Advertising video generation methods, apparatus and computer program products

By using speech recognition and text segmentation technology, matching video clips are automatically selected from a video clip library to generate advertising videos, solving the problem of time-consuming manual editing and improving generation efficiency and matching accuracy.

CN123123638APending Publication Date: 2026-11-17TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510632869.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-15
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In the process of generating advertising videos, manual video editing is time-consuming, resulting in low efficiency in generating advertising videos.

Method used

The advertising text material is segmented by speech recognition, and the video clips are selected from the video clip library based on the segmentation tags and durations to generate advertising video material and synthesize the target video.

Benefits of technology

It reduces the time spent acquiring materials, improves the efficiency of generating advertising videos, ensures the matching degree between video visuals and audio, and avoids manual editing steps.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123638A_ABST
    Figure CN123123638A_ABST
Patent Text Reader

Abstract

This application provides an advertising video generation method, apparatus, and computer program product. The method includes: acquiring advertising text material for an advertising business; converting the advertising text material into advertising audio content; performing speech recognition on the advertising audio content to obtain speech-recognized text; further segmenting the speech-recognized text to obtain text segments; determining the audio segments of the text segments; determining segmentation tags and segmentation durations of the audio segments based on the text segments; filtering video segments from a video segment library based on the segmentation tags and segmentation durations; further determining advertising text material based on the video segments; and finally determining the target video for the advertising business based on the advertising audio content and advertising text material. Using this application embodiment can improve the efficiency of advertising video generation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of multimedia technology, and in particular to an advertising video generation method, apparatus, and computer program product. Background Technology

[0002] Currently, in advertising scenarios targeting advertising businesses, advertising personnel can use video editing software to edit a large number of videos, and then manually select advertising video materials that they believe are suitable for the current advertising scenario from the large number of edited videos. In this way, they can generate advertising videos for advertising placement using the manually edited advertising video materials.

[0003] However, the inventors discovered in practice that the generation of advertising videos requires manual video editing. Based on human experience, they need to select advertising video materials that are suitable for the current advertising business from a large number of edited videos, which consumes a long time to acquire materials. Therefore, how to improve the generation efficiency of advertising videos for specific advertising businesses in advertising scenarios is a technical problem that urgently needs to be solved. Summary of the Invention

[0004] This application provides an advertising video generation method, apparatus, and computer program product, which can improve the efficiency of advertising video generation.

[0005] One embodiment of this application provides a method for generating advertising videos, including:

[0006] The process involves acquiring advertising text materials for advertising business, converting the content of the advertising text materials into advertising voice content, performing speech recognition on the advertising voice content, and obtaining the speech recognition text of the advertising voice content.

[0007] The speech recognition text is segmented to obtain text segments. In the advertising speech content, the corresponding speech segments are identified. The segmentation labels of the speech segments are determined by the content of the text segments, and the segmentation duration of the speech segments is determined.

[0008] Based on segment segmentation tags and segment segmentation duration, video segments that match the advertising business are selected from multiple video segments contained in the video segment library;

[0009] Based on the selected video clips, the advertising video materials for the advertising business are determined, and based on the advertising video materials and advertising audio content, targeted delivery videos for the advertising business are generated.

[0010] One embodiment of this application provides an advertising video generation apparatus, including:

[0011] The speech recognition module is used to acquire advertising text materials for advertising business. When converting the material content in the advertising text materials into advertising speech content, it performs speech recognition on the advertising speech content to obtain the speech recognition text of the advertising speech content.

[0012] The text segmentation module is used to segment the speech recognition text to obtain text segmentation segments. In the advertising speech content, it determines the speech segmentation segment corresponding to the text segmentation segment, and determines the segmentation label of the speech segmentation segment based on the segmentation content of the text segmentation segment, and determines the segmentation duration of the speech segmentation segment.

[0013] The video clip filtering module is used to filter video clips that match the advertising business from multiple video clips in the video clip library based on clip segmentation tags and clip segmentation duration.

[0014] The video delivery generation module is used to determine the advertising video materials for the advertising business based on the selected video clips, and to generate targeted delivery videos for the advertising business based on the advertising video materials and advertising audio content.

[0015] Among them, multiple video clips include segments composed of video footage obtained from target historical delivery videos; target historical delivery videos refer to videos that meet advertising delivery indicators selected from historical delivery videos delivered on advertising delivery platforms related to advertising business.

[0016] The speech recognition module includes:

[0017] The business model acquisition unit is used to acquire a business processing model for advertising business; the business processing model involves multiple tasks, including a first processing task that matches the first type of input data and a second processing task that matches the second type of input data;

[0018] The input data acquisition unit is used to acquire the target input data for inputting the business processing model;

[0019] The first task processing unit is used to obtain the business prompt template and business reference information of the advertising business from the first type of input data if the target input data belongs to the first type of input data. When executing the first processing task through the business processing model, it obtains reference data content from the business reference information according to the template prompt words in the business prompt template, and generates advertising text material for the advertising business through the reference data content.

[0020] The second task processing unit is used to obtain historical advertising text from the target's historical video delivery if the target input data belongs to the second type of input data. When executing the second processing task through the business processing model, the historical advertising text is subjected to text fission processing to obtain fission processed text, which is then used as advertising text material for the advertising business. Fission processing refers to replacing the target text words with related text words when generating associated text words related to the target text words in the historical advertising text.

[0021] Among them, the historical advertising text is the sample advertising text that has been pre-added to the sample storage database; the number of sample advertising texts in the sample storage database is at least N; N is a positive integer;

[0022] The speech recognition module also includes:

[0023] The sample text acquisition unit is used to acquire N sample advertisement texts for training the initial business model from the sample storage database, and select the target sample advertisement text from the N sample advertisement texts;

[0024] The predictive text generation unit is used to determine the target predictive advertising text based on the target sample advertising text and the initial business model.

[0025] The predictive text acquisition unit is used to obtain the target predictive ad text for each sample ad text when each of the N sample ad texts is selected as the target sample ad text.

[0026] The model fine-tuning processing unit is used to fine-tune the initial business model using each sample ad text and the target predicted ad text of each sample ad text, to obtain the business processing model.

[0027] Among them, the predictive text generation unit is also used to perform text content analysis on the target sample advertisement text to obtain sample advertisement demand information for the sample advertisement text, and use the sample advertisement demand information as sample advertisement business.

[0028] The predictive text generation unit is also used to obtain the template engineering construction function for the sample advertising business, input the sample advertising requirement information into the template engineering construction function, and the template engineering construction function constructs the sample business prompt template for the sample advertising requirement according to the template construction function instructions in the template engineering construction function.

[0029] The predictive text generation unit is also used to input the sample business prompt template into the initial business model. When the initial business model performs the first processing task, it predicts and outputs the target predictive advertising text that matches the template prompt words in the sample prompt template.

[0030] Among them, the model fine-tuning processing unit is also used to calculate the model loss of the initial business model by using each sample ad text and the target predicted ad text of each sample ad text;

[0031] The model fine-tuning processing unit is also used to adjust the model parameters of the initial business model if the model loss does not meet the model convergence condition, and use the initial business model with adjusted model parameters as a transitional business model.

[0032] The model fine-tuning processing unit is also used to output the predicted advertising text of each sample advertising text in N sample advertising texts through the transition business model;

[0033] The model fine-tuning processing unit is also used to iteratively train the transition business model using each sample advertisement text and the predicted advertisement text of each sample advertisement text, until the model loss of the iteratively trained transition business model meets the model convergence condition, and then the transition business model that meets the model convergence condition is determined as the business processing model.

[0034] The advertising business includes target businesses for the target product; the business prompt template includes an emotional guidance section, a middle section, and an ending guidance section; the middle section is derived based on the product attributes of the target product.

[0035] The first task processing unit is also used for the emotion guidance part based on the business prompt template, to obtain the emotion guidance text information in the business reference information, and to generate emotion guidance text materials for the target business through the emotion guidance text information; the emotion guidance text information includes historical emotion guidance text obtained based on historical advertising text.

[0036] The first task processing unit is also used to obtain product attribute information about the target product from the business reference information based on the middle segment of the business prompt template, and generate product introduction text material for the target business based on the product attribute information;

[0037] The first task processing unit is also used to obtain the ending guidance text information from the business reference information based on the ending guidance section of the business prompt template, and generate ending guidance text materials for the target business based on the ending guidance text information; the ending guidance text information includes historical ending guidance text obtained based on historical advertising text; the target product includes the target game product, and the ending guidance text materials include download guidance text materials for the target game product;

[0038] The first task processing unit is also used to obtain advertising text materials for advertising business based on emotion-guided text materials, product introduction text materials, and concluding guiding text materials.

[0039] Among them, the speech recognition text carries text segmentation identifiers for text division, and the text segmentation time point of the text segmentation identifiers is determined when performing speech recognition on the advertising speech content;

[0040] The text segmentation module includes:

[0041] The text segment confirmation unit is used to identify the text segmentation markers in the speech recognition text, and to segment the speech recognition text based on the identified text segmentation markers to obtain text segmentation segments carrying text segmentation markers.

[0042] The voice segment confirmation unit is used to identify the voice segment corresponding to the text segment in the advertising voice content, identify the key text content from the segment content of the text segment, and identify the segment segmentation label of the voice segment through the key text content.

[0043] The segment duration confirmation unit is used to obtain the text segmentation time point of the text segmentation identifier in the text segmentation segment, and determine the segmentation duration of the speech segmentation segment based on the text segmentation time point.

[0044] The video clips stored in the video clip library include multiple first-class candidate video clips. A first-class candidate video clip refers to a video clip carrying a video clip tag. A video clip tag refers to a video tag obtained by performing content analysis on the content of a first-class candidate video clip.

[0045] The video clip filtering module includes:

[0046] The video tag lookup unit is used to find video segment tags that match the segment division tags in the video segment library, and to take the first type of candidate video segment corresponding to the found video segment tag as the target first type of candidate video segment.

[0047] The target duration confirmation unit is used to determine the duration of the first type of candidate video segment as the target video segment duration.

[0048] The duration difference determination unit is used to determine the duration difference between the duration of the target video segment and the segment division duration, obtain the target duration difference that meets the duration difference range condition from the duration difference, and take the target first-class candidate video segment corresponding to the target duration difference as the video segment that matches the advertising business.

[0049] The video clip filtering module also includes:

[0050] The general segment acquisition unit is used to acquire the second type of candidate video segment corresponding to the general video tag in the video segment library if no video segment tag matching the segment division tag is found in the video segment library.

[0051] The video segment confirmation unit is used to identify video segments that match the advertising business from the second category of candidate video segments.

[0052] When a speech segment contains multiple sub-speech segments, the advertising video material includes multiple video materials corresponding to the multiple sub-speech segments, with one sub-speech segment corresponding to one video material;

[0053] The video generation module includes:

[0054] The target segment confirmation unit is used to select a target sub-speech segment from multiple sub-speech segments and determine the video material corresponding to the target sub-speech segment as the target video material.

[0055] The target material optimization unit is used to optimize the target video material based on the segment division duration of the target sub-speech segments, resulting in optimized video material for the target sub-speech segments; the playback duration of the optimized video material is consistent with the segment division duration of the target sub-speech segments.

[0056] The optimized material acquisition unit is used to obtain optimized video material for each sub-speech segment when each sub-speech segment in multiple sub-speech segments is selected as the target sub-speech segment.

[0057] The optimized material splicing unit is used to determine the material splicing order of the optimized video material for each sub-speech segment based on the segment segmentation order corresponding to multiple sub-speech segments. The optimized video material for each sub-speech segment is spliced ​​according to the material splicing order to obtain the spliced ​​video.

[0058] The voice merging processing unit is used to merge the voice of the video material with the advertising voice content to obtain the voice-merged video, and then use the voice-synthesized video as the target video for advertising business.

[0059] The video generation module also includes:

[0060] The video text addition unit is used to add the speech recognition text as video text to the speech synthesis video, and to update the speech synthesis video by adding the video text to the speech synthesis video.

[0061] The video generation module also includes:

[0062] The key audio acquisition unit is used to acquire audio files related to advertising business and identify key audio segments in the audio files;

[0063] The key audio addition unit is used to add key audio segments as background audio to the speech-synthesized video.

[0064] The audio in the audio file is music audio, which contains multiple audio segments;

[0065] The key audio acquisition unit is also used to acquire the target audio segment from multiple audio segments, and to identify the audio segments other than the target audio segment from the multiple audio segments as the audio segments to be matched;

[0066] The key audio acquisition unit is also used to determine the segment similarity between the target audio segment and the audio segment to be matched;

[0067] The key audio acquisition unit is also used to determine the target audio segment that is similar to the target audio segment from the audio segments to be matched by segment similarity, and to count the number of similar segments of the target audio segment;

[0068] The key audio acquisition unit is also used to identify the target audio segment as the key audio segment if the number of similar segments meets the key audio determination conditions.

[0069] Among them, the video clip library contains multiple video clips including historical video clips of target historical delivery videos; target historical delivery videos refer to videos that have met the advertising delivery indicators selected from historical delivery videos delivered on advertising delivery platforms related to the advertising business.

[0070] The device also includes:

[0071] The video compliance detection module is used to perform compliance detection on each video frame in the target's historical video feed and obtain the compliance detection results.

[0072] The video frame filtering module is used to filter out illegal video frames from the target historical video feed if the compliance detection result indicates that there are illegal video frames in each video frame, so as to obtain the target historical filtered video.

[0073] The video cleaning module is used to clean the subtitle information in the video frame of the target historical filtered video, and the target historical filtered video with cleaned subtitle information is used as the target historical cleaned video.

[0074] The historical video segmentation module is used to identify the video scene of the target historical cleaning video, segment the target historical cleaning video based on the identified video scene, obtain video segment segments, and add the video segment segments as historical video segments to the video segment library.

[0075] Among them, the video clip library contains multiple video clips including historical video clips of target historical delivery videos; target historical delivery videos refer to videos that have met the advertising delivery indicators selected from historical delivery videos delivered on advertising delivery platforms related to the advertising business.

[0076] The device also includes:

[0077] The target video delivery module is used to deliver target videos to the advertising platform and to count the amount of interaction data of the target video on the advertising platform within the video delivery duration.

