation template can be generated through a multimodal large model, or they can be manually labeled and filled in. In this embodiment of the invention, no specific restrictions are imposed on this.

[0090] In the process of determining the matching target template from a pre-set template database, the presentation document can be input into both a semantic tag extraction model and a feature description generation model. The semantic tag extraction model extracts the first semantic tag from the presentation document, while the feature description generation model extracts the document description text from the presentation document. The first semantic tag can include one or more tags. Both the semantic tag extraction model and the feature description generation model can be constructed using a large language model.

[0091] Subsequently, using the first semantic tag and the template tags corresponding to each presentation template, at least one candidate template with the same template tag is retrieved from the template database. It is understandable that the template database contains a massive number of presentation templates, and each presentation template corresponds to multiple template tags; therefore, there exists at least one presentation template with the same template tag.

[0092] In this embodiment of the invention, the first semantic tag can be used as a retrieval field to perform an Elasticsearch search on the template tags corresponding to each presentation template in order to determine at least one presentation template as a candidate template.

[0093] Next, the document description text is semantically matched with the template description text of each candidate template to obtain a first semantic matching score. For example, the document description text and the template description text of each candidate template can be input into the semantic matching model. The semantic matching model performs semantic matching between the input document description text and the template description text of the candidate templates to obtain the first semantic matching score output by the semantic matching model between the document description text and the template description text of the candidate templates. Here, the first semantic matching score can represent the semantic matching degree between the document description text and the corresponding template description text of the candidate template; the higher the first semantic matching score, the higher the semantic matching degree.

[0094] Finally, the target template is determined from the candidate templates by using the first semantic matching score corresponding to the template description text of each candidate template. For example, the candidate template with the highest first semantic matching score can be directly selected as the target template.

[0095] In this embodiment of the invention, the target template is finally determined by dual retrieval of the first semantic tag and the document description text, which can improve the matching degree between the target template and the presentation document.

[0096] Based on the above embodiments, each presentation template has a first retrieval weight for different types of template tags; Based on the first semantic tag and the template tags corresponding to each presentation template, at least one candidate template is retrieved from the template database, including: Calculate multiple first-degree associations between the first semantic tag and the different types of template tags corresponding to each presentation template; Based on the first retrieval weight of the template tags corresponding to each presentation template, the first relevance is weighted and summed to obtain the first weighted result; Based on the first weighted result corresponding to each presentation template, candidate templates are determined from each presentation template.

[0097] Specifically, to effectively distinguish between presentation templates, different first retrieval weights can be assigned to different types of template tags corresponding to each presentation template. Furthermore, when retrieving at least one candidate template from the template database, multiple first correlation degrees can be calculated between the first semantic tag and the different types of template tags corresponding to each presentation template. That is, each template tag corresponding to each presentation template has a corresponding first correlation degree, which can be the cosine similarity between the first semantic tag and the template tag.

[0098] Subsequently, by utilizing the first retrieval weight of the template tags corresponding to each presentation template, the first relevance is weighted and summed to obtain the first weighted result. Each presentation template corresponds to a first weighted result, which represents the comprehensive relevance between the corresponding presentation template and the first semantic tag.

[0099] Using the first weighted result corresponding to each presentation template, candidate templates are determined from all presentation templates. For example, a specified number of presentation templates with the highest first weighted result can be selected as candidate templates. The specified number can be set as needed and is not specifically limited here. For example, the specified number can be set to 15, or other values.

[0100] In this embodiment of the invention, by introducing a first retrieval weight for each template tag, the retrieved candidate templates can be made to have a high degree of matching with the presentation document.

[0101] Based on the above embodiments, the presentation page types in the target presentation include a cover page; the target presentation is generated page by page based on the presentation explanation document and the target template, including: Based on the paragraph content corresponding to the cover page, the presentation explanation document, and the cover page template in the target template, generate the presentation page corresponding to the cover page.

[0102] Specifically, for the cover page in the target presentation, the paragraph content corresponding to the cover page, the presentation explanation document, and the cover page template in the target template can be input into the large language model, and the large language model can generate the presentation page corresponding to the cover page.

[0103] Here, the paragraph content corresponding to the cover page refers to the sentences contained in the paragraph corresponding to the cover page, that is, the content marked as the opening remarks in the presentation document.

[0104] The cover page template in the target template can include text boxes, their meanings and quantities, and may also include the maximum number of lines and characters for each text box.

[0105] As an example, the relevant information for the presentation page corresponding to the cover page can be the content in the box below: In this embodiment of the invention, some of the content required to generate the presentation page corresponding to the cover page is provided, which can reduce the difficulty of generating the presentation page corresponding to the cover page.

[0106] Based on the above embodiments, the presentation page types in the target presentation include a table of contents page; the target presentation is generated page by page based on the presentation documentation and the target template, including: Based on the paragraph content corresponding to the table of contents, the presentation explanation document, the content outline of the presentation explanation document, and the table of contents template in the target template, generate the presentation page corresponding to the table of contents.

