Learning Sight from Sound: Ambient Sound Provides Supervision for Visual Learning

TL;DR

This paper demonstrates that ambient sounds can serve as a supervisory signal to train visual models, enabling the learning of object and scene representations without explicit labels.

Contribution

It introduces a method to use sound as supervision for visual learning, showing that neural networks can learn meaningful visual features from audio-visual data.

Findings

The learned visual representations are comparable to other unsupervised methods.

The network develops units selective to objects associated with characteristic sounds.

Sound supervision improves visual feature learning without manual labels.

Abstract

The sound of crashing waves, the roar of fast-moving cars -- sound conveys important information about the objects in our surroundings. In this work, we show that ambient sounds can be used as a supervisory signal for learning visual models. To demonstrate this, we train a convolutional neural network to predict a statistical summary of the sound associated with a video frame. We show that, through this process, the network learns a representation that conveys information about objects and scenes. We evaluate this representation on several recognition tasks, finding that its performance is comparable to that of other state-of-the-art unsupervised learning methods. Finally, we show through visualizations that the network learns units that are selective to objects that are often associated with characteristic sounds. This paper extends an earlier conference paper, Owens et al. 2016, with additional experiments and discussion.