Research map / Video / Video generation
AI killed the Video Star. Audio-Driven Diffusion Model for Expressive Talking Head Generation
Baptiste Chopin, Tashvik Dhamija, Pranav Balaji, Yaohui Wang, Antitza Dantcheva
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 text input to Wav2Vec audio features (ablation) | Dimitra · VoxCeleb2 | F-LMD: 6.16 -> 5.79 | Improved |
| add phoneme input to Wav2Vec audio features (ablation) | Dimitra · VoxCeleb2 | F-LMD: 6.16 -> 6.06 | Improved |
| use three separate models for lips, expression and head pose instead of a single model (ablation) | Dimitra · VoxCeleb2 | F-LMD: 6.15 -> 5.72 | Improved |
| replace Wav2Vec+text+phoneme with WavLM features and new training scheme (Dimitra++) (ablation) | Dimitra · VoxCeleb2 | F-LMD: 5.72 -> 5.68 | Improved |
| replace Wav2Vec+text+phoneme with WavLM (Dimitra++) (ablation) | Dimitra · VoxCeleb2 | AHD: 16.50 -> 12.58 (GT 18.91) | Mixed |
| add audio encoding before each transformer layer instead of cross attention (ablation) | Dimitra++ · VoxCeleb2 | F-LMD: 6.23 -> 5.68 | Improved |
| add audio encoding before each layer instead of only once (ablation) | Dimitra++ · VoxCeleb2 | F-LMD: 5.98 -> 5.68 | Improved |
| add audio encoding by addition instead of concatenation (ablation) | Dimitra++ · VoxCeleb2 | F-LMD: 6.02 -> 5.68 | Improved |
| add conditioning on the first pose (ablation) | Dimitra++ · VoxCeleb2 | F-LMD: 6.06 -> 5.68 | Improved |
| use 8 attention heads (default) instead of 4 (ablation) | Dimitra++ · VoxCeleb2 | F-LMD: 5.95 -> 5.68 | Improved |
| use default heads instead of 16 attention heads (ablation) | Dimitra++ · VoxCeleb2 | M-LMD: 4.35 -> 4.36 | No clear effect |
| Dimitra++ vs Dimitra on CelebV-HQ | Dimitra · CelebV-HQ | FID: 67.48 -> 64.51 | Improved |
Closest papers by meaning
- FaceTalk: Audio-Driven Motion Diffusion for Neural Parametric Head Models (CVPR 2024 · similarity 0.92)
- Diffusion-based Realistic Listening Head Generation via Hybrid Motion Modeling (CVPR 2025 · similarity 0.91)
- EmotiveTalk: Expressive Talking Head Generation through Audio Information Decoupling and Emotional Video Diffusion (CVPR 2025 · similarity 0.91)
- MoDiTalker: Motion-Disentangled Diffusion Model for High-Fidelity Talking Head Generation (AAAI 2025 · similarity 0.91)
- DEEPTalk: Dynamic Emotion Embedding for Probabilistic Speech-Driven 3D Face Animation (AAAI 2025 · similarity 0.91)
- DAWN: Dynamic Frame Avatar with Non-autoregressive Diffusion Framework for Talking head Video Generation (ICLR 2025 · similarity 0.90)
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.