[0107] Specifically, for the table of contents page in the target presentation, the paragraph content corresponding to the table of contents page, the presentation explanation document, the content outline of the presentation explanation document, and the table of contents page template in the target template can be input into the large language model, and the large language model can generate the presentation page corresponding to the table of contents page.

[0108] The paragraph content corresponding to the table of contents refers to the statements contained in the paragraph corresponding to the table of contents, that is, the content marked as the table of contents description in the presentation document.

[0109] The table of contents template in the target template may include text boxes, their meanings and quantities, and may also include the maximum number of lines and characters for each text box.

[0110] As an example, the relevant information for the presentation page corresponding to the table of contents page can be the content in the following box: In this embodiment of the invention, some of the content required to generate the presentation page corresponding to the table of contents is provided, which can reduce the difficulty of generating the presentation page corresponding to the table of contents.

[0111] Based on the above embodiments, the presentation page types in the target presentation include at least one title page; the target presentation is generated page by page based on the presentation explanation document and the target template, including: For the current title page in each title page, the presentation page corresponding to the current title page is generated based on the paragraph content corresponding to the current title page, the page content of each presentation page preceding the current title page, the first-level headings corresponding to each presentation page preceding the current title page, and the title page template in the target template.

[0112] Specifically, for each title page in the target presentation, its corresponding presentation page can be determined through the same operation. Taking the current title page as an example, the paragraph content corresponding to the current title page, the page content of each presentation page preceding the current title page, the first-level headings corresponding to each presentation page preceding the current title page, and the title page template in the target template can be input into the large language model, and the large language model can then generate the presentation page corresponding to the current title page.

[0113] The target template's title page template can include text boxes with their meanings and quantities, as well as the maximum number of lines and characters per text box. If the target template contains multiple title page template styles, all title page templates must use the same style to ensure style consistency. If the title page template style includes chapter number boxes, the number boxes should be filled in according to chapter order.

[0114] As an example, the relevant information for the presentation page corresponding to the current title page can be the content in the box below: In this embodiment of the invention, some of the content required to generate the presentation pages corresponding to each title page is provided, which can reduce the difficulty of generating the presentation pages corresponding to the title pages.

[0115] Based on the above embodiments, the presentation page types in the target presentation include at least one detail page; the target presentation is generated page by page based on the presentation explanation document and the target template, including: For the current detail page in each detail page, the presentation page corresponding to the current detail page is generated based on the paragraph content corresponding to the current detail page, the page content of each presentation page preceding the current detail page, the first-level headings corresponding to each presentation page preceding the current detail page, and the detail page template in the target template.

[0116] Specifically, for each detail page in the target presentation, its corresponding presentation page can be determined through the same operation. Taking the current detail page as an example, the paragraph content corresponding to the current detail page, the page content of each presentation page preceding the current detail page, the first-level headings corresponding to each presentation page preceding the current detail page, and the detail page template in the target template can be input into the large language model, and the large language model can then generate the presentation page corresponding to the current detail page.

[0117] The details page template in the target template may include text boxes, their meanings and quantities, as well as the maximum number of lines and characters for each text box.

[0118] As an example, the relevant information for the presentation page corresponding to the current details page can be the content in the following box: In this embodiment of the invention, some of the content required to generate the presentation pages corresponding to each detail page is provided, which can reduce the difficulty of generating the presentation pages corresponding to the detail pages.

[0119] Based on the above embodiments, the target template includes multiple types of detail page templates, each with a corresponding type identifier, which includes one or more identifier elements; based on the paragraph content corresponding to the current detail page, the page content of each presentation page preceding the current detail page, the first-level headings corresponding to each presentation page preceding the current detail page, and the detail page templates in the target template, the presentation page corresponding to the current detail page is generated, including: Perform semantic analysis on the paragraph content corresponding to the current details page to determine the identifier element information in the paragraph content corresponding to the current details page; Based on the identifier element information and the type identifier corresponding to each detail page template, determine the detail page template of the target type from each detail page template; Based on the paragraph content corresponding to the current details page, the presentation pages preceding the current details page, the first-level headings corresponding to the presentation pages preceding the current details page, and the details page template of the target type, generate the presentation page corresponding to the current details page.

[0120] Specifically, the target template may include multiple types of detail page templates, and each detail page template may have a corresponding type identifier, which is used to characterize the type of the corresponding detail page template. The type identifier includes one or more identifier elements. Identifier elements may be at least one of the following: the number of keywords in the detail page template, text tags, the number of different types of text tags, and combinations of keywords and text tags. Among these, text tags are used to characterize the attribute features of text boxes in the detail page template, such as title boxes, content boxes, long text boxes, and short text boxes.