[0078] The historical video update module is used to update the target historical ad video by using the target ad video and update the historical video segment by using the video segment of the target ad video if the video ad duration reaches the video ad duration threshold and the amount of interaction data counted within the video ad duration reaches the ad ad ad metrics.

[0079] The device also includes:

[0080] The video material removal module is used to remove advertising video materials from the video clip library if the video delivery duration reaches the video delivery duration threshold and the amount of interaction data counted within the video delivery duration does not reach the advertising delivery target.

[0081] One aspect of this application provides a computer device, including a memory and a processor. The memory is connected to the processor, the memory is used to store computer programs, and the processor is used to call the computer programs so that the computer device executes the method provided in one aspect of this application.

[0082] One aspect of this application provides a computer-readable storage medium storing a computer program adapted to be loaded and executed by a processor, so that a computer device having a processor performs the method provided in one aspect of this application.

[0083] According to one aspect of this application, a computer program product is provided, comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the method provided in the above aspect.

[0084] This application embodiment can acquire advertising text materials for advertising business and convert the content of the advertising text materials into advertising audio content. It is understood that the advertising text materials are generated based on the business objectives and requirements of the advertising business, and can be manually written or generated by a large language model; the specific generation method of the advertising text materials is not limited here. After obtaining the advertising audio content, speech recognition can be performed on the advertising audio content to obtain speech-recognized text. After obtaining the speech-recognized text, text segmentation can be performed on the speech-recognized text to obtain text segments. By segmenting the speech-recognized text for advertising audio content into text segments, the corresponding audio segments can be identified within the advertising audio content. In other words, each text segment corresponds to an audio segment, and each audio segment is an audio segment present in the advertising audio content. Simultaneously with segmenting the speech-recognized text into text segments, the advertising audio content is also segmented into audio segments, and the text segments and audio segments are in one-to-one correspondence. After identifying the corresponding speech segments to the obtained text segments, the content of the text segments can be analyzed to determine the segmentation labels and durations of the speech segments corresponding to those text segments. It's understood that these segmentation labels are based on the content of the text segments, thus summarizing the speech content of the corresponding speech segments. The text content in the speech recognition text obtained from the advertising speech content carries timestamps. This text content includes text segmentation identifiers that divide the speech recognition text into segments, and the duration of each speech segment can be determined using the timestamps (or the time points of the text segmentation identifiers). After determining the segmentation labels and durations for the speech segments, these labels and durations can be used to search and filter relevant video clips from a video clip library related to the advertising business to find and filter videos that match the advertising business. Here, when searching for video segments in the video segment library using the segment segmentation tags and segment segmentation durations based on speech segmentation, it is ensured that not only do the segment content of the video segment found in the segment library match the segment content of the speech segmentation, but also that the segment duration of the video segment found in the video segment library can be as close as possible to the segment segmentation duration of the speech segmentation, thereby greatly improving the matching degree between the video image and the video speech of the advertising video generated by this application.It is understood that the audio segments here are audio segments obtained from the advertising audio content, so there may be multiple segments. After selecting the corresponding video segments from the video segment library for each audio segment, the advertising video material corresponding to the advertising business can be determined based on the obtained video segments. By combining the obtained advertising video material and advertising audio content, a target delivery video for the advertising business can be generated. This target delivery video can be understood as the generated advertising video that can be placed on the advertising platform. It is understood that the embodiments of this application can convert the advertising text material for the advertising business into advertising audio content, then select matching video segments from the video segment library for each audio segment obtained based on the advertising audio content, and obtain advertising video material for the advertising business based on the selected video segments. Therefore, the process of obtaining advertising video material does not involve manual selection, thereby reducing the material acquisition time. Finally, the target delivery video can be directly generated by combining the video material and advertising audio content without the need for manual video editing, which greatly improves the efficiency of advertising video generation. Attached Figure Description

[0085] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0086] Figure 1 This is a schematic diagram of a network architecture provided in an embodiment of this application;

[0087] Figure 2 This is a schematic diagram of data interaction provided in an embodiment of this application;

[0088] Figure 3 This is a schematic diagram of an advertising video generation method provided in an embodiment of this application;

[0089] Figure 4 This is a schematic diagram illustrating a scenario for generating advertising text materials according to an embodiment of this application;

[0090] Figure 5 This is a schematic diagram of a scenario for training a business processing model provided in an embodiment of this application;

[0091] Figure 6 This is a schematic diagram illustrating a scenario where a business processing model using a business prompt template is provided in an embodiment of this application;

[0092] Figure 7This is a schematic diagram of a scenario for segmenting advertising voice content provided in an embodiment of this application;

[0093] Figure 8 This is a schematic diagram of a scenario for segmenting and filtering video segments based on speech data, provided in an embodiment of this application.

[0094] Figure 9 This is a schematic diagram of a scenario for generating audio-visually merged video according to an embodiment of this application;

[0095] Figure 10 This is a schematic diagram of another advertising video generation method provided in an embodiment of this application;

[0096] Figure 11 This is a schematic diagram of a scenario for determining key audio segments provided in an embodiment of this application;

[0097] Figure 12 This is a schematic diagram of a scenario for acquiring historical video clips provided in an embodiment of this application;

[0098] Figure 13 This is a schematic diagram of a scene for generating advertising video materials according to an embodiment of this application;

[0099] Figure 14 This is a schematic diagram of the structure of an advertising video generation device provided in an embodiment of this application;

[0100] Figure 15 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0101] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0102] Please see Figure 1 , Figure 1 This is a schematic diagram of a network architecture provided in an embodiment of this application. Figure 1 As shown, this network architecture may include a server 10d and a user terminal cluster. The user terminal cluster may include one or more user terminals; the number of user terminals is not limited here. Figure 1As shown, the user terminal cluster can specifically include user terminal 10a, user terminal 10b, and user terminal 10c. Server 10d can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. User terminals 10a, 10b, and 10c can all include: smartphones, tablets, laptops, PDAs, mobile internet devices (MIDs), wearable devices (such as smartwatches and smart bracelets), smart voice interaction devices, smart home appliances (such as smart TVs), and in-vehicle devices, as well as other electronic devices with video / image playback capabilities. Figure 1 As shown, user terminals 10a, 10b, and 10c can each connect to server 10d via a network, so that each user terminal can interact with server 10d through the network connection.

[0103] In this embodiment, the targeted video for advertising can be generated on user terminals (e.g., user terminals 10a, 10b, and 10c). For ease of understanding, user terminal 10a is used as an example. Specifically, after obtaining the advertising text material for the advertising business, user terminal 10a can convert the content of the advertising text material into advertising audio content. It is understood that the advertising text material is generated based on the business objectives and requirements of the advertising business, and can be manually written or generated by a large language model. The specific generation method of the advertising text material is not limited here. Furthermore, user terminal 10a can perform speech recognition on the obtained advertising audio content to obtain the corresponding speech recognition text. It is understood that the speech recognition text contains text segmentation markers, and user terminal 10a can segment the obtained speech recognition text based on these text segmentation markers. The text segmentation markers can be predefined specific text symbols (e.g., spaces) or punctuation marks. The specific form of the text segmentation markers is not limited here. After user terminal 10a performs text segmentation on the speech recognition text to obtain text segments, it can identify the corresponding speech segments within the advertising speech content. Furthermore, by analyzing the content of the text segments, it determines the segmentation label of the corresponding speech segment. In other words, the segmentation label of the speech segment is obtained by analyzing the content of the text segments of the speech segment. It can be understood that the text segmentation identifiers in the speech recognition text obtained from the advertising speech content carry corresponding text segmentation time points. User terminal 10a can determine the segmentation duration of the obtained speech segment based on the text segmentation time points carried in the text segment identifiers. Further, user terminal 10a can use the obtained segmentation labels and segmentation durations of the speech segments to filter corresponding video segments from a video segment library. This video segment library can be understood as a database storing video segments related to the advertising business. User terminal 10a can determine the advertising video material corresponding to the advertising business based on the obtained video clips, and then generate a target delivery video for the advertising business by combining the obtained advertising video material and advertising voice content. Here, the target delivery video can be understood as the generated advertising video that can be placed on the advertising platform for the advertising business.The advertising platform here can be a platform (e.g., a social platform) provided by a server (e.g., server 10d) that can be used to place advertising videos. This platform can be displayed on the client. When users use the platform, they can see the advertisements placed on the platform. Therefore, after the target video is generated through the user terminal 10a, the target video can be placed on the advertising platform provided by server 10d.

[0104] It is understood that in this embodiment, the user terminal can convert advertising text materials for advertising services into advertising voice content, and then select matching video segments from a video segment library for each voice segment obtained based on the advertising voice content. Based on the selected video segments, advertising video materials for the advertising service are obtained. It is understood that the target delivery video finally generated by the user terminal is synthesized from the advertising voice content and advertising video materials. Therefore, by selecting video segments from the video segment library for each voice segment, the video image of the final target delivery video can accurately match the video voice. Furthermore, since this embodiment generates advertising video materials by directly recalling video segments from the video library through voice segmentation, rather than manually thinking about and creating video materials matching the advertising voice content, the efficiency of advertising video generation is greatly improved.

[0105] It is understood that in the specific implementation of this application, business data of users, enterprises, institutions and other objects may be involved. When the above embodiments of this application are applied to specific products or technologies, permission or consent from users, enterprises, institutions and other objects is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant regions.

[0106] For further details, please see Figure 2 , Figure 2 This is a schematic diagram of data interaction provided in an embodiment of this application. For example... Figure 2 As shown, after obtaining the advertising text material (text material 22a) for advertising business in this embodiment of the application, the text material 22a can first be converted into speech to obtain advertising speech content (speech content 23a). Then, speech recognition is performed on the speech content 23a to obtain the corresponding speech recognition text. It can be understood that the speech recognition text can be segmented to further divide the speech content 23a into speech segments. Specifically, the speech recognition text can be segmented using text segmentation markers present in the speech recognition text to obtain text segments, and then the speech segments in the speech content 23a can be determined, such as... Figure 2As shown, the audio content 23a can be divided into multiple audio segments, including audio segments 20a, 20b, 20c, 20d, 20e, and 20f. It is understood that each of these audio segments includes a corresponding segmentation label and segmentation duration. These labels and durations are determined based on the analysis of the content of the corresponding text segments for each audio segment. Specifically, the resulting text segments include keywords (or key text content) that can summarize the main content of the text segment. The segmentation labels can then be determined using these key text content. These labels are tag words that summarize the key text content. Therefore, by using these segmentation labels, video segments matching the audio segment content can be retrieved from the video segment library. Understandably, after segmenting the speech-recognition text using text segmentation markers, the segment duration of each segment can be determined by the text segmentation time point carried by the text segmentation markers. Furthermore, by using the segmentation tag and duration of each speech segment, matching video segments can be retrieved from a video segment library for each segment. This video segment library stores multiple video segments related to advertising, which can be manually created, provided by advertisers, or derived from historical video submissions, etc. The specific source of the video segments in the library is not limited here.

[0107] It's understandable that retrieving video clips for advertising audio segments is done semantically. Specifically, each video clip carries a video clip tag that indicates its content; in other words, the video clip tag is obtained based on the video clip's content. When searching for video clips in the video clip library, clips whose video clip tags match the segment segmentation tags of the advertising audio clip are filtered out. This ensures that the content of the finally filtered video clips matches the content of the advertising audio clip. Since multiple video clips may match the advertising audio clip, the video clip whose duration is closest to the segment segmentation duration of the advertising audio clip is selected as the final target video clip. This method ensures that the content of the found target video clip matches the advertising audio clip as closely as possible, resulting in a higher degree of matching between the video visuals and the audio in the final generated advertising video. Through the above filtering process, it is possible to... Figure 2The audio segments in the video clip library are used to retrieve corresponding video clips. For example, audio segment 20a corresponds to video clip 21b, audio segment 20b corresponds to video clip 21e, audio segment 20c corresponds to video clip 21b, audio segment 20d corresponds to video clip 21a, audio segment 20e corresponds to video clip 21d, and audio segment 20f corresponds to video clip 21c. It is understood that the video clips retrieved from audio segments 20a and 20c are the same, both being video clip 21b. This is because in this embodiment, there may be a situation where the same video clip matches multiple advertising audio segments. In this case, the video clip will be used as the video clip retrieved from each of these multiple advertising audio segments. It is understood that this video clip will also appear multiple times in the final generated target delivery video (finished video).

[0108] After retrieving video segments for multiple ad voice clips, the corresponding video segments can be optimized based on the duration of each voice clip. Specifically, this could involve cropping or multiplying the video segments to ensure that the duration of each retrieved video segment is equal to the duration of its corresponding ad voice clip. Figure 2 Taking audio segment 20a as an example, after selecting a video segment 21b that matches audio segment 20a, video segment 21b can be optimized so that its duration is the same as that of audio segment 20a. Similarly, after optimizing the video segment corresponding to each audio segment, the selected video segments can be evaluated according to their order within the audio content 23a. For example... Figure 2 As shown, the selected video clips can be spliced ​​together in the order of video clip 21b, video clip 21e, video clip 21b, video clip 21a, video clip 21d, video clip 21c… to form a spliced ​​video. This spliced ​​video is then merged with the audio content 23a to obtain a merged audio video. After obtaining the merged audio video, it can be used as the final target video (finished product video) and deployed on an online advertising platform. The online platform can update its video clip library for advertising based on the performance metrics of this finished product video; specifically, it can remove video clips corresponding to substandard finished products from the video clip library.

[0109] It is understood that the embodiments of this application can convert the obtained advertising text material into advertising audio content, and then divide the advertising audio content into audio segments through speech recognition. Based on the content of the obtained audio segments, matching video segments are retrieved from a video segment library, and these video segments are further spliced ​​together to obtain video material. This method can quickly generate video material that matches the advertising audio content, improving the speed of advertising video production and ensuring that the final advertising video's visuals and audio are accurately matched. Furthermore, by placing the obtained target video on an online advertising platform and updating the video segments in the video segment library based on the target video's performance metrics, subsequent finished videos generated for the same advertising business can better meet the needs of the advertising platform, thereby further improving the production quality of advertising videos.

