Research map / 3D vision / Face and shape
TalkingGaussian: Structure-Persistent 3D Talking Head Synthesis via Gaussian Splatting
Jiahe Li, Jiawei Zhang, Xiao Bai, Jin Zheng, Xin Ning, Jun Zhou, Lin Gu
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 |
|---|---|---|---|
| use deformation instead of appearance modification on the 3DGS backbone (ablation) | TalkingGaussian · Self-reconstruction | LPIPS: 0.0355 -> 0.0290 | Improved |
| use deformation instead of appearance modification on NeRF Tri-Hash backbone (ablation) | ER-NeRF Tri-Hash · Self-reconstruction | PSNR: 33.14 -> 31.50 | Hurt |
| add face-mouth decomposition to 3DGS deformation model (ablation) | TalkingGaussian · Self-reconstruction | Sync-C: 5.676 -> 6.516 | Improved |
| add face-mouth decomposition to 3DGS appearance modification model (ablation) | 3DGS appearance · Self-reconstruction | Sync-C: 6.001 -> 6.332 | Improved |
| add incremental sampling to deformation + FMD (ablation) | TalkingGaussian · Self-reconstruction | SSIM: 0.906 -> 0.910 | Improved |
| additionally predict opacity adjustment alongside deformation (ablation) | TalkingGaussian · Self-reconstruction | SSIM: 0.910 -> 0.908 | No clear effect |
| additionally predict RGB color per primitive (ablation) | TalkingGaussian · Self-reconstruction | SSIM: 0.910 -> 0.907 | Hurt |
| use HuBERT instead of DeepSpeech audio features (ablation) | TalkingGaussian · Self-reconstruction | Sync-C: 6.516 -> 6.667 | Improved |
| use TalkingGaussian instead of ER-NeRF | talking head synthesis · Self-reconstruction | Training time: 2.1h -> 0.5h | Improved |
| use TalkingGaussian instead of ER-NeRF on cross-gender audio | talking head synthesis · Lip-sync, Test Audio A, May | Sync-C: 3.639 -> 4.774 | Improved |
| use TalkingGaussian instead of IP-LAP on PSNR | talking head synthesis · Self-reconstruction | PSNR: 35.34 -> 33.61 | Hurt |
Later papers that compared with this method
6 later accepted papers report TalkingGaussian as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- An Interactive Conversational 3D Virtual Human (IJCV 2026 · better on 4 of 4 shared results)
- EmoTaG: Emotion-Aware Talking Head Synthesis on Gaussian Splatting with Few-Shot Personalization (CVPR 2026 · better on 1 of 1 shared result)
- DGTalker: Disentangled Generative Latent Space Learning for Audio-Driven Gaussian Talking Heads (ICCV 2025 · better on 3 of 5 shared results)
- InsTaG: Learning Personalized 3D Talking Head from Few-Second Video (CVPR 2025 · better on 3 of 3 shared results)
- VASA-3D: Lifelike Audio-Driven Gaussian Head Avatars from a Single Image (NeurIPS 2025 · better on 2 of 3 shared results)
- SyncTalk++: High-Fidelity and Efficient Synchronized Talking Heads Synthesis Using Gaussian Splatting (TPAMI 2025 · better on 2 of 3 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- RAD-NeRF — Rad-NeRF: Ray-decoupled Training of Neural Radiance Field (NeurIPS 2024) · better on 2 of 2 shared results
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
- EmoTaG: Emotion-Aware Talking Head Synthesis on Gaussian Splatting with Few-Shot Personalization (CVPR 2026 · similarity 0.92)
- PointTalk: Audio-Driven Dynamic Lip Point Cloud for 3D Gaussian-based Talking Head Synthesis (AAAI 2025 · similarity 0.91)
- Hierarchically Controlled Deformable 3D Gaussians for Talking Head Synthesis (AAAI 2025 · similarity 0.91)
- SyncTalk: The Devil is in the Synchronization for Talking Head Synthesis (CVPR 2024 · similarity 0.89)
- FaceTalk: Audio-Driven Motion Diffusion for Neural Parametric Head Models (CVPR 2024 · similarity 0.89)
- InsTaG: Learning Personalized 3D Talking Head from Few-Second Video (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.
