Literature Review of Audio-Driven 2D Avatar Video Generation Algorithms

Audio-driven 2D avatar video generation algorithms have a wide range of applications in the media field. The technology of generating 2D avatar videos with only the input of compliant audio and images has been a positive boost to the development of online media and other fields. In such generation algorithms, the accurate coupling of speech audio and appearance changes such as faces and gestures in subtle movements has been a point of continuous improvement, with appearance changes moving from an early focus on matching speech content only to starting to incorporate human emotions expressed by speech. There has been a significant improvement in fidelity and synchronization compared to the early experimental results of the study, and the behavioral performance of the 2D avatars in the generated videos is getting closer to that of humans. This paper provides an overview of existing audio-driven 2D avatar generation algorithms and classifies their tasks into two categories: talking face generation and co-speech gesture generation. Firstly, the article describes the task specifically and describes its application areas. Secondly, we analyze the core algorithms in order of technological advancement and briefly describe the performance effects of the methods or models. Thirdly, we present common datasets for both types of tasks as well as evaluation metrics and compare the performance metrics of some recently proposed algorithms. Finally, the paper discusses the opportunities and challenges faced by the field and gives future research directions.