[0110] For further details, please see Figure 3 , Figure 3 This is a schematic diagram illustrating an advertising video generation method provided in an embodiment of this application. It can be understood that... Figure 3 The method for generating advertising videos shown can be executed by a computer device, which can be a user terminal or a server. For ease of understanding, this example uses a computer device as a user terminal. This user terminal can be, for example, a... Figure 1 The user terminal 10a is shown. (As shown in the image) Figure 3 As shown, the advertising video generation method may include at least the following: Figure 3 Steps S101 to S104 in the process.

[0111] Step S101: Obtain advertising text material for advertising business. When converting the material content in the advertising text material into advertising voice content, perform speech recognition on the advertising voice content to obtain the speech recognition text of the advertising voice content.

[0112] In this embodiment, when generating a targeted video for advertising, the relevant advertising text material is first obtained. This advertising text material can be understood as the script used in the final targeted video. It is understood that there are multiple ways to obtain this advertising text material. It can be obtained by inputting a relevant business prompt template for advertising into the business processing model, which then outputs the corresponding advertising text material based on the template. Alternatively, existing advertising text can be input into the business processing model, which performs text fission processing on the existing text and uses the resulting fission-processed text as the advertising text material for the advertising business. It is understood that the business model here can be a Large Language Model (LLM) used to understand and generate human language text. A LLM is a deep learning model, typically based on a neural network architecture, and pre-trained on a large amount of text data to acquire rich language knowledge and reasoning capabilities. It is widely used in natural language processing tasks such as text classification, sentiment analysis, question answering systems, and machine translation.

[0113] Specifically, when generating advertising text materials, the system first obtains the business processing model for the advertising business and inputs the target input data into the model so that it can output the corresponding advertising text materials. It can be understood that this business processing model can handle multiple tasks, including the first processing task of outputting advertising text materials through a business prompt template, and the second processing task of performing fission processing on historical advertising text to obtain fission-processed text. Therefore, the business processing model will judge the received target input data. If the target input data is the first type of input data containing a business prompt template and business reference information, it will call the relevant modules for the first processing task to perform the first processing task on the target input data. It can be understood that the business prompt template contains template prompts for the text structure of the advertising text materials used to prompt the business processing model, while the business reference information is information related to the advertising business.

[0114] Taking advertising for a specific game product as an example, the business reference information could include product selling points, trending topics, target users, etc., while the business prompt template could include multiple parts, such as emotional guidance, product selling point introduction, download guidance, etc. Therefore, when the business processing model receives the first type of input data and performs the first processing task, it can retrieve the corresponding reference data content from the business reference information based on the template prompt words in the business prompt template, and generate advertising text materials for the advertising business corresponding to the text structure indicated by the business prompt template using the retrieved reference data content. Taking advertising for a specific game product as an example, the template prompts in the business prompt template can suggest that the final advertising text generated by the business processing model includes a section introducing the product's selling points. The business processing model can then obtain relevant product selling point information from the received business reference information and generate the product selling point introduction section of the advertising text based on this information. Similarly, the business processing model can generate other parts of the advertising text based on the other parts indicated in the template prompts in the business prompt template, by obtaining the reference data from the business reference information, ultimately resulting in the advertising text for that specific game product.

[0115] When the business processing model receives the target input data and determines that the target input data is the second type of input data containing historical advertising text, the second processing task can be performed using the historical advertising text. It is important to note that the historical advertising text here is text data obtained through speech recognition of the target historical video. The target historical video refers to the video that meets the advertising performance indicators selected from historical videos delivered on advertising platforms related to the advertising business. In other words, when obtaining the second type of input data, this embodiment first searches for historically delivered videos that meet the advertising performance indicators on advertising platforms related to the currently processed advertising business, performs speech recognition on the speech content of the historical video, converts it into historical advertising text, and finally uses the obtained historical advertising text as the second type of input data. It can be understood that "historical video meeting the advertising performance indicators" means that after the historical advertising video was delivered to the advertising platform, the historical advertising video met predetermined requirements as an important indicator for measuring the performance of the historical advertising video. For example, if the click-through rate of the historical advertising video reaches a predetermined value (e.g., 5%), then the historical advertising video can be said to have met the advertising performance indicators. Click-through rate (CTR) here refers to the ratio of the number of times a user clicks on an ad or link to the total number of impressions. CTR is one of the important metrics for measuring advertising effectiveness and user interest. In the field of digital marketing and advertising, CTR is often used to evaluate ad creatives, targeting strategies, and campaign performance in order to optimize advertising strategies and improve ROI.

[0116] Furthermore, it can be understood that the second processing task here is to perform text fission processing on the historical advertising text in the second type of input data to obtain fission-processed text. After obtaining the fission-processed text through the business processing model, the resulting fission-processed text can be used as advertising text material for the current advertising business. It can be understood that the text fission processing here refers to replacing the target text words with related text words when generating associated text words that are related to the target text words in the advertising text contained in the second type of input data. In other words, the business processing model will describe some content in the historical advertising text using a different descriptive method, such as generating some new words (related text words) to replace the original words (target text words) in the historical advertising text. Therefore, after performing text fission on historical advertising text through the business processing model, multiple fission texts can be obtained. The number of fission texts generated by the business processing model in the second type of input data can be specified, so that all the generated fission texts can be used as advertising text materials for the current advertising business. It can be understood that multiple advertising text materials can simultaneously generate multiple target videos. Therefore, this application embodiment has the processing capability to simultaneously generate multiple target videos. This situation will not be elaborated further here; our subsequent processing of advertising text materials will also take an advertising text material as a single advertising text as an example. It is understood that in addition to performing text fission processing on historical advertising text, the business processing model can also perform fission processing on other existing advertising texts, including manually written advertising text. The processing methods for these other existing advertising texts are similar to the processing process of the business processing model for historical advertising text, and will not be elaborated further here.

[0117] For further details, please see Figure 4 , Figure 4 This is a schematic diagram illustrating a scenario for generating advertising text materials according to an embodiment of this application. For example... Figure 4 As shown, after receiving input data 30a and determining that it is the first type of input data, the business processing model can call task processing module 31a to perform the first processing task. Specifically, when the business processing model calls task processing module 31a to perform the first processing task, it first obtains the business prompt template from the input data 30a, and then, based on the prompt words in the template, retrieves the corresponding reference data content from the business reference information contained in the input data 30a. Finally, it generates advertising text material 32a based on the retrieved reference data content. Regarding the business processing model performing the second processing task, as follows... Figure 4As shown, input data 30b at this point is the second type of input data, containing historical advertising text. This historical advertising text can be understood as historical videos that have met the advertising targets, selected from online advertising platforms (i.e., advertising platforms related to the advertising business), and obtained through speech recognition of the audio content of these historical videos. After obtaining input data 30b and determining that it is the second type of input data, the business processing model can call task processing module 31b to perform the second processing task. Specifically, when performing the second processing task, the business processing model will perform text fission processing on the historical advertising text in input data 30b, and finally use the resulting fission-processed text as advertising text material 32b.

[0118] It is important to note that the business processing model here can be trained and fine-tuned using historical advertising text during use, making the advertising text materials generated by the model more consistent with the characteristics of historical advertising text. Understandably, this historical advertising text is obtained through speech recognition of the audio content of target historical videos that have achieved the advertising delivery metrics on the advertising platform. Therefore, training and fine-tuning the business processing model using historical advertising text can improve the quality of the advertising text materials obtained by the model. Correspondingly, using the advertising text materials output by the trained business processing model to create target delivery videos can improve the video quality of the target delivery videos and increase the probability of the target delivery videos achieving the advertising delivery metrics on the advertising platform. Specifically, in this embodiment, a certain number of advertising videos for advertising business are created and delivered to the advertising platform. At this point, multiple historical advertising videos exist on the advertising platform. The target historical advertising videos that have achieved the delivery metrics can be selected from these multiple historical advertising videos, and speech recognition can be performed on the audio content of these target historical advertising videos to obtain historical advertising text. This historical advertising text is then added as sample advertising text to the sample storage database.

[0119] Furthermore, when training the business processing model, N sample advertising texts can be obtained from the sample storage database for training. For ease of understanding, the business processing model that has not yet been trained is referred to as the initial business model. It's understood that N here is a positive integer. In actual training using sample advertising texts, multiple sample advertising texts can be selected to train the initial business model simultaneously. These N sample advertising texts can also be referred to as a batch, and the initial business model is trained in batches. After obtaining the N sample advertising texts, each of these N sample advertising texts can be processed to obtain the predicted advertising text. Regarding the acquisition of the predicted advertising text corresponding to these N sample advertising texts, we can take the target sample advertising text as an example. Here, the target advertising text is a text selected from the N sample advertising texts. After selecting the target advertising text from the N sample advertising texts, the initial business model can output the corresponding target predicted advertising text based on the target advertising text. Specifically, text content analysis can be performed on the target sample advertisement text to obtain sample advertisement demand information contained within it. This sample advertisement demand information can be understood as the business objectives and requirements corresponding to the target advertisement text. Therefore, the obtained sample advertisement demand information can also be regarded as the sample advertisement business, and the target advertisement text can be called the advertisement text material for this sample advertisement business. Here, the target sample advertisement text is taken as the advertisement text material for the business targeting the target product. In this case, the sample advertisement demand information obtained from the text content analysis of the target sample advertisement text may include product information of the target product, target user group information of the target product, etc.

[0120] After obtaining the sample advertising requirement information for the target sample advertising text, this information can be input into the template engineering construction function for the sample advertising business. The template engineering construction function can then use corresponding template construction function instructions to construct a sample business prompt template based on this sample advertising requirement information. It's understandable that this method of constructing the sample business prompt template is used to train the business processing model's ability to output advertising text materials as the first processing task using the business prompt template. The generation of the business prompt template is achieved by inputting the relevant advertising requirement information of the advertising business into the template engineering construction function for the advertising business. This business prompt template is then input into the business processing model, causing the model to output the corresponding advertising text materials. The process of generating advertising text materials can be summarized by the following formula:

[0121] prompt = engineering(advertising demand information) (1)

[0122] script = LLM(Prompt) (2)

[0123] Specifically, regarding formula (1), where `engineering` is the template engineering construction function, when generating a corresponding business prompt template for a sample advertising business or the current advertising business, the sample requirement information is input into `engineering`, so that `engineering` can output the corresponding prompt. Here, the prompt is the generated business prompt template for the advertising business corresponding to the sample requirement information. Further, after obtaining the business prompt template, it can be input into the business processing model (LLM), so that the LLM outputs the corresponding advertising text material script. It is understood that the method used in this application embodiment to construct the corresponding business prompt template through the template engineering construction function employs prompt engineering technology. Here, prompt engineering refers to the process of designing and optimizing input prompts to guide a large language model (LLM) to generate specific responses. Effective prompts can significantly improve the performance and accuracy of the model. In scenarios such as dialogue systems, content generation, and question answering, carefully designed prompts guide the model to output results that meet expectations.

[0124] Furthermore, after obtaining the sample business prompt template corresponding to the target sample advertisement text through the template engineering construction function, this sample business prompt template can be input into the initial business model. This allows the initial business model to perform a first processing task based on the business prompt template, outputting the target predicted advertisement text that matches the template prompt words in the sample prompt template. It can be understood that the process of the initial business model performing the first processing task here is similar to the process described above where the business processing model performs the first processing task based on the first type of input data. The initial business model can obtain business reference information for the advertisement business to be processed based on the template prompt words in the sample prompt template, and predict and output the target predicted advertisement text that matches the template prompt words based on this business reference information. It can be understood that the business reference information used by the initial business model here is the business reference information corresponding to the advertisement business to be performed after the initial business model is trained. In other words, the business reference information of the advertisement business to be performed after the initial business model is trained will also be used to complete the training of this initial business model.

[0125] After selecting N sample ad texts from the sample storage database as target sample ad texts and generating N corresponding target predicted ad texts, the initial business model can be fine-tuned using these N sample ad texts and their target predicted ad texts to obtain a business processing model for generating ad text materials suitable for target ad business. Specifically, the model loss of the initial business model can be calculated using these N sample ad texts and their target predicted ad texts. It's understood that since there are N sample ad texts, there are also N calculated model losses. These N model losses can be used to determine whether to adjust the model parameters of the initial business model. It should be noted that the embodiments of this application perform iterative training on the initial business model. Therefore, there may be multiple rounds of parameter adjustments to the business processing model. For ease of understanding, we refer to the business processing model during the training process as the transitional business model. After obtaining the transitional business model by adjusting the parameters of the initial business model, the sample business prompt templates of the N sample advertising texts obtained earlier need to be input into the transitional business model again, so that the transitional business model outputs the predicted advertising texts corresponding to these N sample advertising texts. Then, the model loss of the transitional business model is calculated using these N sample advertising texts and their predicted advertising texts to determine whether the transitional business model has converged. If it has converged, the transitional business model at this time is taken as the business processing model that has been trained. If it is determined that the transitional business model has not converged, the model parameters are adjusted and the above training process is repeated until the model converges and the trained business processing model is obtained.

[0126] For further details, please see Figure 5 , Figure 5 This is a schematic diagram illustrating a scenario for training a business processing model, as provided in an embodiment of this application. Figure 5 As shown, when training the business model, N sample ad texts to be processed can first be retrieved from the sample storage database containing sample ad texts. Figure 5 When N is 1, the extracted sample advertisement text is sample advertisement text 41a, which can then be used as the target sample advertisement text. By analyzing the text content of sample advertisement text 41a, some sample advertisement demand information is obtained within it. Figure 5 The sample advertisement requirement 42a is used as an example. By inputting the obtained sample advertisement requirement 42a into the template project construction function, the template project construction function can use the corresponding template construction function instructions to construct the sample business prompt template corresponding to the sample advertisement text 41a, i.e., Figure 5The sample prompt template 43a is used as an example. By inputting the sample prompt template 43a into the business model, the business model can output the corresponding predicted advertising text based on the sample prompt template 43a. The model loss of the business model can be calculated using the predicted advertising text output by the business model and the sample advertising text 41a. The model loss can then be used to determine whether the current business model has converged, and the model parameters can be further adjusted until the business model converges to obtain the business processing model for advertising business. The above is the process of iteratively training the business model through model loss. It can be understood that the business model may need to be trained multiple times to achieve convergence. During this process, the sample prompt template will be input into the business model multiple times (this business model is also called the transitional business model), so that the business model is input with the corresponding predicted advertising text, thereby calculating the corresponding model loss, determining whether the business model has converged, and adjusting the model parameters of the business model to obtain the business processing model.

