Translating Visual Art into Music

TL;DR

This paper presents SynVAE, a model that translates visual art into music by learning a shared latent space, achieving high consistency and human-matching accuracy without paired datasets.

Contribution

The research introduces SynVAE, a novel variational autoencoder that enables cross-modal translation between images and music without requiring paired training data.

Findings

SynVAE maintains high information content during translation.

The model achieves up to 73% accuracy in human matching tasks.

It demonstrates effective cross-modal latent space consistency.

Abstract

The Synesthetic Variational Autoencoder (SynVAE) introduced in this research is able to learn a consistent mapping between visual and auditive sensory modalities in the absence of paired datasets. A quantitative evaluation on MNIST as well as the Behance Artistic Media dataset (BAM) shows that SynVAE is capable of retaining sufficient information content during the translation while maintaining cross-modal latent space consistency. In a qualitative evaluation trial, human evaluators were furthermore able to match musical samples with the images which generated them with accuracies of up to 73%.