Long-Video Audio Synthesis with Multi-Agent Collaboration

Yehang Zhang1 *, Xinli Xu1 *, Xiaojie Xu1 *, Li Liu1,2 †, Yingcong Chen1,2 †
1HKUST(GZ), 2HKUST
* Indicates Equal Contribution,
Co-corresponding author

Abstract

Video-to-audio synthesis, which generates synchronized audio for visual content, critically enhances viewer immersion and narrative coherence in film and interactive media. However, video-to-audio dubbing for long-form content remains an unsolved challenge due to dynamic semantic shifts, temporal misalignment, and the absence of dedicated datasets. While existing methods excel in short videos, they falter in long scenarios (e.g., movies) due to fragmented synthesis and inadequate cross-scene consistency. We propose LVAS-Agent, a novel multi-agent framework that emulates professional dubbing workflows through collaborative role specialization. Our approach decomposes long-video synthesis into four steps including scene segmentation, script generation, sound design and audio synthesis. Central innovations include a discussion-correction mechanism for scene/script refinement and a generation-retrieval loop for temporal-semantic alignment. To enable systematic evaluation, we introduce LVAS-Bench, the first benchmark with 207 professionally curated long videos spanning diverse scenarios. Experiments demonstrate superior audio-visual alignment over baseline methods.

Every part of the video is generated at one time by the LVAS-Agent. The video types include: anime fight scenes, video games, real-life videos, and AIGC videos.

Framework

LVAS-Agent Framework

Figure 1: Overview of our LVAS-Agent multi-agent collaboration framework for long-video audio synthesis.

LVAS-Bench

LVAS-Bench

Figure 2: Our LVAS-Bench is presented in the following parts: (a) illustrates sample data from the benchmark, (b) provides statisticaldistributions of audio categories and sub-categories across the dataset, and (c) presents the statistics of video categories within the dataset.

Comparison with Baseline

Our method (right) achieves better audio-visual alignment compared to baseline methods (left)

MMAudio

Prompt: Wind, Tanks moving, Tanks firing

Our Method

Our method achieves better temporal-semantic alignment

MMAudio

Prompt: cat eating, explosion, liquid, arrow release whoosh

Our Method

Reflect more sound effects.


        @misc{zhang2025longvideoaudiosynthesismultiagent,
        title={Long-Video Audio Synthesis with Multi-Agent Collaboration}, 
        author={Yehang Zhang and Xinli Xu and Xiaojie Xu and Li Liu and Yingcong Chen},
        year={2025},
        eprint={2503.10719},
        archivePrefix={arXiv},
        primaryClass={cs.CV},
        url={https://arxiv.org/abs/2503.10719}, 
  }