[0127] It is important to note that when the business processing model receives the first type of input data and performs the first processing task, the business prompt template in this first type of input data is obtained based on the business requirements information of the current advertising business. In this embodiment, the business prompt template also provides a fixed template format applicable to multiple product scenarios, including game products and e-commerce products. It is understood that the advertising business in this embodiment can be a target business targeting a specific product, such as game products or e-commerce products. The business prompt template in this case can include three parts: an emotional guidance section, a middle segment section, and an ending guidance section. Specifically, the emotional guidance section in the business prompt template instructs the business processing model to generate the beginning of the advertising text material as emotional guidance text material that can be used to guide the emotions of the target users, thereby attracting their interest in watching the targeted video. The intermediate segment here is determined based on the product attributes of the target product. In other words, the intermediate segment can contain several sections introducing product attributes. The specific number of product attribute introduction sections can be determined according to the actual business needs of the advertising business and is not limited here. The intermediate segment in the business prompt template can be used to instruct the business processing model to generate the middle part of the advertising text material as product introduction text material for the target product when generating advertising text material. This allows target users to understand the product features of the target product while watching the advertising video, thus further attracting them to the product. Regarding the ending guide in the business prompt template, it can be used to instruct the business processing model to generate the ending part of the advertising text material as ending guide text material that can guide the target users to enter a further advertising scenario. Taking an e-commerce product as an example, the ending guide text material can be click guide text material that guides users to click a link to jump to the sales page of the e-commerce product. Taking a game product as an example, the ending guide text material can be download guide text material that guides users to download the game product.

[0128] Specifically, when the business processing model receives the first type of input data, which includes a business prompt template and business reference information for the target product, and executes the first processing task, it can obtain emotional guidance text information from the business reference information based on the emotional guidance section of the business prompt template, and generate emotional guidance text materials for the target business using the obtained emotional guidance text information. This emotional guidance text information consists of readily available emotional guidance copy related to the target business, such as the emotional guidance section of historical advertising text, or manually created and amplified emotional guidance text by the business processing model. Therefore, the business processing model can refer to this emotional text information to generate emotional guidance text materials. The business processing model can also obtain product attribute information about the target product from the business reference information based on the middle segment of the business prompt template, and generate product introduction text materials for the target business using the obtained product attribute information. Finally, the business processing model can obtain concluding guidance text information from the business reference information based on the concluding segment of the business prompt template, and generate concluding guidance text materials for the target business using the obtained concluding guidance text information. Understandably, the concluding guidance text here is similar to the emotional guidance text; it consists of readily available concluding guidance copy relevant to the target business. Examples include the concluding guidance sections of historical advertising texts, or manually created concluding guidance texts that have been amplified through a business processing model. Therefore, the business processing model can refer to this concluding guidance text information to generate concluding guidance text materials. Furthermore, the business processing model can use the obtained emotional guidance text materials, product introduction text materials, and concluding guidance text materials to generate advertising text materials tailored to the target business.

[0129] For further details, please see Figure 6 , Figure 6 This is a schematic diagram illustrating a scenario where a business processing model using a business prompt template is provided in an embodiment of this application. For example... Figure 6 As shown, when generating a business prompt template for an advertising campaign targeting a specific product, the first step is to organize the product's key selling points and input them into the prompt template diversification tool. It's understandable that the product's key selling points may include multiple points at this stage, such as... Figure 6 The selling points 51a, 51b, 51c, 51d, 51e, and 51f shown are examples of selling points. Taking a game product as an example, these selling points could include gameplay, game activities, game benefits, game characters, game scenes, etc. After inputting the compiled business selling points into the prompt template diversification tool, the tool can be used to construct business prompt template 53a, such as... Figure 6As shown, the constructed business prompt template 53a can be divided into three segments: a beginning segment, a middle segment, and a closing segment. The beginning segment can be created by selecting the emotional guidance section from the prompt template diversification tool, while the closing segment can be created by selecting the closing guidance section. The middle segment can extract 1-3 substructures from the middle segment section of the prompt template diversification tool. Here, the business prompt template 53a is generated by extracting three substructures: selling point 51a, product hotspots, and product knowledge. This means that the middle segment of the final advertising text material generated through the business prompt template 53a will introduce the target product's selling point 51a, product hotspots, and product knowledge respectively. After generating the business prompt template 53a, the text structure indicated in the business prompt template 53a and the business reference information for the advertising business can be input into the business processing model 50a. This allows the business processing model 50a to generate each part of the text structure indicated by the business prompt template based on the business reference information, ultimately resulting in the advertising text material.

[0130] Understandably, the business reference information here is... Figure 6 The business processing model 50a can refer to the emotional guidance text information in the emotional guidance text library, the selling point introduction library, the hot topic information queried by the hot topic query tool, the product knowledge base, and the guidance text library to generate the emotional guidance part of the text indicated by the business prompt template 53a; refer to the selling point introduction library to generate the selling point 51a introduction part of the text indicated by the business prompt template 53a; refer to the hot topic information to generate the product hot topic introduction part of the text indicated by the business prompt template 53a; refer to the product knowledge base to generate the product knowledge introduction part of the text indicated by the business prompt template 53a; and refer to the ending guidance text information in the guidance text library to generate the ending guidance part of the text indicated by the business prompt template 53a. Finally, through the completion of each text part, the advertising text material 52a is generated. It is understood that the business selling point information in the selling point introduction library, the product knowledge information in the product knowledge base, and the hot topic information referenced above are all product attribute information about the target product. For example... Figure 6 As shown, the business processing model 50a can be iteratively trained using sample advertising texts from the sample storage database, thereby improving the quality of the advertising text materials generated by the business processing model 50a based on the business prompt template. For the specific process of iteratively training the business processing model 50a using sample advertising texts from the sample database, please refer to the above... Figure 5 Description of the illustrated embodiment.

[0131] It is understood that, after obtaining the advertising text material for the advertising business in this embodiment, the advertising text material can be converted into advertising voice content. This advertising voice content can be understood as the video voice content to be used in the final generated target video. The technology used to convert the advertising text material into advertising voice content can be text-to-speech (TTS), a technology that converts text data into audible sound. TTS systems typically include multiple steps such as text analysis, prosody processing, and speech synthesis. It is widely used in fields such as intelligent assistants, accessibility technology, education, and entertainment, providing users with convenient voice output services. The system architecture and processing flow of TTS-related processing systems are not fixed; when using a specific TTS processing system, it can be selected according to the actual business needs of the advertising business, and no specific limitations are made here. Furthermore, the obtained advertising voice content can be subjected to speech recognition to obtain the corresponding speech-recognized text. The technology used for speech recognition of the advertising voice content can be automatic speech recognition (ASR), a technology that converts human speech signals into text data. ASR systems typically include steps such as speech signal processing, feature extraction, and pattern recognition. They are widely used in scenarios such as voice assistants, voice input methods, real-time caption generation, and automated telephone answering. The system architecture and processing flow of ASR-related processing systems are not fixed. When using a specific ASR processing system, you can choose according to the actual business needs of the advertising business; no specific limitations are set here.

[0132] Step S102: Perform text segmentation on the speech recognition text to obtain text segmentation segments. In the advertising speech content, determine the speech segmentation segment corresponding to the text segmentation segment, and determine the segmentation label of the speech segmentation segment through the segmentation content of the text segmentation segment, and determine the segmentation duration of the speech segmentation segment.

[0133] Specifically, after obtaining the corresponding speech-recognized text by performing speech recognition on the advertising audio content, text segmentation markers present in the speech-recognized text can be identified. Based on these markers, the speech-recognized text is then segmented into text segments carrying the segmentation markers. It's understood that the speech-recognized text obtained through speech recognition carries text segmentation markers. These markers can be punctuation marks, spaces, or other characters; in short, they divide the corresponding speech-recognized text into multiple segments. Taking punctuation marks as an example, different segmentation methods can be chosen according to the business needs of the current advertising operation, including using periods, commas, etc. No specific limitations are imposed here. After segmentation, each resulting text segment will also carry a corresponding text segmentation marker at the end, with the segment ending with a punctuation mark. Understandably, after dividing the speech-recognition text into multiple text segments, the corresponding speech segments can be identified within the advertising speech content. Since there's a one-to-one correspondence between the advertising speech content and the speech-recognition text, there's also a one-to-one correspondence between the text segments and the speech segments. In other words, if each speech segment is converted into text using speech recognition technology, the content of that text and its corresponding text segment are identical. Essentially, we first divide the speech-recognition text into text segments, and then identify the corresponding speech segment for each text segment to achieve the division of the advertising speech content into speech segments.

[0134] After obtaining the speech segments, the content of the corresponding text segments can be analyzed to determine the key content of the text segments. Then, a segmentation tag is assigned to each speech segment based on the key content of the text segments. Specifically, each text segment may contain multiple words, some of which are key words. These key words can be used to summarize the main content of the text segment regarding the current advertising business; these key words can also be called the text key content. Therefore, the main content of each text segment regarding the current advertising business can be determined through the text key content, and thus, the segmentation tag for each corresponding speech segment can be assigned based on the text key content. This segmentation tag can also be understood as a label summarizing the main content of the speech segment to which that tag belongs. For example, if the advertising business is targeting a specific game product, and a speech segment is introducing game benefits, then the segmentation tag for that speech segment could be "benefits," and the main content of that speech segment can be determined to be an introduction to game benefits.

[0135] Understandably, when performing speech recognition on advertising audio content to obtain speech-recognized text, the speech-recognized text can carry the time point corresponding to each text segmentation marker in the advertising audio content, also known as the text segmentation time point. Since each text segment ends with a text segmentation marker, the segment duration of the corresponding audio segment can be calculated using the text segmentation time point corresponding to the text segmentation marker at the end of the current text segment, and the text segmentation time point corresponding to the text segmentation marker at the end of the previous text segment. In other words, the segment duration of each obtained audio segment can be determined using the above method. It is important to note that the duration of the audio segment corresponding to the first text segment in the speech-recognized text can be directly obtained using the text segmentation time point of the text segmentation marker at the end of that text segment, because the start time point of that audio segment can be understood as starting from zero. It is important to note that the reason this embodiment does not directly use the text segmentation markers present in the advertising text material to divide the advertising text material into text fragments is because the text segmentation markers themselves do not possess text segmentation time points suitable for direct text segmentation. Therefore, to obtain speech content that can be used for narration, this embodiment proposes to convert the content of the currently obtained advertising text material into advertising speech content through text-to-speech conversion. This allows for the determination of the speech recognition text corresponding to the advertising speech content through speech recognition. In other words, this embodiment, through speech recognition, can ensure that the text segmentation markers in the currently obtained speech recognition text carry specific text segmentation time points. Therefore, when performing text segmentation on speech recognition text, text segments of corresponding text durations can be dynamically segmented based on the text segmentation time points carried by the text segmentation identifiers. This allows for the identification of speech segments matching the corresponding text duration segments from the advertising speech content, thus obtaining the segmentation duration of each speech segment. In this way, when retrieving video using the segmentation duration and segmentation tag of each speech segment, this embodiment can retrieve video segments that match both the segmentation tag and the segmentation duration as much as possible. Consequently, when obtaining advertising video materials through the retrieved video segments, it can ensure that the final generated advertising video achieves audio-visual synchronization between the video image and the audio.

[0136] For further details, please see Figure 7 , Figure 7 This is a schematic diagram illustrating a scenario for segmenting advertising voice content, provided in an embodiment of this application. For example... Figure 7As shown, after obtaining the advertising voice content 60a, speech recognition can be performed on the advertising voice content to obtain speech-recognized text 60b. The obtained speech-recognized text 60b carries multiple text segmentation identifiers, such as... Figure 7 As shown, the speech recognition text 60b carries segmentation identifiers 62a, 62b, and 62c. These segmentation identifiers can divide the speech recognition text 60b into multiple text segments, including text segment 65a, text segment 65b, and text segment 65c. It can be understood that the text segments obtained through these segmentation identifiers each carry a corresponding segmentation identifier at the end. Text segment 65a carries segmentation identifier 62a at the end, text segment 65b carries segmentation identifier 62b at the end, and text segment 65c carries segmentation identifier 62c at the end. Alternatively, segmentation identifier 62a can be used to segment text segment 65a, segmentation identifiers 62a and 62b can be used to segment text segment 65b, and segmentation identifiers 62b and 62c can be used to segment text segment 65c. By obtaining these multiple text segments, the corresponding speech segment can be determined from the advertising speech content 60a, such as... Figure 7 As shown, the audio segment 66a in the advertising audio content 60a can be determined by the text segment 65a, the audio segment 66b in the advertising audio content 60a can be determined by the text segment 65b, and the audio segment 66c in the advertising audio content 60a can be determined by the text segment 65c.

[0137] After dividing the advertising audio content 60a into multiple audio segments by determining audio segments for each text segment, the corresponding segmentation label and segmentation duration can be determined for each audio segment by the segment content of the text segment and the text segmentation identifier it carries. Specifically, the segment tag 63a of the audio segment 66a can be determined by the text content 61a in the text segment 65a, and the segment duration 64a of the audio segment 66a can be determined by the time point corresponding to the segmentation marker 62a in the text segment 65a. Similarly, the segment tag 63b of the audio segment 66b can be determined by the text content 61b in the text segment 65b, and the segment duration 64b of the audio segment 66b can be determined by the time points corresponding to the segmentation marker 62a and 62b in the text segment 65a and text segment 65b. Likewise, the segment tag 63c of the audio segment 66c can be determined by the text content 61c in the text segment 65c, and the segment duration 64c of the audio segment 66c can be determined by the time points corresponding to the segmentation marker 62b and 62c in the text segment 65b and text segment 65c. This allows us to determine the segment segmentation tag and segment segmentation duration of each audio segment obtained from the segmentation of the advertising audio content.