[0121] Furthermore, when generating the presentation page corresponding to the current details page, semantic analysis can be performed on the paragraph content corresponding to the current details page to determine the identifier element information in the paragraph content, that is, to determine the number of keywords and at least one of the text tags in the paragraph content.

[0122] Furthermore, the identifier elements in the paragraph content corresponding to the current details page can be matched with the type identifiers corresponding to each details page template. The successfully matched details page template is then selected as the details page template for the target type. The type identifier for this target type can be the number of keywords, text tags, or a combination of the number of keywords and text tags.

[0123] After that, the paragraph content corresponding to the current details page, the presentation pages preceding the current details page, the first-level headings corresponding to the presentation pages preceding the current details page, and the details page template of the target type can be input into the large language model, and the large language model can generate the presentation page corresponding to the current details page.

[0124] As an example, the relevant information on the presentation page corresponding to the current details page can be the content in the box below: In this embodiment of the invention, a process for determining the detail page template of the target type is provided, which can improve the matching degree between the detail page template of the target type and the corresponding paragraph content of the detail page in the presentation document.

[0125] Based on the above embodiments, the target type detail page template includes a first type of text box and a second type of text box, wherein the maximum character count of the first type of text box is greater than the maximum character count of the second type of text box; based on the paragraph content corresponding to the current detail page, the presentation pages preceding the current detail page, the first-level headings corresponding to the presentation pages preceding the current detail page, and the target type detail page template, the presentation page corresponding to the current detail page is generated, which also includes: Perform semantic analysis on the paragraph content corresponding to the current details page to identify multiple text contents within the paragraph content corresponding to the current details page; For any text content, if the number of characters in any text content is greater than the maximum number of characters in the second type of text box, then the text content is simplified to obtain the simplified result; If the simplified result has fewer than or equal to the maximum number of characters in the second type of text box, then the second type of text box is filled based on any text content; If the number of characters in the simplified result exceeds the character limit of the second type of text box, then the first type of text box will be filled based on any text content.

[0126] Specifically, the detail page template for the target type can include a first type of text box and a second type of text box, with the first type of text box having a greater character limit than the second type of text box. For example, the first type of text box is a long text box, and the second type of text box is a short text box. The character limit of a text box refers to the maximum total number of characters that the entire text box can hold.

[0127] For presentation pages, short text is more concise and clear than long text, so short text boxes can be used preferentially in presentation pages.

[0128] Based on this, when generating the presentation page corresponding to the current details page, semantic analysis can be performed on the paragraph content corresponding to the current details page to determine multiple text contents in the paragraph content corresponding to the current details page, and each text content can correspond to a keyword.

[0129] For any text content, if the number of characters in the text content exceeds the character limit of the second type of text box, the text content can be simplified using the simplification model to obtain a simplified result. Here, the simplification model can be constructed using a large language model, and the simplified result can include one or more texts, each of which is a short text.

[0130] If the simplified result has a character count less than or equal to the character count limit of the second type of text box, the simplified result is considered to meet the character count requirement of the second type of text box, and can therefore be used to fill the second type of text box. Each text in the simplified result can be considered a line in the second type of text box.

[0131] If the simplified result has more characters than the maximum number of characters in the second type of text box, it is considered that the simplified result does not meet the character count requirement of the second type of text box. In this case, only any text content can be used to fill the first type of text box.

[0132] In this embodiment of the invention, when generating presentation pages corresponding to each detail page, the second type of text box, namely short text box, is used first to make the presentation text page as concise and clear as possible.

[0133] Based on the above embodiments, the presentation page type in the target presentation also includes an end page; generating the target presentation page by page based on the presentation explanation document and the target template also includes: Based on the paragraph content corresponding to the ending page, the page content and paragraph content of the presentation page corresponding to the cover page, and the ending page template in the target template, generate the presentation page corresponding to the ending page.

[0134] Specifically, for the last page in the target presentation, the paragraph content corresponding to the last page, the page content and paragraph content of the presentation page corresponding to the cover page, and the last page template in the target template can be input into the large language model, and the presentation page corresponding to the last page can be generated through the large language model.

[0135] Here, the paragraph content corresponding to the last page refers to the sentences contained in the paragraph corresponding to the last page, that is, the content marked as the conclusion in the presentation document.

[0136] The end page template in the target template can include text boxes, their meanings and quantities, and may also include the maximum number of lines and characters for each text box.

[0137] As an example, the relevant information for the presentation page corresponding to the final page could be the content shown in the box below: In this embodiment of the invention, some of the content required to generate the presentation page corresponding to the ending page is provided, which can reduce the difficulty of generating the presentation page corresponding to the ending page.

[0138] Based on the above embodiments, the target presentation is generated page by page based on the presentation document, including: If the current page of the target presentation contains an image frame, then the matching target image is determined from the pre-set image database based on the page content of the presentation page corresponding to the previous page of the current page, and the paragraph content of the current page and the next page in the presentation explanation document. Fill the image frame based on the target image.

