Research map / Video / Video generation
VoiceCraft-Dub: Automated Video Dubbing with Neural Codec Language Models
Kim Sung-Bin, Jeongsoo Choi, Puyuan Peng, Joon Son Chung, Tae-Hyun Oh, David Harwath
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-visual fusion layers instead of prepending video tokens (ablation) | VoiceCraft-Dub · LRS3 | WER: 2.81 -> 1.38 | Improved |
| add audio-visual fusion layers instead of prepending video tokens (ablation) | VoiceCraft-Dub · LRS3 | LSE-D: 7.58 -> 6.59 | Improved |
| share positional encoding between speech and video tokens (ablation) | prepend-video baseline · LRS3 | WER: 2.85 -> 2.81 | No clear effect |
| add facial expression features alongside lip features (ablation) | VoiceCraft-Dub · LRS3 | UTMOS: 3.66 -> 3.86 | Improved |
| add facial expression features alongside lip features (ablation) | VoiceCraft-Dub · LRS3 | WER: 1.38 -> 1.68 | Hurt |
| add face input on expressive data (ablation) | VoiceCraft-Dub · CelebV-Dub | emoSIM: 0.778 -> 0.782 | Mixed |
| use VoiceCraft-Dub in place of HPMDubbing | automated dubbing · LRS3 | MOS naturalness: 3.12 -> 4.30 | Improved |
| use VoiceCraft-Dub in place of HPMDubbing | automated dubbing · LRS3 | LSE-D: 6.58 -> 6.59 | Hurt |
| LSE-D as proxy for human lip-sync judgments | evaluation metric · CelebV-Dub subset (50 samples) | correlation coefficient with MOS: 0.36 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports VoiceCraft-Dub as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- JAM-Flow: Joint Audio-Motion Synthesis with Flow Matching (ECCV 2026 · 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
- From Faces to Voices: Learning Hierarchical Representations for High-quality Video-to-Speech (CVPR 2025 · similarity 0.87)
- EmoDubber: Towards High Quality and Emotion Controllable Movie Dubbing (CVPR 2025 · similarity 0.87)
- Prosody-Enhanced Acoustic Pre-training and Acoustic-Disentangled Prosody Adapting for Movie Dubbing (CVPR 2025 · similarity 0.87)
- RVCBench: Benchmarking Robustness of Voice Cloning Across Modern Audio Generation Models (NeurIPS 2026 · similarity 0.86)
- MMAudio: Taming Multimodal Joint Training for High-Quality Video-to-Audio Synthesis (CVPR 2025 · similarity 0.86)
- Vevo: Controllable Zero-Shot Voice Imitation with Self-Supervised Disentanglement (ICLR 2025 · similarity 0.86)
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.
