[4C2-GS-13] AI application: Entertainment

[4C2-GS-13-05] Automatic Chord Recognition Using Semi-Supervised Learning with Variational AutoEncoder

〇Shu Kumata¹, Masahiro Suzuki¹, Yutaka Matsuo¹ (1. University of Tokyo)

Keywords:Music, Automatic Chord Recognition, Semi-Supervised Learning

In recent years, with the advancement of machine learning technology, methods of data-driven music analysis have been widely studied. Among them, the automatic chord recognition task is an important task as a music summary. However, the annotation of music data requires the annotator having an advanced ability in music, and it takes a long time to annotate music since the data is longer than other time-series data. Therefore, there is not enough teacher data compared with data such as images and natural languages. On the other hand, music data without annotation can be widely collected on the Internet. Therefore, in this research, we try automatic chord recognition by semi-supervised learning using VAE. As a result of the experiment, the learning of VAE is not successful, but the result of the supervised learning using CNN shows that the semi-supervised learning may be effective.