[0139] Specifically, the target presentation may need to insert images, i.e., image frames may exist. If an image frame exists on the current page of the target presentation, the page content of the previous page, the corresponding paragraph content of the current page and the next page in the presentation documentation, can be input into the large language model. The large language model then determines the matching target image from a pre-set image database.

[0140] It is understandable that an image database can include multiple presentation images, and different presentation images can correspond to different content.

[0141] Furthermore, by using the target image, the image frames on the current page can be filled, ensuring that the current page contains images and guaranteeing the diversity of the content displayed on the current page.

[0142] Based on the above embodiments, the image database includes multiple presentation images, each corresponding to an image description text and multiple types of image tags; based on the page content of the previous page of the current page and the paragraph content of the current page and the next page in the presentation document, a matching target image is determined from the pre-set image database, including: Based on the page content of the presentation page corresponding to the previous page of the current page, and the paragraph content of the current page and the next page in the presentation document, the feature description text and the second semantic tag are determined. Based on the second semantic tag and the image tags corresponding to each presentation image, at least one candidate image is retrieved from the image database. The feature description text is semantically matched with the image description text of each candidate image to obtain a second semantic matching score. Based on the second semantic matching score, the target image is determined from each candidate image.

[0143] Specifically, each presentation image in the image database can correspond to an image description text and multiple types of image tags. The image description text is used to describe the style, content, and applicable industries of the corresponding presentation image. For example, "This is a realistic image that mainly depicts a woman wearing a beige cotton apron arranging a bunch of herbs on a granite countertop. It is suitable for cooking and food presentations, health education, and the life services industry."

[0144] Image tags can be in JSON format, and their types can include, but are not limited to, industry, style, aspect ratio, and main color scheme. Each presentation image can have one or more image tags of the same type; no specific limit is imposed here. Different types of image tags are shown in Table 2.

[0145] Table 2. Image Tags for Different Types The image description text and image tags for each presentation template can be generated by a multimodal large model, or they can be manually labeled and filled in. In this embodiment of the invention, no specific restrictions are imposed on this.

[0146] In the process of determining the target image from the candidate images, the page content of the previous page of the presentation document and the paragraph content of the current page and the next page in the presentation document can be input into the semantic tag extraction model and the feature description generation model. The semantic tag extraction model extracts the second semantic tag, and the feature description generation model extracts the feature description text. The second semantic tag can include one or more.

[0147] Subsequently, using the second semantic tag and the image tags corresponding to each presentation image, at least one candidate image with the same image tag is retrieved from the image database. It is understandable that the image database contains a massive number of presentation images, and each presentation image corresponds to multiple image tags; therefore, at least one presentation image with the same image tag exists.

[0148] In this embodiment of the invention, the second semantic tag can be used as a retrieval field to perform an Elasticsearch search on the image tags corresponding to each presentation image, so as to determine at least one presentation image as a candidate image.

[0149] Next, the feature description text is semantically matched with the image description text of each candidate image to obtain a second semantic matching score. For example, the feature description text and the image description text of each candidate image can be input together into the semantic matching model. The semantic matching model performs semantic matching between the input feature description text and the image description text of the candidate images to obtain a second semantic matching score between the feature description text and the image description text of the candidate images output by the semantic matching model. Here, the second semantic matching score can be considered as the semantic matching degree between the feature description text and the corresponding image description text of the candidate images; the higher the second semantic matching score, the higher the semantic matching degree.

[0150] Finally, the target image is determined from the candidate images using the second semantic matching score corresponding to the image description text of each candidate image. For example, the candidate image with the highest second semantic matching score can be directly selected as the target image.

[0151] In this embodiment of the invention, the target image is finally determined through dual retrieval using the second semantic tag and the feature description text, which can improve the matching degree between the target image and the presentation document.

[0152] Based on the above embodiments, each presentation image has a second search weight corresponding to different types of image tags; Based on the second semantic tag and the image tags corresponding to each presentation image, at least one candidate image is retrieved from the image database, including: Calculate multiple second-degree associations between the second semantic tags and the different types of image tags corresponding to each presentation image; Based on the second retrieval weight of the image tags corresponding to each presentation image, the second relevance is weighted and summed to obtain the second weighted result. Based on the second weighted result corresponding to each presentation image, candidate images are determined from each presentation image.

[0153] Specifically, to effectively distinguish between presentation images, different second retrieval weights can be assigned to the different types of image tags corresponding to each presentation image. Furthermore, when retrieving at least one candidate image from the image database, multiple second correlation degrees can be calculated between the second semantic tag and the different types of image tags corresponding to each presentation image. That is, each image tag corresponding to each presentation image has a corresponding second correlation degree, which can be the cosine similarity between the second semantic tag and the image tag.

