面向教育内容生成的动画多镜头音视频数据集构建

doi:10.11959/j.issn.2096-0271.BDR26166

Chinese

您当前的位置：

首页 >

文章列表页 >

面向教育内容生成的动画多镜头音视频数据集构建

专题 | 更新时间：2026-09-20

- 面向教育内容生成的动画多镜头音视频数据集构建
- Construction of an animated multi-shot audio-visual dataset for educational content generation
- 大数据 2026年
- 作者机构：
- 作者简介：
- 基金信息：
- DOI：10.11959/j.issn.2096-0271.BDR26166
  中图分类号： TP391.41
- 收稿：2026-04-30，
  
  修回：2026-09-15，
  
  录用：2026-09-20，
- 稿件说明：
移动端阅览
面向教育内容生成的动画多镜头音视频数据集构建[J/OL]. 大数据, 2026.

Construction of an animated multi-shot audio-visual dataset for educational content generation[J/OL]. Big Data Research, 2026.
面向教育内容生成的动画多镜头音视频数据集构建[J/OL]. 大数据, 2026. DOI： 10.11959/j.issn.2096-0271.BDR26166.

Construction of an animated multi-shot audio-visual dataset for educational content generation[J/OL]. Big Data Research, 2026. DOI： 10.11959/j.issn.2096-0271.BDR26166.

摘要

面向情景化教育视频生成，现有音视频数据集在多镜头结构、动画域适配和音画一致性方面仍存在不足。构建动画多镜头音视频数据集 Edu-Anime-MultiShot，并形成“场景—镜头”层次组织、分阶段质量过滤和“全局—局部”结构化标注相结合的半自动构建方法。方法重点设计动画多镜头数据治理规则与双阈值门控，并通过场景级角色原型和统一角色 ID 约束降低跨镜头指代漂移。与 TalkCuts 相比，该数据集规模较小，但增加双层结构、对白—角色映射和角色音色信息。人工抽样显示，画外音比例由 24.6% 降至 1.2% 以内。结果表明，该流程可改善数据结构化程度和音画一致性，为教育内容生成提供可复用的数据治理方案。

Abstract

Contextualized educational video generation requires training data with multi-shot structure

animation-domain adaptation

and reliable audio-visual correspondence. To address these requirements

Edu-Anime-MultiShot was constructed

and a semi-automatic construction method integrating scene-shot hierarchical organization

staged quality filtering

and global-local structured annotation was developed. Instead of modifying backbone models

task-specific governance rules and threshold design were emphasized. Visual-subject coverage within each speech interval was used as a local constraint

and the proportion of qualified speech duration was used as a global constraint; a two-level gate was adopted to balance sample retention and audio-visual consistency. Scene-level character prototypes and shot-level unified character IDs were used to reduce cross-shot reference drift. Compared with TalkCuts

which contains more than 500 h of multi-shot human speech videos

the constructed dataset is smaller but provides a scene-shot hierarchy

dialogue-character mapping

and character-level voice attributes for animated educational scenarios. Manual sampling showed that off-screen speech decreased from 24.6% in raw candidates to below 1.2% after governance. The resulting dataset provides a reusable data-governance and structured-annotation framework for educational animation generation

while the platform results should be interpreted as observational evidence rather than causal proof of learning improvement.

关键词

Keywords

references

Wang S , Xu T L , Li H , et al . Large language models for education: A survey and outlook [J ] . arXiv preprint arXiv: 2403.18105 , 2024 .

Kasneci E , Sessler K , Küchemann S , et al . ChatGPT for good? On opportunities and challenges of large language models for education [J ] . Learning and Individual Differences , 2023 , 103 : 102274 .

Vajjala S , et al . Exploring the role of generative AI in second language education: Insights for instruction, learning, and assessment [J ] . Language Learning & Technology , 2024 , 28 ( 1 ): 1 - 15 .

Xue H , Hang T , Zeng Y , et al . Advancing high-resolution video-language representation with large-scale video transcriptions [C ] // Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 2022 : 5036 - 5045 .

Wang Y , He Y , Li Y , et al . Internvid: A large-scale video-text dataset for multimodal understanding and generation [J ] . arXiv preprint arXiv: 2307.06942 , 2023 .

Chen T S , Siarohin A , Menapace W , et al . Panda-70m: Captioning 70m videos with multiple cross-modality teachers [C ] // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 2024 : 13320 - 13331 .

Chen Z , Liu H , Yu W , et al . M3AV: A multimodal, multigenre, and multipurpose audio-visual academic lecture dataset [C ] // Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) . 2024 : 9041 - 9060 .

