Research map / Video / Video generation
TARO: Timestep-Adaptive Representation Alignment with Onset-Aware Conditioning for Synchronized Video-to-Audio Synthesis
Tri Ton, Ji Woo Hong, Chang D. Yoo
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 Timestep-Adaptive Representation Alignment (TRA) (ablation) | TARO · VGGSound test | FAD: 1.24 -> 0.78 | Improved |
| add Onset-Aware Conditioning (OAC) (ablation) | TARO · VGGSound test | CLIP: 13.73 -> 14.05 | Improved |
| add both TRA and OAC vs neither (ablation) | TARO · VGGSound test | FD: 0.55 -> 0.47 | Improved |
| add OAC to TRA-only model (ablation) | TARO · VGGSound test | FAD: 0.78 -> 0.94 | Hurt |
| add timestep-adaptive weighting to alignment loss (ablation) | TRA · VGGSound test | FD: 0.51 -> 0.47 | Improved |
| use convolution projection instead of global pooling for sequence-length mismatch (ablation) | TRA · VGGSound test | FAD: 1.16 -> 0.94 | Improved |
| use convolution projection instead of interpolation (ablation) | TRA · VGGSound test | FAD: 1.09 -> 0.94 | Improved |
| use EAT audio encoder instead of wav2vec 2.0 (ablation) | TRA · VGGSound test | FAD: 1.41 -> 0.94 | Improved |
| use EAT instead of BEATs encoder (ablation) | TRA · VGGSound test | FAD: 0.74 -> 0.94 | Mixed |
| inject encoder features at 4th block instead of 2nd (ablation) | TRA · VGGSound test | FAD: 1.49 -> 0.94 | Improved |
| inject at later blocks (8th) instead of 4th (ablation) | TRA · VGGSound test | CLIP: 14.10 -> 13.73 | Hurt |
| use onset cues instead of energy control or down-sampled mel (ablation) | TARO · VGGSound test | Alignment accuracy (%): 96.10 -> 97.19 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Frieren — Frieren: Efficient Video-to-Audio Generation Network with Rectified Flow Matching (NeurIPS 2024) · better on 5 of 6 shared results
- Diff-Foley — Diff-Foley: Synchronized Video-to-Audio Synthesis with Latent Diffusion Models (NeurIPS 2023) · better on 1 of 1 shared result
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
- Native Audio-Visual Alignment for Generation (NeurIPS 2026 · similarity 0.86)
- MMAudio: Taming Multimodal Joint Training for High-Quality Video-to-Audio Synthesis (CVPR 2025 · similarity 0.85)
- Model-Guided Dual-Role Alignment for High-Fidelity Open-Domain Video-to-Audio Generation (NeurIPS 2025 · similarity 0.84)
- CAV-MAE Sync: Improving Contrastive Audio-Visual Mask Autoencoders via Fine-Grained Alignment (CVPR 2025 · similarity 0.83)
- Foley-Flow: Coordinated Video-to-Audio Generation with Masked Audio-Visual Alignment and Dynamic Conditional Flows (CVPR 2025 · similarity 0.83)
- Hallo2: Long-Duration and High-Resolution Audio-Driven Portrait Image Animation (ICLR 2025 · similarity 0.82)
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.
