A Survey on Audio-Driven Talking Face Generation

In recent years, advancements in deep learning have driven the development of audio-driven talking face generation. This technology has widespread applications in virtual avatars, intelligent assistants, film and television special effects, remote education, digital human live streaming, and more. Audio-driven talking face generation methods synthesize synchronized talking videos based on input speech signals, ensuring that the mouth movements, facial expressions, and head poses of virtual characters naturally align with the speech content. This paper provides an overview of the major research progress in this field and analyzes the advantages and limitations of different techniques. Additionally, it introduces evaluation metrics used to assess model performance in detail. Finally, the paper summarizes current challenges and potential future research directions.