[0138] Step S103: Based on the segment segmentation tags and segment segmentation duration, select video segments that match the advertising business from multiple video segments contained in the video segment library;

[0139] Specifically, after obtaining the segmentation tag and duration for each speech segment, the system can use these tags and durations to filter out corresponding video segments from a video segment library relevant to the current advertising business. This video segment library can be understood as a database specifically designed for the current advertising business, storing video segments. The sources of these video segments can include those provided by advertisers, manually created segments, and segments based on historical target videos, etc. We do not impose excessive restrictions on the sources of the video segments in this library. It's important to note that the video segments in the library can be categorized into two types: first-type candidate video segments carrying video segment tags, and second-type candidate video segments carrying general video tags. Regarding the first type of candidate video clips, the video clip tag carried by these clips indicates their main content. In other words, the video clip tag is obtained through content analysis of the video clip's content. Taking advertising targeting a specific game product as an example, the content of the video clip might be an introduction to a scene within the target game; in this case, the video clip tag could be "scene." The specific process of performing content analysis on the first type of candidate videos and configuring video clip tags can be done manually or through intelligent models, etc. The method of completion depends on the business needs of the advertising business, and no specific limitations are made here.

[0140] Regarding the second type of candidate video segments in the video segment library, these video segments can be understood as general video segments applicable to advertising. In other words, the biggest difference between the second type and the first type is that these video segments do not require specific video segment tags. Assigning different advertising-related audio content segments to these video segments will not appear out of place. Therefore, when a matching first-type candidate video segment cannot be found for an audio segment, the second-type candidate video segment, i.e., the general video segment, can be used as the matching video segment for that audio segment. It can be understood that the second-type candidate video segments do not need to carry video segment tags to indicate the content of the video segment, or it can be understood that all second-type candidate video segments carry general video tags that can be used to indicate that the video segment is a general video segment.

[0141] This embodiment first searches a video segment library for video segment tags that match the segment segmentation tags of the voice segment. The first type of candidate video segments corresponding to the found video segment tags are then used as target first type candidate video segments. Video segments matching the segment length of the voice segment can then be selected from these target first type candidate video segments. Specifically, when selecting video segments to obtain voice segment segments, the segment segmentation tags of the voice segment are first obtained. Then, video segmentation tags with similar or identical content to the segment segmentation tags are searched in the video segment library. Taking an advertising business targeting a game as an example, the segment segmentation tag of the voice segment might be "character," indicating that the content of the voice segment mainly introduces the characters of the target game. At this time, video segment tags with the same or similar semantics as "character" can be searched in the video segment library. It can be understood that the main content of the first type candidate video segments corresponding to these found video segment tags also mainly introduces the characters of the target game. After finding video segment tags that match the segmentation tags of the speech segments, the first-category candidate video segments corresponding to these tags can be used as target first-category candidate video segments. It's understandable that there may be multiple video segment tags matching the segmentation tags, resulting in multiple target first-category candidate video segments. The next step is to filter these video segments to select those whose duration difference from the speech segment's segment duration meets the specified range. This is because the selected video segments correspond one-to-one with the speech segments. When the difference between the video segment's duration and the speech segment's duration is small, the final target video will have a higher degree of matching between the video image and the speech content. Although video optimization can be performed to make the video segment's duration consistent with the speech segment's duration, when the difference between the video segment's duration and the speech segment's duration is relatively small, the optimized video segment will be more complete.

[0142] Therefore, the video segment duration of the first-category candidate video can be used as the target video segment duration, and the duration difference between the target video segment duration and the segment division duration of the speech segment can be determined. It is understandable that since there may be multiple first-category candidate videos selected, there may be multiple target video segment durations, and thus multiple corresponding target video segment duration differences. In this case, the target duration difference that meets the duration difference range condition can be obtained from these multiple target video segment duration differences. Then, the target first-category candidate video segment corresponding to the target duration difference is used as the video segment recalled corresponding to the current speech segment. Here, the duration difference range condition is a range condition pre-set based on the business needs of the current advertising business. When the target video segment duration difference is within this range, it can be considered a target duration difference that meets the condition, and the video segment corresponding to this difference is also a video segment that can match the current speech segment. It is understandable that the above process may filter out multiple video segments for certain speech segments within each speech segmentation. These multiple video segments can then be used to create multiple target delivery videos. In other words, by recalling video segments for speech segments through the embodiments of this application, multiple target delivery videos can be obtained in the end. It is understood that the production process for each of these multiple target delivery videos is similar, and this situation will not be elaborated upon further here. Of course, if there are multiple target first-class candidate video segments whose video segment lengths meet the condition, the target first-class candidate video segment with the smallest difference between its video segment length and the segmentation length of the current speech segment can be directly selected as the video segment recalled by the current speech segment. In this way, a corresponding video segment will be recalled for each speech segment, and correspondingly, the final target delivery video generated based on these speech segments will also be one.

[0143] Understandably, when searching for video segment tags matching the segment segmentation tags in the video segment library, it's possible that no matches will be found, or the duration of the first-category candidate video segments found may not meet the requirements. In this case, a second-category candidate video segment—that is, a video segment carrying a general video tag—can be retrieved from the video segment library and used as the recalled video segment corresponding to the current speech segment. Of course, if there are multiple second-category candidate video segments, the one with the smallest difference in duration between its video segment duration and the duration of the current speech segment segment can be selected as the recalled video segment corresponding to the current speech segment. For further details, please refer to [link to relevant documentation]. Figure 8 , Figure 8 This is a schematic diagram illustrating a scenario for segmenting and filtering video segments based on speech data, as provided in an embodiment of this application. Figure 8As shown, after obtaining the speech segment (speech segment 76a), the corresponding video segment can be filtered from the video segment library using the segment segmentation tag (speech tag 761a) carried by speech segment 76a and the segment segmentation duration (speech duration 762a). Specifically, as... Figure 7 As shown, the video clip library at this time stores first-class candidate video clips, including video clip 77a with video clip tag 771a and video clip duration 772a, video clip 77b with video clip tag 771b and video clip duration 772b, and video clip 77c with video clip tag 771c and video clip duration 772c. The video clip library also stores second-class candidate video clips, including video clip 77d with video clip duration 772d and video clip 77e with video clip duration 772e. When filtering video clips, it is first necessary to check if there is a video clip in the first-class candidate video clips whose video tag matches the clip tag 761a of the audio clip 76a. If so, ... Figure 8 As shown, at this time, the video tag 771a of video segment 77a and the video tag 771b of video segment 77b match the segment tag 761a. Then, the difference can be calculated between the segment duration 762a of audio segment 76a and the video duration 772a of video segment 77a and the video duration 772b of video segment 77b, respectively. The video segment 77b with the smallest duration difference is selected as the video segment corresponding to audio segment 76a. If no video segment matching the video tag of audio segment 76a's segment tag 761a is found in the first category of candidate video segments, video segments will be filtered from the second category of candidate video segments, namely video segments 77d and 77e. In this case, the difference can be calculated between the segment duration 762a of audio segment 76a and the video duration 772d of video segment 77d and the video duration 772e of video segment 77e, respectively. The video segment 77d with the smallest duration difference is selected as the video segment corresponding to audio segment 76a.

[0144] Step S104: Determine the advertising video materials for the advertising business based on the selected video clips, and generate a target delivery video for the advertising business based on the advertising video materials and advertising voice content.

[0145] It's understandable that the audio segments obtained from the advertising audio content may consist of multiple audio segments. Alternatively, it can be said that the obtained audio segments contain multiple sub-audio segments, and these sub-audio segments together constitute the advertising audio content. After retrieving the corresponding video segments from the video segment library for each of these sub-audio segments, these video segments can be used to construct advertising video materials for the advertising business. In other words, the advertising video materials at this time include multiple video materials corresponding to multiple sub-audio segments. Further, one sub-audio segment can be selected from these multiple sub-audio segments as the target sub-audio segment, and the video material corresponding to the target sub-audio segment in the advertising video materials can be identified as the target video material. At this point, the target video material can be optimized to obtain optimized video material corresponding to the target sub-audio segment. The video length of the optimized video material is consistent with the segment length of the target sub-audio segment. In other words, the material optimization here will optimize the duration of the target video material. This duration optimization can be achieved by cropping or increasing the video length of the target video material, thereby obtaining the optimized video material corresponding to the target sub-audio segment.

[0146] After selecting multiple sub-speech segments as target sub-speech segments, optimized video materials can be developed for each sub-speech segment. It's understood that these sub-speech segments have a specific order within the ad's audio content; this order can be referred to as the segment segmentation order. The segment segmentation order can then be used to determine the splicing order of the optimized video materials for each sub-speech segment. These materials are then spliced ​​together according to this order to obtain a spliced ​​video. In other words, the order of the optimized video materials in the spliced ​​video is consistent with the order of the corresponding sub-speech segments within the ad's audio content. The resulting spliced ​​video and the ad's audio content are then merged to produce a merged video. This merged video uses the ad's audio content as the audio feed and the spliced ​​video as the video frame. Finally, the merged video can be used as the target video for the current advertising campaign. Understandably, before using the obtained speech-merged video as the target video for the current advertising business, the obtained speech recognition text can be used as the video text of the speech-synthesized video. Then, the speech-synthesized video can be updated using this video text. Here, the video text can be understood as video subtitles. Updating the speech-synthesized video involves adding subtitles to the speech-synthesized video using the speech recognition text.

[0147] For further details, please see Figure 9 , Figure 9 This is a schematic diagram illustrating a scenario for generating audio-visually merged video, as provided in an embodiment of this application. Figure 9 As shown, by performing speech recognition and segmentation on the advertising speech content 73a, speech segments corresponding to the advertising speech content can be obtained. These speech segments can include multiple sub-speech segments, including... Figure 9 The audio segments are 72a, 72b, 72c, 72d, 72e, and 72f. Further, matching video segments can be retrieved from a video segment library for these multiple sub-speech segments, and the video segment corresponding to each sub-speech segment can be optimized to obtain multiple optimized video segments, ensuring that the duration of these optimized video segments matches the duration of the corresponding sub-speech segments. For example... Figure 9 As shown, optimized video clip 70b retrieved from audio clip 72a can be optimized to obtain optimized video clip 71a; optimized video clip 70e retrieved from audio clip 72b can be optimized to obtain optimized video clip 71b; optimized video clip 70b retrieved from audio clip 72c can be optimized to obtain optimized video clip 71c; optimized video clip 70a retrieved from audio clip 72d can be optimized to obtain optimized video clip 71d; optimized video clip 70d retrieved from audio clip 72e can be optimized to obtain optimized video clip 71e; and optimized video clip 70c retrieved from audio clip 72f can be optimized to obtain optimized video clip 71f. Furthermore, the obtained optimized video clips can be sorted according to the segmentation order of the sub-speech segments in the advertising audio content. It can be understood that the splicing order of the optimized video clips after sorting is as follows: optimized video clip 71a, optimized video clip 71b, optimized video clip 71c, optimized video clip 71d, optimized video clip 71e, and optimized video clip 71f. By splicing the sorted and optimized video clips, we can obtain spliced ​​video 74a. Then, we can perform voice merging processing on the obtained spliced ​​video 74a and the advertising voice content 73a to obtain voice merged video 75a.

[0148] It is understood that, in the embodiments of this application, advertising text materials for advertising business can be converted into advertising voice content, and then the voice segments obtained based on the advertising voice content can be divided into segments and video segments that match each segment can be selected from the video segment library. Based on the selected video segments, advertising video materials for advertising business can be obtained, so that the video screen of the final target video can be accurately matched with the video voice. Furthermore, by directly dividing the voice into segments and recalling video segments from the video library to obtain advertising video materials, the production efficiency of advertising videos can be greatly improved.

[0149] For further details, please see Figure 10 , Figure 10 This is a schematic diagram illustrating another advertising video generation method provided in an embodiment of this application. It is understood that... Figure 3 The method for generating advertising videos shown can be executed by a computer device, which can be a user terminal or a server. For ease of understanding, this example uses a computer device as a user terminal. This user terminal can be, for example, a... Figure 1 The user terminal 10a is shown. (As shown in the image) Figure 10 As shown, the advertising video generation method may include at least the following: Figure 10 Steps S201 to S214 in the process.

[0150] Step S201: Obtain advertising text material for advertising business. When converting the material content in the advertising text material into advertising voice content, perform speech recognition on the advertising voice content to obtain the speech recognition text of the advertising voice content.

[0151] In this embodiment, when generating a target video for advertising, the relevant advertising text material is first obtained. There are several ways to obtain this advertising text material. One method is to input a relevant business prompt template for advertising into the business processing model, allowing the model to output the corresponding advertising text material based on the template. Another method is to input existing advertising text into the model, allowing it to undergo text splitting processing, and then use the resulting split text as the advertising text material for the advertising business. After obtaining the advertising text material, text-to-speech technology can be used to convert it into advertising speech content. Then, speech recognition is performed on the advertising speech content to obtain the speech-recognized text corresponding to the advertising speech content. For a detailed explanation of step S201, please refer to the above. Figure 3 The description of step S101 in the corresponding embodiments will not be repeated here.

[0152] Step S202: Perform text segmentation on the speech recognition text to obtain text segmentation segments. In the advertising speech content, determine the speech segmentation segment corresponding to the text segmentation segment. Determine the segmentation label of the speech segmentation segment through the segmentation content of the text segmentation segment and determine the segmentation duration of the speech segmentation segment.

[0153] After obtaining the speech recognition text of the advertising audio content, the speech recognition text can be segmented using text segmentation markers present in the speech recognition text to obtain text segmentation segments carrying text segmentation markers. Further, after dividing the speech recognition text into multiple text segmentation segments, the corresponding audio segmentation segments can be determined within the advertising audio content. In other words, this embodiment of the application achieves the segmentation of advertising audio content into audio segmentation segments by first dividing the speech recognition text into text segmentation segments and then determining the corresponding audio segmentation segments for each text segmentation segment. After obtaining the audio segmentation segments, the segment content of the text segmentation segments corresponding to each audio segmentation segment can be analyzed to determine the segmentation tag corresponding to each audio segmentation segment. Alternatively, the segmentation duration of each audio segmentation segment can be calculated using the text segmentation time point corresponding to the text segmentation markers carried in the audio segmentation segments. For the specific process of step S202, please refer to the above. Figure 3 The description of step S102 in the corresponding embodiments will not be repeated here.

