Learning and Evaluating Musical Features with Deep Autoencoders

TL;DR

This paper explores methods to learn musical embeddings using deep autoencoders, evaluating their effectiveness on prediction and classification tasks to improve musical feature representation.

Contribution

It introduces autoencoding-based techniques for learning musical embeddings from symbolic data, with comprehensive evaluation on predictive and classification tasks.

Findings

Autoencoder methods effectively learn meaningful musical features.

Embeddings improve performance on forward prediction tasks.

Embeddings enhance classification accuracy for musical data.

Abstract

In this work we describe and evaluate methods to learn musical embeddings. Each embedding is a vector that represents four contiguous beats of music and is derived from a symbolic representation. We consider autoencoding-based methods including denoising autoencoders, and context reconstruction, and evaluate the resulting embeddings on a forward prediction and a classification task.