AI Paper Atlas

Research map / Video / Video generation

AlignSep: Temporally-Aligned Video-Queried Sound Separation with Flow Matching

Xize Cheng, Chenyuhao Wen, Slytherin Wang, Yongqi Wang, Zehan Wang, Rongjie Huang, Tao Jin, Zhou Zhao

ICLR 2026 · Video › Video generation · approach: Video-to-audio synthesis

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 changedWhereMeasuredEffect
add flow matching instead of diffusion (ablation)AlignSep · VGGSound-HardTA-V (alignment acc): 92.28 -> 95.76Improved
add flow matching instead of diffusion (ablation)AlignSep · VGGSound-CleanSA-A (CLAP): 64.12 -> 73.38Improved
add CAVP temporal visual encoder (ablation)AlignSep · VGGSound-HardTA-V (alignment acc): 76.27 -> 95.76Improved
add CAVP temporal visual encoder (ablation)AlignSep · VGGSound-CleanTA-V (alignment acc): 94.71 -> 96.88Improved
use concat fusion instead of cross-attention (ablation)AlignSep · VGGSound-HardTA-V (alignment acc): 73.38 -> 95.76Improved
use concat fusion instead of cross-attention (ablation)AlignSep · VGGSound-CleanSA-A (CLAP): 70.13 -> 73.38Improved
increase denoising steps from 5 to 25 (ablation)AlignSep · VGGSound-HardTA-V (alignment acc): 88.14 -> 95.76Improved
increase denoising steps from 25 to 100 (ablation)AlignSep · VGGSound-HardTA-V (alignment acc): 95.76 -> 93.22No clear effect
increase denoising steps from 25 to 100 (ablation)AlignSep · VGGSound-Hardinference throughput (FPS): 2.17 -> 0.72Hurt
use flow matching instead of rectified flow (100 steps) (ablation)AlignSep · VGGSound-CleanSA-A (CLAP): 57.36 -> 73.64Improved
increase reference video frame rate from 0.25 to 4 FPS (ablation)AlignSep · VGGSound-Hardalignment accuracy: 0.76 -> 0.95Improved
increase reference video frame rate (semantic-only baseline) (ablation)CLIPSep · VGGSound-Hardalignment accuracy: 0.81No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy