20 Years of Automatic Chord Recognition from Audio

In 1999, Fujishima published "Realtime Chord Recognition of Musical Sound: a System using Common Lisp Music". This paper kickstarted an active research topic that has been popular in and around the ISMIR community. The field of Automatic Chord Recognition (ACR) has evolved considerably from early knowledge-based systems towards data-driven methods, with neural network approaches arguably being central to current ACR research. Nonetheless, many of its core issues were already addressed or referred to in the Fujishima paper. In this paper, we review those twenty years of ACR according to these issues. We furthermore attempt to frame current directions in the field in order to establish some perspective for future research.