Research map / Video / Video generation
StreamTalk: Streaming Co-Speech Gesture Generation with Key-Pose Anchoring
Jianfang Li, Xiangyue Zhang, Jiaxu Zhang, Kaixing Yang, Steven Hoi
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 SPG key-pose anchoring at inference on top of SAM (ablation) | StreamTalk · BEAT2 1-Speaker | FGD: 0.503 -> 0.383 | Improved |
| add SAM alone without SPG (ablation) | StreamTalk · BEAT2 1-Speaker | FGD: 0.478 -> 0.503 | Hurt |
| add SPG alone (ablation) | StreamTalk · BEAT2 1-Speaker | FGD: 0.478 -> 0.455 | Improved |
| add SAM and SPG (ablation) | StreamTalk · BEAT2 All-Speakers | FGD: 0.391 -> 0.293 | Improved |
| use retrieved anchor instead of random anchor (ablation) | StreamTalk · BEAT2 1-Speaker | FGD: 0.673 -> 0.503 | Improved |
| use flow-matching refinement instead of linear refinement (ablation) | StreamTalk · BEAT2 1-Speaker | FGD: 0.471 -> 0.383 | Improved |
| use 1 tail key pose instead of 8 (ablation) | StreamTalk · BEAT2 Speaker-2 | FGD: 0.575 -> 0.383 | Improved |
| use 1 tail key pose instead of 4 (ablation) | StreamTalk · BEAT2 Speaker-2 | FGD: 0.443 -> 0.383 | Improved |
| place anchor at tail instead of middle (ablation) | StreamTalk · BEAT2 Speaker-2 | FGD: 0.408 -> 0.383 | Improved |
| use 3-branch Part-aware DiT instead of single branch (ablation) | StreamTalk · BEAT2 Speaker-2 | FGD: 0.612 -> 0.478 | Improved |
| add SPG to reduce self-intersection (ablation) | StreamTalk · BEAT2 | self-intersection frames: 387 -> 39 | Improved |
| scale database 20x (ablation) | StreamTalk · Efficiency | FPS: 76 -> 36 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 5 of 5 on most shared results.
- SemTalk — SemTalk: Holistic Co-speech Motion Generation with Frame-level Semantic Emphasis (ICCV 2025) · better on 2 of 2 shared results
- ReMoDiffuse — ReMoDiffuse: Retrieval-Augmented Motion Diffusion Model (ICCV 2023) · better on 2 of 2 shared results
- SemGes — SemGes: Semantics-aware Co-Speech Gesture Generation using Semantic Coherence and Relevance Learning (ICCV 2025) · better on 1 of 1 shared result
- GestureLSM — GestureLSM: Latent Shortcut based Co-Speech Gesture Generation with Spatial-Temporal Modeling (ICCV 2025) · better on 1 of 1 shared result
- MambaTalk — MambaTalk: Efficient Holistic Gesture Synthesis with Selective State Space Models (NeurIPS 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Co-Speech Gesture Video Generation with Implicit Motion-Audio Entanglement (CVPR 2025 · similarity 0.89)
- Co$^{\mathbf{3}}$Gesture: Towards Coherent Concurrent Co-speech 3D Gesture Generation with Interactive Diffusion (ICLR 2025 · similarity 0.88)
- DiffSHEG: A Diffusion-Based Approach for Real-Time Speech-driven Holistic 3D Expression and Gesture Generation (CVPR 2024 · similarity 0.88)
- TANGO: Co-Speech Gesture Video Reenactment with Hierarchical Audio Motion Embedding and Diffusion Interpolation (ICLR 2025 · similarity 0.87)
- Co-Speech Gesture Video Generation via Motion-Decoupled Diffusion Model (CVPR 2024 · similarity 0.87)
- ConvoFusion: Multi-Modal Conversational Diffusion for Co-Speech Gesture Synthesis (CVPR 2024 · similarity 0.87)
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.
