Research map / Video / Video generation
Audio-driven Talking Face Generation with Stabilized Synchronization Loss
Dogucan Yaman, Fevziye Irem Eyiokur, Leonard Bärmann, HAZIM KEMAL EKENEL, Alexander Waibel
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 pretrained frozen audio encoder instead of jointly trained audio encoder (ablation) | talking face generator with lip-sync loss · LRS2 test | FID: 12.25 -> 9.82 | Improved |
| use Wav2Vec2 audio encoder instead of sync-pretrained audio encoder (ablation) | talking face generator · LRS2 test | FID: 9.82 -> 10.56 | Hurt |
| add silent-lip generator to produce closed-lip identity references (ablation) | talking face generator · LRS2 test | LMD: 2.325 -> 1.741 | Improved |
| replace lip-sync loss with stabilized synchronization loss (ablation) | talking face generator · LRS2 test | LSE-C: 7.752 -> 8.370 | Improved |
| replace lip-sync loss with stabilized synchronization loss (ablation) | talking face generator · LRS2 test | FID: 8.16 -> 5.27 | Improved |
| add vanilla triplet loss (ablation) | talking face generator · LRS2 test | FID: 5.27 -> 6.24 | Hurt |
| add adaptive triplet loss (ablation) | talking face generator · LRS2 test | FID: 5.27 -> 4.02 | Improved |
| replace SyncNet with AVSyncNet (ablation) | talking face generator · LRS2 test | FID: 4.02 -> 3.83 | Improved |
| add VQFR face restoration post-processing (ablation) | full model · LRS2 test | FID: 3.83 -> 5.23 | Hurt |
| choose VQFR over GPEN and GFPGAN for post-processing (ablation) | full model · LRS2 test | FID: 58.77 -> 5.23 | Improved |
| use silent-lip generator instead of VideoReTalking for silent face generation (ablation) | silent identity reference generation · Silent face generation (LRS2) | FID: 33.60 -> 4.41 | Improved |
Closest papers by meaning
- SyncTalkFace: Talking Face Generation with Precise Lip-Syncing via Audio-Lip Memory (AAAI 2022 · similarity 0.93)
- Seeing What You Said: Talking Face Generation Guided by a Lip Reading Expert (CVPR 2023 · similarity 0.91)
- Perceptually Accurate 3D Talking Head Generation: New Definitions, Speech-Mesh Representation, and Evaluation Metrics (CVPR 2025 · similarity 0.90)
- SyncTalk: The Devil is in the Synchronization for Talking Head Synthesis (CVPR 2024 · similarity 0.90)
- OmniSync: Towards Universal Lip Synchronization via Diffusion Transformers (NeurIPS 2025 · similarity 0.88)
- DiVAS: Video and Audio Synchronization with Dynamic Frame Rates (CVPR 2024 · 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.
