Core Concepts
Innovative approach for emotional 3D talking heads generation using speech-driven techniques.
Abstract
The content discusses the development of EmoVOCA, a dataset for emotional 3D talking heads generation. It introduces a novel method combining speech and expression deformations to create realistic animations. The study compares different models and datasets, showcasing superior results with the proposed approach.
Structure:
Introduction to Emotional 3D Talking Heads Generation
Challenges in Existing Methods
Proposed Approach: EmoVOCA Dataset Creation
Double Encoder/Shared Decoder Architecture Explanation
Experimental Results and Evaluation Metrics
Comparison with State-of-the-Art Models (Faceformer, S2L+S2D)
User-Based Studies Comparing E-S2L+S2D with EmoTalk and EMOTE
Conclusions and Future Directions
Stats
Fig. 1: Method generates emotional 3D talking heads.
arXiv:2403.12886v1 [cs.CV] 19 Mar 2024