[0154] Subsequently, by utilizing the second retrieval weights of the image tags corresponding to each presentation image, the second relevance is weighted and summed to obtain the second weighted result. Each presentation image corresponds to a second weighted result, which represents the comprehensive relevance between the corresponding presentation image and the second semantic tag.

[0155] Using the second weighted result corresponding to each presentation image, candidate images are determined from all presentation images. For example, a preset number of presentation images with the highest second weighted result can be selected as candidate images. The preset number can be set as needed and is not specifically limited here. For example, the preset number can be set to 30, or other values.

[0156] In this embodiment of the invention, by introducing a second retrieval weight for each image tag, the retrieved candidate images can have a high degree of matching with the presentation document.

[0157] Based on the above embodiments, the image tags corresponding to each presentation image all include the image's main color tone; based on each second semantic matching score, the target image is determined from each candidate image, including: Based on the color scheme of the target template and the main color scheme of each candidate image, color matching rules are applied to adjust the second semantic matching scores to obtain the adjustment results. Based on the adjustment results, the target image is determined from the candidate images.

[0158] Specifically, the image tags for each presentation image can include the image's primary color tone. Therefore, when determining the target image, the color scheme of the target template and the primary color tone of each candidate image can be used to apply color matching rules and adjust the secondary semantic matching scores to obtain the adjusted results. For example, for candidate images whose primary color tone matches the target template's color scheme, their secondary semantic matching scores can be increased; for candidate images whose primary color tone clashes with the target template's color scheme, their secondary semantic matching scores can be decreased.

[0159] The color matching rules are shown in Table 3.

[0160] Table 3 Color Scheme Matching Rules Finally, the candidate image with the highest score among the adjusted images can be selected as the target image.

[0161] In this embodiment of the invention, by introducing color matching rules, the images in each presentation page can be better matched with the color scheme of the target template, thereby improving the sensory effect of the target presentation.

[0162] Based on the above embodiments, the target presentation is generated page by page based on the presentation document, and then the following is also included: For any statement in the text content of the target text box of the target presentation, if the number of characters in any statement exceeds the single-line character limit of the target text box, then the statement is segmented into multiple segmentation units. Bind punctuation marks in any sentence to adjacent word segmentation units, and determine the line break position of any sentence based on the maximum number of characters per line; At the newline position, a newline is created for any statement.

[0163] Specifically, after generating the target presentation page by page, each page of the target presentation can be optimized, such as the layout and style of each page. For each sentence in the text content of the target text box in the target presentation, if the number of characters in the sentence exceeds the single-line character limit of the target text box, the sentence can be segmented into multiple segmentation units. Each segmentation unit can be a character, a word, or a phrase, etc.

[0164] Furthermore, punctuation marks in the sentence can be bound to adjacent word segments; for example, left quotation marks and left parentheses can be bound to the following words, while other punctuation marks can be bound to the preceding words.

[0165] Then, using the maximum number of characters per line, the line break position of the statement is determined, and the statement is broken into new lines at the line break position.

[0166] In this embodiment of the invention, line breaks are achieved by segmenting sentences into words and binding punctuation marks with adjacent segmentation units. This improves the clarity of the text structure, makes it easier to understand and read, and enables the standardized expression of sentences.

[0167] Based on the above embodiments, such as Figure 2 As shown, this embodiment of the invention also provides a method for generating digital human videos, including: S21, Obtain the target presentation document; the target presentation document is generated based on the presentation document generation method provided in the above embodiments; S22, determine the presentation explanation document with paragraph structure corresponding to the target presentation, and generate a digital human video based on the presentation explanation document.

[0168] Specifically, the digital human video generation method provided in this embodiment of the invention is executed by a digital human video generation device, which can be configured in a computer. The computer can be a local computer or a cloud computer. The local computer can be a computer, tablet, etc., and no specific limitation is made here.

[0169] First, step S21 is executed, which generates a target presentation document based on the user input information using the presentation document generation methods provided in the above embodiments.

[0170] Then, step S22 is executed to determine the presentation explanation document with paragraph structure corresponding to the target presentation. This presentation explanation document can be determined when the target presentation is generated in step S21. Furthermore, the content of the presentation explanation document is used as the digital human's presentation content, and a digital human video is generated through text-driven video generation.

[0171] The digital human video generation method provided in this embodiment of the invention first obtains a target presentation, then determines the corresponding presentation explanation document with paragraph structure, and generates a digital human video based on the presentation explanation document. This method can not only display a high-quality target presentation in the digital human video, but also provide a speech script for the digital human's presentation through the presentation explanation document, making the digital human's speaking actions in the digital human video more natural and fluid.

[0172] Based on the above embodiments, a digital human video is generated based on the presentation document, including: Convert presentation slides into audio; Generate digital human videos based on audio, or audio and presentation documentation.

