Research map / 3D vision / Face and shape
MeshTalk: 3D Face Animation From Speech Using Cross-Modality Disentanglement
Alexander Richard, Michael Zollhöfer, Yandong Wen, Fernando de la Torre, Yaser Sheikh
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 audio to the encoder input (with l2 loss) (ablation) | MeshTalk latent learning · Latent space strategies (in-house test set) | autoregressive model perplexity: 1.853 -> 1.879 | No clear effect |
| add audio to the encoder input (with l2 loss) (ablation) | MeshTalk latent learning · Latent space strategies (in-house test set) | reconstruction error (mm): 1.156 -> 1.124 | No clear effect |
| add cross-modality loss on top of expression+audio encoder (ablation) | MeshTalk latent learning · Latent space strategies (in-house test set) | autoregressive model perplexity: 1.879 -> 1.669 | Improved |
| add cross-modality loss on top of expression+audio encoder (ablation) | MeshTalk latent learning · Latent space strategies (in-house test set) | reconstruction error (mm): 1.124 -> 1.244 | Hurt |
| use audio and expression encoder with cross-modality loss instead of expression-only encoder (ablation) | MeshTalk latent learning · Latent space strategies (in-house test set) | autoregressive model perplexity: 1.853 -> 1.669 | Improved |
| use categorical instead of continuous latent space (ablation) | MeshTalk · Latent space type (in-house test set) | vertex error (mm): 1.975 -> 1.244 | Improved |
| use categorical instead of continuous latent space (ablation) | MeshTalk · Latent space type (in-house test set) | lip error (mm): 4.578 -> 3.184 | Improved |
| use MeshTalk instead of VOCA | audio-driven face animation · Lip sync vs VOCA (in-house test set) | lip vertex error (mm): 3.720 -> 3.184 | Improved |
| replace DeepSpeech features in VOCA with Mel spectrograms and our audio encoder (ablation) | VOCA · Lip sync vs VOCA (in-house test set) | lip vertex error (mm): 3.720 -> 3.472 | Improved |
| use MeshTalk instead of VOCA | audio-driven face animation · Perceptual study, full-face | % ours better or equal: 75.3% | Improved |
| use MeshTalk instead of VOCA | audio-driven face animation · Perceptual study, upper face | % ours better or equal: 66.4% | Improved |
| compare MeshTalk with tracked ground truth | MeshTalk · Perceptual study, upper face | % ours better or equal: 31.5% | Hurt |
Later papers that compared with this method
2 later accepted papers report MeshTalk as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- EmoTalk3D: High-Fidelity Free-View Synthesis of Emotional 3D Talking Head (ECCV 2024 · better on 0 of 1 shared result)
- EmoTalk: Speech-Driven Emotional Disentanglement for 3D Face Animation (ICCV 2023 · better on 3 of 3 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
- Mimic: Speaking Style Disentanglement for Speech-Driven 3D Facial Animation (AAAI 2024 · similarity 0.91)
- Perceptually Accurate 3D Talking Head Generation: New Definitions, Speech-Mesh Representation, and Evaluation Metrics (CVPR 2025 · similarity 0.91)
- Probabilistic Speech-Driven 3D Facial Motion Synthesis: New Benchmarks Methods and Applications (CVPR 2024 · similarity 0.90)
- FaceTalk: Audio-Driven Motion Diffusion for Neural Parametric Head Models (CVPR 2024 · similarity 0.90)
- Seeing Speech: Learning Visible Articulatory Dynamics for Speech-Driven 3D Facial Animation (NeurIPS 2026 · similarity 0.89)
- FaceChain-ImagineID: Freely Crafting High-Fidelity Diverse Talking Faces from Disentangled Audio (CVPR 2024 · similarity 0.89)
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.