[0154] Step S203: Based on the segment segmentation tags and segment segmentation duration, select video segments that match the advertising business from multiple video segments contained in the video segment library;

[0155] After obtaining the segmentation tags and durations of the speech segments, the corresponding video segments matching the current advertising business can be filtered from a video segment library related to the current advertising business using these tags and durations. Specifically, in this embodiment, after obtaining the advertising text material, the advertising text material can be digitized into speech, and then multiple speech segments can be obtained using speech recognition technology. Matching video segments can then be retrieved from the video segment library based on the duration and tags of these multiple speech segments. The above process can be summarized by the following formula:

[0156] l video =Label(v) (3)

[0157] s = TTS(t) (4)

[0158] t1,t2,t3,t4,…=ASR(s) (5)

[0159] recall = similarity semantic (l video ,t n=1,2,3,4,… (6)

[0160] In formula (3), v represents the video segment in the input video segment library, Label represents the process of tagging the video segment, and lvideo The tags are for video clips; in formula (4), t is the advertising text material, TTS is the text-to-speech function, and s is the generated advertising audio content; in formula (5), ASR is the speech recognition function, and t1, t2, t3, t4, ... are the timestamped text segments obtained by performing speech recognition on the advertising audio content. Here, the duration of the audio segment corresponding to the text segment can be obtained through the timestamp. Furthermore, the semantic similarity matching function similarity can be used in formula (6). semantic For each text segment, video clips whose tags match the text content are retrieved. The recall is the final recall result.

[0161] For details on step S203, please refer to the description of step S103 above, which will not be repeated here.

[0162] Step S204: Determine the advertising video materials for the advertising business based on the selected video clips;

[0163] Step S205: Select the target sub-speech segment from multiple sub-speech segments, and determine the video material corresponding to the target sub-speech segment as the target video material;

[0164] Step S206: Based on the segment segmentation duration of the target sub-speech segments, optimize the target video material to obtain optimized video material for the target sub-speech segments;

[0165] Step S207: When each of the multiple sub-speech segments is selected as the target sub-speech segment, optimized video material for each sub-speech segment is obtained.

[0166] Specifically, the speech segments obtained from the advertising audio content can include multiple sub-speech segments. After retrieving corresponding video segments from the video segment library for each of these sub-speech segments, these video segments can be used to construct advertising video materials for the advertising business. In other words, the advertising video materials at this time include multiple video materials corresponding to multiple sub-speech segments. Further, a sub-speech segment can be selected as the target sub-speech segment from these multiple sub-speech segments, and the video material corresponding to the target sub-speech segment in the advertising video materials can be determined as the target video material. Then, the target video material is optimized to obtain the optimized video material for the target sub-speech segment. After selecting multiple sub-speech segments as target sub-speech segments, optimized video materials can be obtained for each sub-speech segment.

[0167] Step S208: Based on the segmentation order corresponding to multiple sub-speech segments, determine the material splicing order of the optimized video material for each sub-speech segment, and splice the optimized video material for each sub-speech segment according to the material splicing order to obtain the spliced ​​video.

[0168] Step S209: Perform audio merging processing on the spliced ​​video material using the advertising audio content to obtain an audio-merged video;

[0169] Understandably, the order of the multiple sub-speech segments within the advertising audio content can be called the segmentation order. This order can then be used to determine the splicing order of the optimized video materials corresponding to each sub-speech segment. The optimized video materials for each sub-speech segment are then spliced ​​together according to this splicing order to obtain a spliced ​​video. It's also understood that the order of the optimized video materials in the spliced ​​video is consistent with the order of the corresponding sub-speech segments within the advertising audio content. Furthermore, by merging the spliced ​​video and the advertising audio content, a merged audio video can be obtained. The video visuals in this merged audio video are based on the spliced ​​video, while the audio is based on the advertising audio content.

[0170] For details on steps S204 to S209, please refer to the description of step S104 above, which will not be repeated here.

[0171] Step S210: Obtain audio files related to the advertising business and identify key audio segments in the audio files;

[0172] Step S211: Add key audio segments as background audio to the speech-synthesized video, and use the speech-synthesized video as the target video for advertising business.

[0173] Understandably, after obtaining the speech-synthesized video, background audio related to the current advertising business can be added to it. This background audio can be background music. Specifically, firstly, an audio file related to the current advertising business can be obtained; this audio file could be an audio file of a specific piece of music. After obtaining this audio file, it can be segmented into multiple audio segments based on the periodicity of the audio signal. Further, a target audio segment can be selected from these multiple audio segments, while the other audio segments are identified as audio segments to be matched. By determining the segment similarity between the target audio segment and these audio segments to be matched, the number of similar segments in the audio segments to be matched that are similar to the target audio segment is determined. The audio segments whose number of similar audio segments among the multiple audio segments meets the key audio determination criteria are identified as key audio. In fact, the above process can be understood as classifying multiple audio segments of an audio file to obtain several target audio segments, and counting the number of similar segments among these target audio segments. Then, the target audio segment that meets the criteria for determining key audio is identified as the key audio. Here, the criteria for determining key audio can be that the number of similar segments of the target audio segment is the highest among the classified target audio segments. In other words, through the above process, the audio segment that appears most frequently in the audio file can be selected as the key audio segment. The specific process for obtaining key audio segments can be summarized by the following formula:

[0174] x1,x2,x3,x4,…=selfcorrelation(x) (7)

[0175] x′1,x′2,x′3,…=cluster(x1,x2,x3,x4,…) (8)

[0176] x′=max(count(x′1,x′2,x′3,…)) (9)

[0177] In formula (7), x represents the audio source signal of the input audio file, selfcorrelation is an autocorrelation function, a method for measuring signal similarity, which can be used to determine the periodic components in the signal. In this embodiment, it is mainly used to segment audio segments, and x1, x2, x3, x4, ... are the segmented audio segments. In formula (8), cluster is a clustering function, which can classify audio segments with high similarity, while x1 ′ x2 ′ ,x ′3,… are target audio segments with high similarity after clustering. In formula (9), the count function is mainly used to count the number of times similar audio appears, the max function is used to obtain the audio segment with the most occurrences, and x… ′ It is the key audio segment in the final output;

[0178] Furthermore, after identifying the key audio segment, this key audio segment can be used as the background audio for the current speech-synthesized video and added to the video.

[0179] For further details, please see Figure 11 , Figure 11 This is a schematic diagram illustrating a scenario for determining key audio segments, provided in an embodiment of this application. Figure 11 As shown, after obtaining audio file 80a for the current advertising business, the audio file 80a can be divided into multiple audio segments based on the periodicity of its audio signal, including audio segment 81a, audio segment 81b, audio segment 81c, audio segment 81d, audio segment 81e, and audio segment 81f. Furthermore, audio segments with segment similarity can be categorized to obtain the target audio segment. For example... Figure 8 As shown, audio segments 81a and 81f are similar segments and can be categorized as target audio segment 82a. It's understandable that, because audio segments 81a and 81f are similar, either one can be arbitrarily chosen as target audio segment 82a. Audio segments 81b, 81c, and 81d are similar segments and can be categorized as target audio segment 82b. It's understandable that, because audio segments 81b, 81c, and 81d are similar, either one can be arbitrarily chosen as target audio segment 82b. Because audio segment 81e has no similar segments in the audio file, it can be categorized as target audio segment 82c. After obtaining the target audio segments, the target audio segment with the most similar segments, or the one that appears most frequently, can be selected as the key audio segment. Since target audio segment 82a appears twice, target audio segment 82b appears three times, and target audio segment 82c appears once, the target audio segment 82b with the most frequent appearance will be selected as the key audio segment 83a.

[0180] Step S212: Deliver the target video to the advertising platform. During the video delivery duration of the target video, count the amount of interaction data of the target video on the advertising platform.

[0181] Step S213: If the video delivery duration reaches the video delivery duration threshold and the amount of interaction data counted within the video delivery duration reaches the advertising delivery metric, then update the target historical advertising video through the target delivery video and update the historical video segment through the video segment of the target delivery video.

[0182] Step S214: If the video delivery duration reaches the video delivery duration threshold, and the amount of interaction data counted within the video delivery duration does not reach the advertising delivery target, then remove the advertising video material from the video clip library.

[0183] Understandably, after generating a target video, it can be placed on an advertising platform specific to the current advertising business. This target video can then be referred to as a historical video. Within a certain duration, the data interaction volume of the target video on the advertising platform can be measured. This data interaction volume primarily refers to click-through rate (CTR). The data interaction volume obtained here helps determine the performance of the target video on the advertising platform. If the target video's duration exceeds the threshold and the measured data interaction volume reaches the advertising metrics, the target video will be designated as a historical video, and its video clips will be added to the video clip library as historical video clips. If the video duration exceeds the threshold, but the data interaction volume does not reach the advertising metrics, the advertising video assets used to generate the target video can be removed from the video clip library. It is understood that the video segments in the video segment library of this application embodiment may be historical video segments obtained by segmenting the target historical delivery video. Here, the target historical delivery video refers to the video that meets the advertising delivery index selected from the historical delivery videos on the advertising delivery platform. In other words, the data interaction volume of the target historical delivery video on the advertising delivery platform has reached the pre-set advertising delivery index. Here, the advertising delivery index is a fixed value about the data interaction volume set based on the current advertising business.

[0184] This application embodiment can obtain historical campaign videos that meet advertising metrics from an advertising platform as target historical campaign videos. Then, it performs a legality check on the video frames in the obtained target historical campaign videos. If the legality check result indicates the presence of illegal video frames in the target historical campaign videos, the corresponding illegal video frames will be filtered out, resulting in a filtered target historical campaign video. The specific process for filtering the target historical campaign videos is as follows:

[0185]

[0186] Among them, Grounddino is an open-set detection algorithm that can detect content within descriptive terms. If it detects inappropriate content A in a video frame, it will filter that video frame. Face++ is a facial age detection algorithm, primarily used to detect facial age. Gemini is a multimodal large language model, mainly used to identify whether there are visually-oriented intentions in a video. Gemini will also filter out inappropriate content B detected in a video frame. The "approved" output obtained through this filtering process represents the compliant frames from the target's historical video feed. These compliant frames can be used to obtain the target's historical filtered videos.

[0187] After obtaining the target historical filtered video, the subtitle information in the target historical filtered video can be cleaned, and the target historical filtered video after subtitle information cleanup is used as the target historical cleaned video. Further, video scenes in the target historical cleaned video can be identified, and video segmentation can be performed based on the video scenes to obtain historical video clips, which can then be added to a video clip library. It is understandable that this segmentation based on video scenes mainly involves finding the transition points between video scenes. In other words, the main content of these historical video clips is from the same scene, allowing for better content summarization and video tag configuration.

[0188] For further details, please see Figure 12 , Figure 12 This is a schematic diagram illustrating a scenario for acquiring historical video clips, provided in an embodiment of this application. Figure 12 As shown, historical videos that meet advertising metrics can be filtered from the online advertising platform. The filtered historical video is the target historical video 84a. Then, video frames in target historical video 84a that do not conform to the current advertising business rules can be detected and filtered to obtain the target filtered historical video 84b. Further, the target filtered historical video 84b can be cleaned to obtain the target cleaned historical video 84c. This cleaning mainly involves removing the video subtitles from the target filtered historical video; therefore, it can be understood that the target cleaned historical video 84c does not contain subtitles. After obtaining the target cleaned historical video 84c, video analysis can be performed to identify video scenes present in the video and segment the video based on these scenes to obtain historical video segments. These historical video segments can include multiple segments, such as... Figure 12As shown, the historical video clips include video clips 85a, 85b, 85c, 85d, and 85e, which can then be added to the video clip library.

[0189] Understandably, after a target video is delivered to an advertising platform, if the data interaction volume of the target video reaches the advertising target within a certain preset duration, or if the video delivery duration is within a video delivery duration threshold, then the target video can be considered as a target historical delivery video, and its video clips can be used to update the video clip library. If the video delivery duration exceeds the video delivery duration threshold, and the data interaction volume of the target video does not reach the advertising target, then the advertising video materials used to generate the target video can be removed from the video clip library. Understandably, this method ensures that the video clips in the video clip library are of higher quality and better meet the business needs of the advertising operations.

[0190] For further details, please see Figure 13 , Figure 13 This is a schematic diagram illustrating a scenario for generating advertising video materials according to an embodiment of this application. For example... Figure 13As shown, the Creative Trend Component 90a provides access to high-quality creative examples for current advertising campaigns on the online advertising platform 95a. The Intelligent Creative Deconstruction Component 91a then deconstructs these creative examples to obtain corresponding video template paradigms. These video template paradigms are used in the creation of the target video. After generating the voice-merged video, effects and editing can be added using this template. These templates are general, high-level templates with excellent historical data. In other words, editing the voice-merged video using this template does not require analyzing the specific video content (including scene transitions and visual content), thus increasing editing efficiency. Furthermore, after obtaining the video template paradigm, the video material generation component 92a can send the advertising video materials obtained based on the advertising voice content to the advertising video generation component 93a. The advertising video generation component 93a will generate a corresponding voice-merged video based on the obtained advertising voice content and advertising video materials related to the advertising business. Then, by using the obtained video template paradigm on the voice-merged video, the target delivery video (finished material) is obtained. Furthermore, the finished material can be evaluated by the material quality assessment component 94a, and high-quality materials can be delivered to the online advertising platform 95a. It can be understood that, based on the evaluation results obtained by the material quality assessment component 94a on the currently generated target delivery video, the creative direction component 90a and the video material generation component 92a can be optimized accordingly, so that the template paradigm and advertising video materials involved in the production of the target delivery video are more in line with the current quality assessment system. Understandably, by optimizing the performance of targeted videos on the online advertising platform 95a, the creative trend component 90a, the video creative generation component 92a, and the creative quality assessment component 94a can be updated and optimized, making the video templates, advertising video creatives, and the assessment of creative quality more in line with the current advertising environment.

[0191] It is understood that the embodiments of this application can convert the obtained advertising text material into advertising audio content, and then divide the advertising audio content into audio segments through speech recognition. Based on the content of the obtained audio segments, matching video segments are retrieved from a video segment library, and these video segments are further spliced ​​together to obtain video material. This method can quickly generate video material that matches the advertising audio content, improving the speed of advertising video production and ensuring that the final advertising video's visuals and audio are accurately matched. Furthermore, by placing the obtained target video on an online advertising platform and updating the video segments in the video segment library based on the target video's performance metrics, subsequent finished videos generated for the same advertising business can better meet the needs of the advertising platform, thereby further improving the production quality of advertising videos.

