The team constructed SyncReward-Data-80K using a three-level taxonomy spanning four sound-event families: speech, onset events such as impacts, instruments, and ambient sound.
Human-rated clips improve rewards for synchronizing generated sound and video
SyncReward learned from professional ratings to evaluate synchronization and guide selection or post-training of audio-visual generators.
Chinese Tech
Kaisi Guan · Ying Ba · Wenjing Wang · Xihua Wang · Chen Li · Jing Lyu · +1 more
Renmin University of China · WeChat, Tencent Inc.
Research Digest··2 min read
The authors assembled 80,000 professional synchronization ratings covering outputs from eight audio-visual generators, then built a separate 1,000-clip benchmark with multiple ratings per clip.
Why this paper
From WeChat, Tencent Inc. and Renmin University of China · Part of Video Diffusion Scaling, now 19 papers
In one line
SyncReward is a human-aligned reward model that improves audio-visual synchronization in generated content.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§