[0173] Specifically, in the process of generating digital human videos, speech synthesis technology can be used to convert presentation documents into audio, which can then be used as the audio for the digital human's speech.

[0174] Then, using audio, or audio combined with presentation slides, voice-animation synthesis technology is employed to generate digital human animation data. This digital human animation data is used to drive the digital human to perform different actions and expressions.

[0175] When using audio and presentation slides to generate digital human animation data, since both audio and digital human animation data are generated based on the presentation slides and have a temporal sequence, the paragraphs in the presentation slides correspond to the audio segments and animation segments, respectively. That is, there is a correspondence between different presentation pages in the target presentation and the audio segments and animation segments. This allows the digital human to give a speech using different presentation pages in the target presentation, thus obtaining a complete digital human video.

[0176] In addition, the audio can be equipped with multilingual translation, enabling the digital human to explain in different languages, breaking down geographical and linguistic barriers and allowing audiences worldwide to understand the content.

[0177] like Figure 3 As shown, based on the above embodiments, this embodiment of the invention provides a presentation generation device, including: Information acquisition module 31 is used to acquire user input information; The presentation document determination module 32 is used to determine the presentation document with paragraph structure based on user input information; each paragraph in the presentation document corresponds to a presentation page; The presentation generation module 33 is used to generate the target presentation page by page based on the presentation explanation document.

[0178] Specifically, the functions of each module in the presentation generation device provided in this embodiment of the invention correspond one-to-one with the operation flow of each step in the above method-like embodiments, and the achieved effects are also the same. For details, please refer to the above embodiments, and this will not be repeated in this embodiment of the invention.

[0179] like Figure 4 As shown, based on the above embodiments, this embodiment of the invention provides a digital human video generation device, comprising: Document acquisition module 41 is used to acquire a target presentation document; the target presentation document is generated based on the presentation document generation method provided in the above embodiments; The digital human video generation module 42 is used to determine the presentation document with paragraph structure corresponding to the target presentation and generate a digital human video based on the presentation document.

[0180] Specifically, the functions of each module in the digital human video generation device provided in this embodiment correspond one-to-one with the operation flow of each step in the above method-like embodiments, and the achieved effects are also the same. For details, please refer to the above embodiments, and this will not be repeated in this embodiment.

[0181] Figure 5An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute the presentation generation method or the digital human video generation method provided in the above embodiments.

[0182] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to related technologies, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0183] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the presentation generation method or the digital human video generation method provided in the above embodiments.

[0184] In another aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the presentation generation method or the digital human video generation method provided in the above embodiments. This computer-readable storage medium can be either a non-transitory computer-readable storage medium or a transient computer-readable storage medium, and no specific limitation is made herein.

[0185] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0186] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0187] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating presentation slides, characterized in that, include: Obtain user input information; Based on the user input information, a presentation document with paragraph structure is determined; Each paragraph in the presentation document corresponds to one presentation page; Based on the presentation document, the target presentation is generated page by page.

2. The presentation generation method according to claim 1, characterized in that, The user input information includes user commands; the step of determining a presentation document with a paragraph structure based on the user input information includes: Based on the user instructions, a content outline is generated; Based on the user instructions and the content outline, the presentation document is generated.

3. The presentation generation method according to claim 2, characterized in that, The user input information also includes a knowledge base; The process of generating the content outline based on the user instruction includes, prior to: The knowledge base is divided into knowledge blocks to obtain multiple knowledge blocks and the knowledge block index corresponding to each knowledge block. Based on the indexes of each knowledge block, determine the target knowledge block that matches the user instruction; The process of generating the content outline based on the user instruction includes: The content outline is generated based on the user instructions and the target knowledge block; And / or, generating the presentation document based on the user instructions and the content outline includes: Based on the user instructions, the content outline, and the target knowledge blocks, the presentation document is generated.

4. The presentation generation method according to claim 1, characterized in that, The user input information includes an initial explanation document; the step of determining a presentation explanation document with a paragraph structure based on the user input information includes: Extract the outline of the initial explanation document; Based on the content outline, the initial explanation document is segmented to obtain the presentation explanation document.

5. The presentation generation method according to claim 4, characterized in that, The extraction of the content outline of the initial explanation document includes: The initial explanatory document is segmented into sentences to obtain multiple statements; Perform semantic analysis on each of the statements to obtain the semantic analysis results; Based on the semantic analysis results, extract headings of a specified level from the initial explanatory document; Based on the titles mentioned, the outline of the content is determined.

6. The presentation generation method according to claim 5, characterized in that, The content outline includes the titles corresponding to each presentation page type in the target presentation; The process of determining the content outline based on each of the titles includes: Each title is marked according to its corresponding presentation page type to obtain the content outline.

7. The presentation generation method according to claim 6, characterized in that, The initial explanatory document is segmented into sentences to obtain multiple statements, which then include: Number each of the statements. The step of extracting the content outline of the initial explanation document then includes: Based on the sequence number of each statement, establish the correspondence between each statement and each title in the content outline.

