Talking Head Generation Driven by Speech-Related Facial Action Units and Audio- Based on Multimodal Representation Fusion

TL;DR

A novel generative framework that contains a dilated non-causal temporal convolutional self-attention network as a multimodal fusion module to promote the relationship learning of cross-modal features and outperforms existing methods in terms of both image quality and lip-sync accuracy.

Abstract

Talking head generation is to synthesize a lip-synchronized talking head video by inputting an arbitrary face image and corresponding audio clips. Existing methods ignore not only the interaction and relationship of cross-modal information, but also the local driving information of the mouth muscles. In this study, we propose a novel generative framework that contains a dilated non-causal temporal convolutional self-attention network as a multimodal fusion module to promote the relationship learning of cross-modal features. In addition, our proposed method uses both audio- and speech-related facial action units (AUs) as driving information. Speech-related AU information can guide mouth movements more accurately. Because speech is highly correlated with speech-related AUs, we propose an audio-to-AU module to predict speech-related AU information. We utilize pre-trained AU classifier to ensure that the generated images contain correct AU information. We verify the effectiveness of the proposed model on the GRID and TCD-TIMIT datasets. An ablation study is also conducted to verify the contribution of each component. The results of quantitative and qualitative experiments demonstrate that our method outperforms existing methods in terms of both image quality and lip-sync accuracy.

Figures (13)

• Figure 1: Proposed talking head generation system, which uses audio and speech-related AU information to drive a face image to generate speech video.
• Figure 2: Pipeline of our proposed system. The TCSAN module can effectively integrate multimodal features, and the audio-to-AU module can obtain speech-related facial AUs as local driving information of the mouth movement.
• Figure 3: Interpretation of speech-related AUs
• Figure 4: Proposed dilated non-causal temporal convolutional self-attention network. "Res attention" means the residual attention block.
• Figure 5: Residual attention block in TCSAN. In the multi-head self-attention module, Q = K = V.