[0192] For further details, please see Figure 14 , Figure 14 This is a schematic diagram of the structure of an advertising video generation device provided in an embodiment of this application. The advertising video generation device 1 can be a computer program (including program code) running on a computer device; for example, the advertising video generation device 1 can be an application software. The advertising video generation device 1 can be used to execute the functions provided in the embodiment of this application. Figure 3 and Figure 10 The corresponding steps in the method.

[0193] The advertising video generation device 1 may include: a speech recognition module 11, a text segmentation module 12, a video segment filtering module 13, and a video generation module 14.

[0194] The speech recognition module 11 is used to acquire advertising text materials for advertising business. When converting the material content in the advertising text materials into advertising speech content, it performs speech recognition on the advertising speech content to obtain the speech recognition text of the advertising speech content.

[0195] The text segmentation module 12 is used to segment the speech recognition text to obtain text segmentation segments. In the advertising speech content, it determines the speech segmentation segment corresponding to the text segmentation segment, and determines the segmentation label of the speech segmentation segment through the segmentation content of the text segmentation segment, and determines the segmentation duration of the speech segmentation segment.

[0196] The video clip filtering module 13 is used to filter video clips that match the advertising business from multiple video clips contained in the video clip library based on clip segmentation tags and clip segmentation duration.

[0197] The video delivery generation module 14 is used to determine the advertising video materials for the advertising business based on the selected video clips, and to generate a target delivery video for the advertising business based on the advertising video materials and advertising voice content.

[0198] Among them, multiple video clips include segments composed of video footage obtained from target historical delivery videos; target historical delivery videos refer to videos that meet advertising delivery indicators selected from historical delivery videos delivered on advertising delivery platforms related to advertising business.

[0199] The speech recognition module 11 includes: a business model acquisition unit 111, an input data acquisition unit 112, a first task processing unit 113, and a second task processing unit 114.

[0200] The business model acquisition unit 111 is used to acquire a business processing model for advertising business; the business processing model involves multiple tasks including a first processing task that matches the first type of input data and a second processing task that matches the second type of input data;

[0201] The input data acquisition unit 112 is used to acquire target input data for inputting the business processing model.

[0202] The first task processing unit 113 is used to obtain the business prompt template and business reference information of the advertising business from the first type of input data if the target input data belongs to the first type of input data. When executing the first processing task through the business processing model, it obtains reference data content from the business reference information according to the template prompt words in the business prompt template, and generates advertising text material for the advertising business through the reference data content.

[0203] The second task processing unit 114 is used to obtain historical advertising text from the target historical video if the target input data belongs to the second type of input data. When executing the second processing task through the business processing model, the historical advertising text is subjected to text fission processing to obtain fission processed text, which is then used as advertising text material for the advertising business. Fission processing refers to replacing the target text words with related text words when generating associated text words related to the target text words in the historical advertising text.

[0204] Among them, the historical advertising text is the sample advertising text that has been pre-added to the sample storage database; the number of sample advertising texts in the sample storage database is at least N; N is a positive integer;

[0205] The speech recognition module 11 also includes: a sample text acquisition unit 115, a predicted text generation unit 116, a predicted text acquisition unit 117, and a model fine-tuning processing unit 118.

[0206] The sample text acquisition unit 115 is used to acquire N sample advertisement texts for training the initial business model from the sample storage database, and select the target sample advertisement text from the N sample advertisement texts;

[0207] The prediction text generation unit 116 is used to determine the target prediction advertising text based on the target sample advertising text and the initial business model.

[0208] The predicted text acquisition unit 117 is used to obtain the target predicted advertising text for each sample advertising text when each of the N sample advertising texts is selected as the target sample advertising text.

[0209] The model fine-tuning processing unit 118 is used to fine-tune the initial business model using each sample advertisement text and the target prediction advertisement text of each sample advertisement text to obtain the business processing model.

[0210] Among them, the predictive text generation unit is also used to perform text content analysis on the target sample advertisement text to obtain sample advertisement demand information for the sample advertisement text, and use the sample advertisement demand information as sample advertisement business.

[0211] The predictive text generation unit 116 is also used to obtain the template engineering construction function for the sample advertising business, input the sample advertising requirement information into the template engineering construction function, and construct the sample business prompt template for the sample advertising requirement according to the template construction function instructions in the template engineering construction function.

[0212] The predictive text generation unit 116 is also used to input the sample business prompt template into the initial business model, and when the initial business model performs the first processing task, it predicts and outputs the target predictive advertising text that matches the template prompt words in the sample prompt template.

[0213] Among them, the model fine-tuning processing unit 118 is also used to calculate the model loss of the initial business model by using each sample ad text and the target prediction ad text of each sample ad text.

[0214] The model fine-tuning processing unit 118 is also used to adjust the model parameters of the initial business model if the model loss does not meet the model convergence condition, and use the initial business model with adjusted model parameters as a transitional business model.

[0215] The model fine-tuning processing unit 118 is also used to output the predicted advertising text of each sample advertising text in N sample advertising texts through the transition business model;

[0216] The model fine-tuning processing unit 118 is also used to iteratively train the transition business model using each sample advertisement text and the predicted advertisement text of each sample advertisement text, until the model loss of the iteratively trained transition business model meets the model convergence condition, and then the transition business model that meets the model convergence condition is determined as the business processing model.

[0217] The advertising business includes target businesses for the target product; the business prompt template includes an emotional guidance section, a middle section, and an ending guidance section; the middle section is derived based on the product attributes of the target product.

[0218] The first task processing unit 113 is also used for the emotion guidance part based on the business prompt template, to obtain the emotion guidance text information in the business reference information, and to generate emotion guidance text material for the target business through the emotion guidance text information; the emotion guidance text information includes historical emotion guidance text obtained based on historical advertising text.

[0219] The first task processing unit 113 is also used to obtain product attribute information about the target product from the business reference information based on the intermediate segment of the business prompt template, and generate product introduction text material for the target business through the product attribute information;

[0220] The first task processing unit 113 is also used to obtain the ending guidance text information in the business reference information based on the ending guidance part of the business prompt template, and generate ending guidance text material for the target business through the ending guidance text information; the ending guidance text information includes historical ending guidance text obtained based on historical advertising text; the target product includes the target game product, and the ending guidance text material includes download guidance text material for the target game product.

[0221] The first task processing unit 113 is also used to obtain advertising text materials for advertising business based on emotion-guided text materials, product introduction text materials, and ending guidance text materials.

[0222] For the specific implementation details of the business model acquisition unit 111, input data acquisition unit 112, first task processing unit 113, second task processing unit 114, sample text acquisition unit 115, predicted text generation unit 116, predicted text acquisition unit 117, and model fine-tuning processing unit 118, please refer to the above. Figure 3 The description of step S101 in the corresponding embodiments will not be repeated here.

[0223] Among them, the speech recognition text carries text segmentation identifiers for text division, and the text segmentation time point of the text segmentation identifiers is determined when performing speech recognition on the advertising speech content;

[0224] The text segmentation module 12 includes: a text segment confirmation unit 121, a speech segment confirmation unit 122, and a segment duration confirmation unit 123;

[0225] The text segment confirmation unit 121 is used to identify the text segmentation markers in the speech recognition text, and to divide the speech recognition text into text segments based on the identified text segmentation markers to obtain text segmentation segments carrying text segmentation markers.

[0226] The voice segment confirmation unit 122 is used to determine the voice segment corresponding to the text segment in the advertising voice content, determine the key text content from the segment content of the text segment, and determine the segment segmentation label of the voice segment through the key text content.

[0227] The segment duration confirmation unit 123 is used to obtain the text segmentation time point of the text segmentation identifier in the text segmentation segment, and determine the segmentation duration of the speech segmentation segment based on the text segmentation time point.

[0228] For details on the implementation of the text segment confirmation unit 121, the speech segment confirmation unit 122, and the segment duration confirmation unit 123, please refer to the above. Figure 3 The description of step S102 in the corresponding embodiments will not be repeated here.

[0229] The video clips stored in the video clip library include multiple first-class candidate video clips. A first-class candidate video clip refers to a video clip carrying a video clip tag. A video clip tag refers to a video tag obtained by performing content analysis on the content of a first-class candidate video clip.

[0230] The video clip filtering module 13 includes: a video tag search unit 131, a target duration confirmation unit 132, and a duration difference determination unit 133;

[0231] The video tag lookup unit 131 is used to find video segment tags that match the segment division tags in the video segment library, and to take the first type of candidate video segment corresponding to the found video segment tag as the target first type of candidate video segment.

[0232] The target duration confirmation unit 132 is used to take the video segment duration of the first type of candidate video segment as the target video segment duration;

[0233] The duration difference determination unit 133 is used to determine the duration difference between the duration of the target video segment and the segment division duration, obtain the target duration difference that meets the duration difference range condition from the duration difference, and take the target first-class candidate video segment corresponding to the target duration difference as the video segment that matches the advertising business.

[0234] The video clip filtering module 13 also includes: a general clip acquisition unit 134 and a video clip confirmation unit 135;

[0235] The general segment acquisition unit 134 is used to acquire the second type of candidate video segment corresponding to the general video tag in the video segment library if no video segment tag matching the segment division tag is found in the video segment library.

[0236] The video segment confirmation unit 135 is used to determine the video segment that matches the advertising business through the second type of candidate video segments.

[0237] For details on the implementation of the video tag lookup unit 131, target duration confirmation unit 132, duration difference determination unit 133, general segment acquisition unit 134, and video segment confirmation unit 135, please refer to the above. Figure 3 The description of step S103 in the corresponding embodiment will not be repeated here.

[0238] When a speech segment contains multiple sub-speech segments, the advertising video material includes multiple video materials corresponding to the multiple sub-speech segments, with one sub-speech segment corresponding to one video material;

[0239] The video generation module 14 includes: a target segment confirmation unit 141, a target material optimization unit 142, an optimized material acquisition unit 143, an optimized material splicing unit 144, and an audio merging processing unit 145.

[0240] The target segment confirmation unit 141 is used to select a target sub-speech segment from multiple sub-speech segmentation segments and determine the video material corresponding to the target sub-speech segmentation segment as the target video material.

[0241] The target material optimization unit 142 is used to optimize the target video material based on the segment division duration of the target sub-speech segments, so as to obtain optimized video material for the target sub-speech segments; the playback duration of the optimized video material is consistent with the segment division duration of the target sub-speech segments.

[0242] The optimized material acquisition unit 143 is used to obtain optimized video material for each sub-speech segment when each sub-speech segment in multiple sub-speech segments is selected as the target sub-speech segment.

[0243] The optimized material splicing unit 144 is used to determine the material splicing order of the optimized video material of each sub-speech segment based on the segment segmentation order corresponding to multiple sub-speech segments, and to splice the optimized video material of each sub-speech segment according to the material splicing order to obtain the spliced ​​video material.

[0244] The voice merging processing unit 145 is used to perform voice merging processing on the spliced ​​video material based on the advertising voice content to obtain a voice-merged video, and to use the voice-synthesized video as the target video for advertising business.

[0245] The video generation module 14 also includes a video text addition unit 146.

[0246] The video text adding unit 146 is used to add the speech recognition text as video text to the speech synthesis video, and update the speech synthesis video by adding the video text to the speech synthesis video.

[0247] The video generation module 14 also includes: a key audio acquisition unit 147 and a key audio addition unit 148.

[0248] The key audio acquisition unit 147 is used to acquire audio files related to advertising business and determine key audio segments in the audio files;

[0249] The key audio adding unit 148 is used to add key audio segments as background audio to the speech synthesis video.

[0250] The audio in the audio file is music audio, which contains multiple audio segments;

[0251] The key audio acquisition unit 147 is also used to acquire a target audio segment from multiple audio segments and to identify audio segments other than the target audio segment from the multiple audio segments as audio segments to be matched.

[0252] The key audio acquisition unit 147 is also used to determine the segment similarity between the target audio segment and the audio segment to be matched;

[0253] The key audio acquisition unit 147 is also used to determine the target audio segment that is similar to the target audio segment from the audio segments to be matched by segment similarity, and to count the number of similar segments of the target audio segment to be matched;

[0254] The key audio acquisition unit 147 is also used to select the target audio segment as the key audio segment if the number of similar segments meets the key audio determination condition.

[0255] For the specific implementation methods of the target segment confirmation unit 141, target material optimization unit 142, optimized material acquisition unit 143, optimized material splicing unit 144, speech merging processing unit 145, video text addition unit 146, key audio acquisition unit 147, and key audio addition unit 148, please refer to the above. Figure 3 The description of step S104 in the corresponding embodiments will not be repeated here.

[0256] Among them, the video clip library contains multiple video clips including historical video clips of target historical delivery videos; target historical delivery videos refer to videos that have met the advertising delivery indicators selected from historical delivery videos delivered on advertising delivery platforms related to the advertising business.

[0257] Device 1 also includes: a video compliance detection module 15, a video image filtering module 16, a video image cleaning module 17, and a historical video segmentation module 18;

[0258] The video compliance detection module 15 is used to perform compliance detection on each video frame in the target's historical video and obtain the compliance detection result.

[0259] The video frame filtering module 16 is used to filter out illegal video frames from the target historical video if the compliance detection result indicates that there are illegal video frames in each video frame, so as to obtain the target historical filtered video.

[0260] The video image cleaning module 17 is used to clean the subtitle information in the video image of the target historical filtered video, and to use the target historical filtered video with cleaned subtitle information as the target historical cleaned video.

[0261] The historical video segmentation module 18 is used to identify the video scene of the target historical cleaning video, segment the target historical cleaning video based on the identified video scene, obtain video segmentation segments, and add the video segmentation segments as historical video segments to the video segment library.

[0262] For details on the implementation of the video compliance detection module 15, video image filtering module 16, video image cleaning module 17, and historical video segmentation module 18, please refer to the above. Figure 10 The description of step S213 in the corresponding embodiments will not be repeated here.

[0263] Among them, the video clip library contains multiple video clips including historical video clips of target historical delivery videos; target historical delivery videos refer to videos that have met the advertising delivery indicators selected from historical delivery videos delivered on advertising delivery platforms related to the advertising business.

[0264] Device 1 also includes: target video delivery module 19 and historical video update module 20;

[0265] The target video delivery module 19 is used to deliver the target video to the advertising platform and to count the amount of interaction data of the target video on the advertising platform during the video delivery duration.

