A Survey on Talking Head Generation

In recent years, the advancement of virtual digital human technology has been significantly accelerated by deep learning-based generative techniques. This paper offers a comprehensive review of the current hot topic in virtual digital human research: talking head generation. It emphasizes the promising applications of this technology in domains s such as film dubbing, animation production, and virtual assistants. From the perspectives of dataset availability, key technologies, and evaluation strategies, this paper presents a systematic overview and summary of the current state of talking head generation technology and research. It introduces pivotal artificial intelligence technologies involved in the generation process, encompassing visual generation, image recognition, speech recognition, and cross-modal analysis, along with their progressive developments. The paper identifies pressing issues that requires attention in this field, such as data, models, and evaluation metrics, and offers a future outlook based on these challenges. Its objective is to provide insightful guidance and promote the advancement of this field for researchers in the domain.