Masked Generative Video-to-Audio Transformers with Enhanced Synchronicity

TL;DR

This paper introduces MaskVAT, a video-to-audio generative model that achieves high-quality, semantically accurate, and temporally synchronized audio generation from video inputs by integrating a high-quality audio codec with a sequence-to-sequence masked generative approach.

Contribution

The paper presents a novel V2A model combining a full-band audio codec with a sequence-to-sequence masked generator to improve synchronization without sacrificing quality.

Findings

Achieves highly synchronized audio with video actions.

Competitive with state-of-the-art non-codec models.

Demonstrates high audio quality and semantic matching.

Abstract

Video-to-audio (V2A) generation leverages visual-only video features to render plausible sounds that match the scene. Importantly, the generated sound onsets should match the visual actions that are aligned with them, otherwise unnatural synchronization artifacts arise. Recent works have explored the progression of conditioning sound generators on still images and then video features, focusing on quality and semantic matching while ignoring synchronization, or by sacrificing some amount of quality to focus on improving synchronization only. In this work, we propose a V2A generative model, named MaskVAT, that interconnects a full-band high-quality general audio codec with a sequence-to-sequence masked generative model. This combination allows modeling both high audio quality, semantic matching, and temporal synchronicity at the same time. Our results show that, by combining a high-quality codec with the proper pre-trained audio-visual features and a sequence-to-sequence parallel structure, we are able to yield highly synchronized results on one hand, whilst being competitive with the state of the art of non-codec generative audio models. Sample videos and generated audios are available at https://maskvat.github.io .