V2A-DPO: Omni-Preference Optimization for Video-to-Audio Generation
This work addresses the challenge of improving audio quality and alignment in video-to-audio generation for applications in multimedia and AI, representing an incremental advancement with specific optimizations for flow-based models.
The paper tackles the problem of aligning generated audio with human preferences in video-to-audio generation by introducing V2A-DPO, a Direct Preference Optimization framework, which results in state-of-the-art performance on the VGGSound dataset, outperforming baselines like DDPO and pre-trained models.
This paper introduces V2A-DPO, a novel Direct Preference Optimization (DPO) framework tailored for flow-based video-to-audio generation (V2A) models, incorporating key adaptations to effectively align generated audio with human preferences. Our approach incorporates three core innovations: (1) AudioScore-a comprehensive human preference-aligned scoring system for assessing semantic consistency, temporal alignment, and perceptual quality of synthesized audio; (2) an automated AudioScore-driven pipeline for generating large-scale preference pair data for DPO optimization; (3) a curriculum learning-empowered DPO optimization strategy specifically tailored for flow-based generative models. Experiments on benchmark VGGSound dataset demonstrate that human-preference aligned Frieren and MMAudio using V2A-DPO outperform their counterparts optimized using Denoising Diffusion Policy Optimization (DDPO) as well as pre-trained baselines. Furthermore, our DPO-optimized MMAudio achieves state-of-the-art performance across multiple metrics, surpassing published V2A models.