8. The presentation generation method according to claim 7, characterized in that, The specified level of headings includes first-level headings, and the presentation page type includes at least one title page; Based on the content outline, the initial explanation document is segmented to obtain the presentation explanation document, including: Based on the outline, the initial explanation document is segmented to obtain the segmentation results; Based on the starting statements of each of the first-level headings and the starting statements of each of the title pages, the segmentation results are calibrated to obtain the presentation document.

9. The presentation generation method according to any one of claims 1-8, characterized in that, The process of generating the target presentation page by page based on the presentation document includes: Based on the presentation document, a matching target template is determined from a pre-set template database; Based on the presentation document and the target template, the target presentation is generated page by page.

10. The presentation generation method according to claim 9, characterized in that, The template database includes multiple presentation templates, each of which corresponds to template description text and multiple types of template tags; The step of determining a matching target template from a pre-set template database based on the presentation document includes: Extract the document description text and the first semantic tag from the presentation document; Based on the first semantic tag and the template tag corresponding to each of the presentation templates, at least one candidate template is retrieved from the template database; The document description text is semantically matched with the template description text of each candidate template to obtain a first semantic matching score, and the target template is determined from each candidate template based on the first semantic matching score.

11. The presentation generation method according to claim 10, characterized in that, Each of the aforementioned presentation templates has a first retrieval weight corresponding to different types of template tags; The step of retrieving at least one candidate template from the template database based on the first semantic tag and the template tags corresponding to each of the presentation templates includes: Calculate multiple first correlation degrees between the first semantic tag and the template tags of different types corresponding to each of the presentation templates; Based on the first retrieval weight of the template tag corresponding to each of the presentation templates, the first relevance is weighted and summed to obtain a first weighted result; Based on the first weighted result corresponding to each of the presentation templates, each candidate template is determined from each of the presentation templates.

12. The presentation generation method according to claim 9, characterized in that, The presentation page types in the target presentation include a cover page; the process of generating the target presentation page by page based on the presentation explanation document and the target template includes: Based on the paragraph content corresponding to the cover page, the presentation document, and the cover page template in the target template, the presentation page corresponding to the cover page is generated.

13. The presentation generation method according to claim 9, characterized in that, The presentation page types in the target presentation include table of contents pages; the process of generating the target presentation page by page based on the presentation explanation document and the target template includes: Based on the paragraph content corresponding to the directory page, the presentation explanation document, the content outline of the presentation explanation document, and the directory page template in the target template, the presentation page corresponding to the directory page is generated.

14. The presentation generation method according to claim 9, characterized in that, The presentation page type in the target presentation includes at least one title page; the step of generating the target presentation page by page based on the presentation explanation document and the target template includes: For the current title page in each of the title pages, the presentation page corresponding to the current title page is generated based on the paragraph content corresponding to the current title page, the page content of each of the presentation pages preceding the current title page, the first-level headings corresponding to each of the presentation pages preceding the current title page, and the title page template in the target template.

15. The presentation generation method according to claim 9, characterized in that, The presentation page types in the target presentation include at least one detail page; The step of generating the target presentation page by page based on the presentation document and the target template includes: For the current detail page in each of the aforementioned detail pages, the presentation page corresponding to the current detail page is generated based on the paragraph content corresponding to the current detail page, the page content of each of the aforementioned presentation pages preceding the current detail page, the first-level headings corresponding to each of the aforementioned presentation pages preceding the current detail page, and the detail page template in the target template.

16. The presentation generation method according to claim 15, characterized in that, The target template includes multiple types of detail page templates, and each detail page template has a corresponding type identifier, which includes one or more identifier elements. The process of generating the presentation page corresponding to the current details page based on the paragraph content corresponding to the current details page, the page content of each presentation page preceding the current details page, the first-level headings corresponding to each presentation page preceding the current details page, and the details page template in the target template includes: Perform semantic analysis on the paragraph content corresponding to the current details page to determine the identifier element information in the paragraph content corresponding to the current details page; Based on the identification element information and the type identifier corresponding to each of the detail page templates, a detail page template of the target type is determined from each of the detail page templates; Based on the paragraph content corresponding to the current details page, the presentation pages preceding the current details page, the first-level headings corresponding to the presentation pages preceding the current details page, and the details page template of the target type, the presentation page corresponding to the current details page is generated.