[0266] The historical video update module 20 is used to update the target historical ad video through the target ad video and update the historical video segment through the video segment of the target ad video if the video ad duration reaches the video ad duration threshold and the amount of interaction data counted within the video ad duration reaches the ad ad ad metrics.

[0267] The device 1 also includes: a video material removal module 21;

[0268] The video material removal module 21 is used to remove advertising video materials from the video clip library if the video delivery duration reaches the video delivery duration threshold and the amount of interaction data counted within the video delivery duration does not reach the advertising delivery target.

[0269] For details on the implementation of the target video delivery module 19, the historical video update module 20, and the video material removal module 21, please refer to the above. Figure 10 The descriptions of steps S212 to S214 in the corresponding embodiments will not be repeated here.

[0270] Further, please see Figure 15 , Figure 15 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Figure 15 As shown, the computer device 1000 can be a user terminal, for example, the one described above. Figure 1 The user terminal 10a in the corresponding embodiment can also be a server, for example, as described above. Figure 1The server 10d in the corresponding embodiment will not be limited here. For ease of understanding, this application takes a computer device as a user terminal as an example. The computer device 1000 may include: a processor 1001, a network interface 1004, and a memory 1005. In addition, the computer device 1000 may also include: a user interface 1003, and at least one communication bus 1002. The communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device. The memory 1005 may also optionally be at least one storage device located away from the aforementioned processor 1001. Figure 15 As shown, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a device control application.

[0271] The network interface 1004 in the computer device 1000 can also provide network communication functionality. Figure 15 In the computer device 1000 shown, the network interface 1004 provides network communication functionality; the user interface 1003 is mainly used to provide an input interface for the user; and the processor 1001 can be used to call the device control application stored in the memory 1005 to achieve: (as described above) Figure 3 and Figure 10 The description of the data processing method in the corresponding embodiments can also be performed as described above. Figure 14 The description of the advertising video generation device 1 in the corresponding embodiments will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated.

[0272] Furthermore, it should be noted that this application embodiment also provides a computer-readable storage medium, which stores a computer program executed by the aforementioned advertising video generation device 1. The computer program includes computer instructions, and when the processor executes the computer instructions, it can execute the aforementioned... Figure 3 and Figure 10 The description of the data processing method in the corresponding embodiments is already provided and will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer-readable storage medium embodiments related to this application, please refer to the description of the method embodiments of this application.

[0273] Furthermore, it should be noted that this application also provides a computer program product, which may include a computer program that can be stored in a computer-readable storage medium. The processor of a computer device reads the computer program from the computer-readable storage medium, and the processor can execute the computer program, causing the computer device to perform the aforementioned... Figure 3 and Figure 10 The description of the data processing method in the corresponding embodiments is already provided and will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer program products or computer program embodiments related to this application, please refer to the description of the method embodiments of this application.

[0274] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0275] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0276] The steps in the method of this application embodiment can be adjusted, combined, or deleted according to actual needs.

[0277] The modules in the device of this application embodiment can be merged, divided, and deleted according to actual needs.

[0278] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0279] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A method for generating advertising videos, characterized in that, include: The process involves acquiring advertising text materials for advertising business, converting the content of the advertising text materials into advertising voice content, performing speech recognition on the advertising voice content, and obtaining the speech recognition text of the advertising voice content. The speech recognition text is segmented to obtain text segmentation segments. In the advertising speech content, the speech segmentation segment corresponding to the text segmentation segment is determined. The segmentation tag of the speech segmentation segment is determined by the segmentation content of the text segmentation segment, and the segmentation duration of the speech segmentation segment is determined. Based on the segment segmentation tags and the segment segmentation duration, video segments that match the advertising business are selected from multiple video segments contained in the video segment library; Based on the selected video clips, the advertising video materials for the advertising business are determined, and based on the advertising video materials and the advertising audio content, a target delivery video for the advertising business is generated.

2. The method according to claim 1, characterized in that, The multiple video segments include segments composed of video frames obtained from target historical delivery videos; the target historical delivery videos refer to videos that meet the advertising delivery indicators selected from historical delivery videos delivered on advertising delivery platforms related to the advertising business. The acquisition of advertising text materials for advertising business includes: Obtain a business processing model for advertising operations; the business processing model involves multiple tasks including a first processing task that matches the first type of input data and a second processing task that matches the second type of input data; Obtain the target input data used to input the business processing model; If the target input data belongs to the first type of input data, then the business prompt template and the business reference information of the advertising business are obtained from the first type of input data. When the first processing task is executed through the business processing model, the reference data content is obtained from the business reference information according to the template prompt words in the business prompt template, and the advertising text material for the advertising business is generated through the reference data content. If the target input data belongs to the second type of input data, then the historical advertising text from the target historical video is obtained from the second type of input data. When the second processing task is executed through the business processing model, the historical advertising text is subjected to text fission processing to obtain fission processed text. The fission processed text is used as advertising text material for the advertising business. The fission processing refers to replacing the target text word with the related text word when generating associated text words related to the target text word in the historical advertising text.

3. The method according to claim 2, characterized in that, The historical advertising text is a sample advertising text pre-added to the sample storage database; the number of sample advertising texts in the sample storage database is at least N; N is a positive integer; Before obtaining the business processing model for the advertising business, the method further includes: Obtain N sample advertising texts from the sample storage database for training the initial business model, and select the target sample advertising text from the N sample advertising texts; Based on the target sample ad text and the initial business model, determine the target predicted ad text for the target sample ad text; When each of the N sample ad texts is selected as the target sample ad text, the target predicted ad text for each sample ad text is obtained. The initial business model is fine-tuned using each sample ad text and the target predicted ad text for each sample ad text to obtain the business processing model.

4. The method according to claim 3, characterized in that, The step of determining the target predicted advertising text based on the target sample advertising text and the initial business model includes: The target sample advertisement text is analyzed to obtain sample advertisement demand information for the target sample advertisement text, and the sample advertisement demand information is used as sample advertisement business. Obtain the template project construction function for the sample advertising business, input the sample advertising requirement information into the template project construction function, and the template project construction function constructs the sample business prompt template for the sample advertising business according to the template construction function instructions in the template project construction function; The sample business prompt template is input into the initial business model. When the initial business model executes the first processing task, it predicts and outputs target predicted advertising text that matches the template prompt words in the sample prompt template.

5. The method according to claim 3, characterized in that, The process of fine-tuning the initial business model using each sample ad text and the target predicted ad text of each sample ad text to obtain the business processing model includes: The model loss of the initial business model is calculated using each sample ad text and the target predicted ad text for each sample ad text. If the model loss does not meet the model convergence condition, the model parameters of the initial business model are adjusted, and the initial business model with adjusted model parameters is used as the transitional business model. The transition business model outputs the predicted advertising text for each of the N sample advertising texts. The transition business model is iteratively trained using the sample advertisement texts and the predicted advertisement texts of the sample advertisement texts until the model loss of the iteratively trained transition business model satisfies the model convergence condition. The transition business model that satisfies the model convergence condition is then determined as the business processing model.

6. The method according to claim 2, characterized in that, The advertising service includes a target service for the target product; the service prompt template includes an emotional guidance section, a middle section, and an ending guidance section; the middle section is derived based on the product attributes of the target product; The step of obtaining reference data content from the business reference information based on the template prompt words in the business prompt template, and generating advertising text material for the advertising business using the reference data content, includes: Based on the emotional guidance section of the business prompt template, emotional guidance text information is obtained from the business reference information, and emotional guidance text material for the target business is generated through the emotional guidance text information; the emotional guidance text information includes historical emotional guidance text obtained based on the historical advertising text. Based on the middle segment of the business prompt template, obtain the product attribute information about the target product from the business reference information, and generate product introduction text material for the target business using the product attribute information; Based on the closing guidance section of the business prompt template, the closing guidance text information in the business reference information is obtained, and closing guidance text material for the target business is generated through the closing guidance text information; the closing guidance text information includes historical closing guidance text obtained based on the historical advertising text; the target product includes the target game product, and the closing guidance text material includes download guidance text material for the target game product; Based on the aforementioned emotional guidance text material, product introduction text material, and concluding guidance text material, advertising text material for the aforementioned advertising business is obtained.

7. The method according to claim 1, characterized in that, The speech recognition text carries a text segmentation identifier for text division, and the text segmentation time point of the text segmentation identifier is determined when performing speech recognition on the advertising speech content; The process of segmenting the speech-recognized text to obtain text segments, identifying the corresponding speech segments within the advertising speech content, determining the segmentation tags of the speech segments based on the content of the text segments, and determining the segment duration of the speech segments includes: The text segmentation identifier in the speech recognition text is identified, and the speech recognition text is segmented based on the identified text segmentation identifier to obtain text segmentation fragments carrying the text segmentation identifier; In the advertising audio content, determine the audio segment corresponding to the text segment, determine the key text content from the segment content of the text segment, and determine the segment segmentation tag of the audio segment through the key text content; Obtain the text segmentation time point of the text segmentation identifier in the text segmentation segment, and determine the segmentation duration of the speech segmentation segment based on the text segmentation time point.

8. The method according to claim 1, characterized in that, The video clips stored in the video clip library include multiple first-class candidate video clips. A first-class candidate video clip refers to a video clip carrying a video clip tag. A video clip tag refers to a video tag obtained by performing content analysis on the content of a first-class candidate video clip. The step of selecting video segments from the video segment library that match the advertising business based on the segment segmentation tags and the segment segmentation duration includes: Search for video segment tags that match the segment division tags in the video segment library, and take the first type of candidate video segments corresponding to the found video segment tags as the target first type of candidate video segments; The video segment duration of the first type of candidate video segment is taken as the target video segment duration; Determine the duration difference between the target video segment duration and the segment division duration, obtain the target duration difference that meets the duration difference range condition from the duration difference, and take the target first-class candidate video segment corresponding to the target duration difference as the video segment that matches the advertising business.

9. The method according to claim 8, characterized in that, The video clips stored in the video clip library include a second type of candidate video clips in addition to the plurality of first-type candidate video clips. The second type of candidate video clips refers to video clips carrying a general video tag; the general video tag is a video tag that is different from the video dispatch tag. The method further includes: If no video segment tag matching the segment division tag is found in the video segment library, then the second type of candidate video segment corresponding to the general video tag is obtained from the video segment library; The video segments that match the advertising business are determined using the second type of candidate video segments.

10. The method according to claim 1, characterized in that, When the speech segment contains multiple sub-speech segments, the advertising video material includes multiple video materials corresponding to the multiple sub-speech segments, with one sub-speech segment corresponding to one video material; The step of generating a targeted video for the advertising business based on the advertising video material and the advertising voice content includes: Select a target sub-speech segment from the plurality of sub-speech segments, and determine the video material corresponding to the target sub-speech segment as the target video material; Based on the segment division duration of the target sub-speech segment, the target video material is optimized to obtain optimized video material for the target sub-speech segment; the playback duration of the optimized video material is consistent with the segment division duration of the target sub-speech segment. When each of the plurality of sub-speech segments is selected as the target sub-speech segment, optimized video material is obtained for each sub-speech segment. Based on the segmentation order corresponding to the multiple sub-speech segments, the splicing order of the optimized video materials for each sub-speech segment is determined, and the optimized video materials for each sub-speech segment are spliced ​​according to the splicing order to obtain a spliced ​​video. The video material is spliced ​​together using the advertising audio content to perform audio merging processing, resulting in an audio-merged video. This audio-generated video is then used as the target video for the advertising business.

11. The method according to claim 10, characterized in that, Before using the speech-synthesized video as the targeted video for the advertising business, the method further includes: Obtain audio files related to the advertising business and identify key audio segments in the audio files; The key audio segments are used as background audio for the speech-synthesized video and added to the speech-synthesized video.

12. The method according to claim 11, characterized in that, The audio in the audio file is music audio, and the music audio contains multiple audio segments; The process of determining the key audio segments in the audio file includes: Obtain the target audio segment from the plurality of audio segments, and determine the audio segments other than the target audio segment from the plurality of audio segments as the audio segments to be matched; Determine the segment similarity between the target audio segment and the audio segment to be matched; The similarity of the segments determines the target audio segments that are similar to the target audio segment, and the number of similar segments to the target audio segment is counted. If the number of similar segments meets the key audio determination condition, then the target audio segment is taken as the key audio segment.

13. The method according to claim 1, characterized in that, The video clip library contains multiple video clips, including historical video clips of the target historical delivery video; the target historical delivery video refers to the video that meets the advertising delivery indicators selected from the historical delivery videos delivered on the advertising delivery platform related to the advertising business. The method further includes: Perform compliance checks on each video frame in the target's historical video feed to obtain compliance check results; If the compliance test result indicates that there is an illegal video frame in each of the video frames, then the illegal video frames are filtered from the target historical video to obtain the target historical filtered video; Clean the subtitle information from the video frame of the target historical filtered video, and use the target historical filtered video after the subtitle information is cleaned as the target historical cleaned video. Identify the video scene of the target historical cleaning video, segment the target historical cleaning video based on the identified video scene to obtain video segment segments, and add the video segment segments as the historical video segments to the video segment library.

14. An advertising video generation device, characterized in that, include: The speech recognition module is used to acquire advertising text materials for advertising business. When converting the material content in the advertising text materials into advertising speech content, the module performs speech recognition on the advertising speech content to obtain the speech recognition text of the advertising speech content. The text segmentation module is used to segment the speech recognition text to obtain text segmentation segments. In the advertising speech content, the module determines the speech segmentation segment corresponding to the text segmentation segment, determines the segmentation tag of the speech segmentation segment through the segmentation content of the text segmentation segment, and determines the segmentation duration of the speech segmentation segment. The video clip filtering module is used to filter video clips that match the advertising business from multiple video clips contained in the video clip library based on the clip segmentation tags and the clip segmentation duration. The video delivery generation module is used to determine the advertising video material for the advertising business based on the selected video clips, and to generate a target delivery video for the advertising business based on the advertising video material and the advertising voice content.

15. A computer program product, characterized in that, The computer program product includes a computer program stored in a computer-readable storage medium, the computer program being adapted to be read and executed by a processor to cause a computer device having the processor to perform the method of any one of claims 1-13.