A video course enhancement technique utilizing generated talking heads

Download PDF
Open Webpage

Zixiang Lu, Bujia Tian, Ping Gao, Qiguang Miao, Kun Xie, Ruyi Liu, Yining Quan

Published: 01 Jul 2025, Last Modified: 06 Nov 2025Neural Computing and ApplicationsEveryoneRevisionsCC BY-SA 4.0
Abstract: In the field of intelligent education, course videos integrate an instructor’s image, voice, and instructional content. These videos play a vital role in the teaching and learning process, but their production requires a substantial investment of time and effort from instructors. To reduce the stress experienced by instructors during lesson preparation, we propose a course video generation solution that is specifically designed to efficiently produce high-quality educational videos. Unlike traditional methods, our solution requires only the instructor to record a video of the course screen and provide an image of themselves, such as a photo or short video. Our method can be divided into two modules: the talking head generation module and the fusion module for generated video and screen-recorded video. In regard to the crucial talking head generation module, existing methods have certain limitations in terms of clarity and naturalness. Consequently, there is a need for further improvement in this area. To address these problems, we propose a transformer-based network to generate talking head videos with the instructor’s image and then combine screen-recorded video and talking head video to obtain the final course video. We performed separate comparison experiments and ablation experiments on the talking head videos obtained using our proposed method. In the comparison experiments, we compared different methods, all of which yielded better results. In the ablation experiments, we presented the methods we used as well as some optimization modules and compared them in terms of objective metrics. Moreover, we also conducted survey experiments with instructors and students to demonstrate the effectiveness of the generated videos using our framework. In addition, we have developed an application through a series of processes using our framework. The application has been deployed on the smart education platform of our university, which has received good feedback.