17. The presentation generation method according to claim 16, characterized in that, The target type of detail page template includes a first type of text box and a second type of text box, wherein the maximum number of characters in the first type of text box is greater than the maximum number of characters in the second type of text box; The step of generating the presentation page corresponding to the current details page based on the paragraph content corresponding to the current details page, the presentation pages preceding the current details page, the first-level headings corresponding to the presentation pages preceding the current details page, and the details page template of the target type, further includes: Perform semantic analysis on the paragraph content corresponding to the current details page to determine multiple text contents in the paragraph content corresponding to the current details page; For any text content, if the number of characters in the text content is greater than the maximum number of characters in the second type of text box, then the text content is simplified to obtain a simplified result; If the number of characters in the simplified result is less than or equal to the maximum number of characters in the second type of text box, then the second type of text box is filled based on the simplified result; If the number of characters in the simplified result is greater than the upper limit of the number of characters in the second type of text box, then the first type of text box is filled based on any of the text contents.

18. The presentation generation method according to claim 12, characterized in that, The presentation page types in the target presentation also include a final page; the step of generating the target presentation page by page based on the presentation explanation document and the target template also includes: Based on the paragraph content corresponding to the ending page, the page content and paragraph content of the presentation page corresponding to the cover page, and the ending page template in the target template, the presentation page corresponding to the ending page is generated.

19. The presentation generation method according to claim 9, characterized in that, The process of generating the target presentation page by page based on the presentation document includes: If the current page of the target presentation contains an image frame, then a matching target image is determined from a pre-set image database based on the page content of the presentation page corresponding to the previous page of the current page and the paragraph content of the current page and the next page in the presentation explanation document. The image frame is filled based on the target image.

20. The presentation generation method according to claim 19, characterized in that, The image database includes multiple presentation images, each of which corresponds to image description text and multiple types of image tags; the process of determining matching target images from a pre-set image database based on the page content of the previous page of the presentation and the paragraph content of the current page and the next page in the presentation document includes: Based on the page content of the presentation page corresponding to the previous page of the current page, and the paragraph content of the current page and the next page in the presentation document, the feature description text and the second semantic tag are determined. Based on the second semantic tag and the image tags corresponding to each of the presentation images, at least one candidate image is retrieved from the image database. The feature description text is semantically matched with the image description text of each candidate image to obtain a second semantic matching score, and the target image is determined from each candidate image based on the second semantic matching score.

21. The presentation generation method according to claim 20, characterized in that, Each of the presentation images has a second search weight corresponding to a different type of image tag. The step of retrieving at least one candidate image from the image database based on the second semantic tag and the image tags corresponding to each of the presentation images includes: Calculate multiple second correlation degrees between the second semantic tag and the image tags of different types corresponding to each of the presentation images; Based on the second retrieval weight of the image tags corresponding to each of the presentation images, the second relevance is weighted and summed to obtain a second weighted result; Based on the second weighted result corresponding to each of the presentation images, each candidate image is determined from each of the presentation images.

22. The presentation generation method according to claim 20, characterized in that, The image tags corresponding to each of the presentation images all include the image's main color tone; determining the target image from each of the candidate images based on the second semantic matching score includes: Based on the color scheme of the target template and the main color tone of each candidate image, color matching rules are applied to adjust the second semantic matching score to obtain the adjustment result; Based on the adjustment results, the target image is determined from each of the candidate images.

23. The presentation generation method according to any one of claims 1-8, characterized in that, The process of generating the target presentation page by page based on the presentation document also includes: For any statement in the text content of the target text box of the target presentation, if the number of characters in any statement exceeds the single-line character limit of the target text box, then the statement is segmented into multiple segmentation units. The punctuation marks in any of the statements are bound to the adjacent word segmentation units, and the line break position of any of the statements is determined based on the single-line character limit. At the given line break position, perform a line break on any of the statements.

24. A method for generating digital human videos, characterized in that, include: Obtain the target presentation; The target presentation is generated based on the presentation generation method as described in any one of claims 1-23; Identify the presentation explanation document with paragraph structure corresponding to the target presentation, and generate a digital human video based on the presentation explanation document.

25. The digital human video generation method according to claim 24, characterized in that, The process of generating a digital human video based on the presentation document includes: Convert the presentation document into audio; The digital human video is generated based on the audio, or the audio and the presentation document.

26. A presentation generation device, characterized in that, include: The information acquisition module is used to acquire user input information; The presentation document determination module is used to determine a presentation document with a paragraph structure based on the user input information. Each paragraph in the presentation document corresponds to one presentation page; The presentation generation module is used to generate the target presentation page by page based on the presentation explanation document.

27. A digital human video generation device, characterized in that, include: The document acquisition module is used to acquire the target presentation document; The target presentation is generated based on the presentation generation method as described in any one of claims 1-23; The digital human video generation module is used to determine the presentation document with paragraph structure corresponding to the target presentation document, and generate a digital human video based on the presentation document.

28. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the presentation generation method as described in any one of claims 1-23, or the digital human video generation method as described in any one of claims 24-25.

29. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the presentation generation method as described in any one of claims 1-23, or the digital human video generation method as described in any one of claims 24-25.

30. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the presentation generation method as described in any one of claims 1-23, or the digital human video generation method as described in any one of claims 24-25.

Images