Rossetto L , Bailer W , Dang-Nguyen D T , et al . The castle 2024 dataset: Advancing the art of multimodal understanding [C ] // Proceedings of the 33rd ACM International Conference on Multimedia . 2025 : 12629 - 12635 .

Chen J , Wang Z , Zeng A , et al . Talkcuts: A large-scale dataset for multi-shot human speech video generation [J ] . arXiv preprint arXiv: 2510.07249 , 2025 .

Soucek T , Lokoc J . Transnet v2: An effective deep network architecture for fast shot transition detection [C ] // Proceedings of the 32nd ACM International Conference on Multimedia . 2024 : 11218 - 11221 .

Girdhar R , El-Nouby A , Liu Z , et al . Imagebind: One embedding space to bind them all [C ] // Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 2023 : 15180 - 15190 .

Lin B , Zhu B , Ye Y , et al . Video-LLaVA: Learning united visual representation by alignment before projection [J ] . arXiv preprint arXiv: 2311.10122 , 2023 .

Yang Z , Yu Y , Zhao Y , et al . Timeexpert: An expert-guided video llm for video temporal grounding [C ] // Proceedings of the IEEE/CVF International Conference on Computer Vision . 2025 : 24286 - 24296 .

Wang Q , Shi X , Li B , et al . Multishotmaster: A controllable multi-shot video generation framework [J ] . arXiv preprint arXiv: 2512.03041 , 2025 .

Radford A , Kim J W , Xu T , et al . Robust speech recognition via large-scale weak supervision [C ] // International conference on machine learning . PMLR , 2023 : 28492 - 28518 .

Redmon J , Divvala S , Girshick R , et al . You only look once: Unified, real-time object detection [C ] // Proceedings of the IEEE conference on computer vision and pattern recognition . 2016 : 779 - 788 .

曾志平 , 张旭龙 , 瞿晓阳 , 等 . 深度伪造音频生成与鉴伪技术综述 [J ] . 大数据 , 2025 , 11 ( 5 ): 130 - 151 .

Zeng Z , Zhang X , Qu X , et al . Review of deepfake audio generation and detection technologies [J ] . Big Data Research , 2025 , 11 ( 5 ): 130 - 151 .

Chen L , Li J , Dong X , et al . ShareGPT4Video: Improving video understanding and generation with better captions [J ] . arXiv preprint arXiv: 2406.01463 , 2024 .

Wang S , Xu T , Li H , et al . Large language models for education: A survey and outlook [J ] . IEEE Signal Processing Magazine , 2026 , 42 ( 6 ): 51 - 63 .

Li W , Wang F , Mayer R E . How to guide learners’ processing of multimedia lessons with pedagogical agents [J ] . Learning and Instruction , 2023 , 84 : 101729 .

Carlotto T , Jaques P A . The effects of animated pedagogical agents in an English-as-a-foreign-language learning environment [J ] . International Journal of Human-Computer Studies , 2016 , 95 : 15 - 26 .

Li W , Wang F , Mayer R E , et al . Animated pedagogical agents enhance learning outcomes and brain activity during learning [J ] . Journal of Computer Assisted Learning , 2022 , 38 ( 3 ): 621 - 637 .

Davis R O . The impact of pedagogical agent gesturing in multimedia learning environments: A meta-analysis [J ] . Educational Research Review , 2018 , 24 : 193 - 209 .

Chung J S , Zisserman A . Out of time: automated lip sync in the wild [C ] // Asian conference on computer vision . Cham : Springer International Publishing , 2016 : 251 - 263 .

Bai J , Bai S , Yang S , et al . Qwen-VL: A versatile vision-language model for understanding, localization, text reading, and beyond [J ] . arXiv preprint arXiv: 2308.12966 , 2023 .

Bai S , Chen K , Liu X , et al . Qwen2.5-VL Technical Report [J ] . arXiv preprint arXiv: 2502.13923 , 2025 .

Xu J , Guo Z , He J , et al . Qwen2.5-Omni Technical Report [J ] . arXiv preprint arXiv: 2503.20215 , 2025 .

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

面向成本收益协同优化的高质量数据准备方法

本体驱动的电信行业多模态数据集构建与治理方法

面向人工影响天气的机载多源微物理数据集构建与泄露控制评估

企业数据爬取行为的法律风险与规制路径——基于权益均衡的视角

基于湖仓一体技术的“多湖一中台”工业大数据平台研究与应用实践