Impact of visual assistance for automated audio captioning

TL;DR

This paper investigates how visual features influence automated audio captioning, revealing that semantic visual embeddings improve performance, whereas other features do not, highlighting task-dependent optimal visual information use.

Contribution

The study demonstrates the task-specific effectiveness of different pretrained visual features in audiovisual captioning, emphasizing the importance of selecting appropriate embeddings for improved results.

Findings

Semantic visual features improve captioning performance

Other pretrained features show no significant gains

Optimal visual embeddings depend on the specific audio task

Abstract

We study the impact of visual assistance for automated audio captioning. Utilizing multi-encoder transformer architectures, which have previously been employed to introduce vision-related information in the context of sound event detection, we analyze the usefulness of incorporating a variety of pretrained features. We perform experiments on a YouTube-based audiovisual data set and investigate the effect of applying the considered transfer learning technique in terms of a variety of captioning metrics. We find that only one of the considered kinds of pretrained features provides consistent improvements, while the others do not provide any noteworthy gains at all. Interestingly, the outcomes of prior research efforts indicate that the exact opposite is true in the case of sound event detection, leading us to conclude that the optimal choice of visual embeddings is strongly dependent on the task at hand. More specifically, visual features focusing on semantics appear appropriate in the context of automated audio captioning, while for sound event detection, time information seems to be more important.