Research map / Image generation / Autoregressive and latent generation
LivelySpeaker: Towards Semantic-Aware Co-Speech Gesture Generation
Yihao Zhi, Xiaodong Cun, Xuelin Chen, Xi Shen, Wen Guo, Shaoli Huang, Shenghua Gao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add diffusion rhythm-aware refinement to semantic-aware generator (ablation) | SAG · TED Gesture | FGD: 56.878 -> 8.446 | Improved |
| use FFT-synthesized data beat alignment network instead of diffusion RAG (ablation) | SAG · TED Gesture | FGD: 56.878 -> 65.718 | Hurt |
| replace MLP denoiser with transformer decoder (ablation) | RAG · TED Gesture | BC: 0.656 -> 0.418 | Hurt |
| use 1D conv waveform audio encoder instead of Mel-spectrogram 2D conv (ablation) | RAG · TED Gesture | FGD: 4.951 -> 2.152 | Improved |
| add KL divergence loss on style embedding (ablation) | RAG · TED Gesture | FGD: 5.256 -> 2.152 | Improved |
| use SiLU activation instead of ReLU or LeakyReLU (ablation) | RAG MLP blocks · TED Gesture | FGD: 3.956 -> 2.152 | Improved |
| use 4 MLP layers instead of 2, 6 or 8 (ablation) | RAG · TED Gesture | FGD: 3.047 -> 2.152 | Improved |
| increase classifier-free guidance weight w (ablation) | Ours Rhythm · TED Gesture | BC and Diversity: BC 0.656 -> 0.699; FGD 2.152 -> 6.622 | Mixed |
| add beat empowerment to SAG gestures in the full system vs rhythm-only | Ours Full vs Ours Rhythm · TED Gesture | FGD: 2.152 -> 11.310 | Hurt |
| two-stage semantic plus rhythm generator vs end-to-end baselines | LivelySpeaker · TED Gesture user study | naturalness preference: 42.7% for Ours Full vs 41.6% for S2G | Improved |
Closest papers by meaning
- Co$^{\mathbf{3}}$Gesture: Towards Coherent Concurrent Co-speech 3D Gesture Generation with Interactive Diffusion (ICLR 2025 · similarity 0.91)
- Retrieving Semantics from the Deep: an RAG Solution for Gesture Synthesis (CVPR 2025 · similarity 0.91)
- ConvoFusion: Multi-Modal Conversational Diffusion for Co-Speech Gesture Synthesis (CVPR 2024 · similarity 0.91)
- DiffSHEG: A Diffusion-Based Approach for Real-Time Speech-driven Holistic 3D Expression and Gesture Generation (CVPR 2024 · similarity 0.90)
- Co-Speech Gesture Video Generation with Implicit Motion-Audio Entanglement (CVPR 2025 · similarity 0.89)
- HOP: Heterogeneous Topology-based Multimodal Entanglement for Co-Speech Gesture Generation (CVPR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
