Taming wild horses with Essentia Music Extractor

Abstract

We overview the work done on automatic music audio description using the Essentia Music Extractor. We have successfully applied this tool to the analysis on a large scale in the AcousticBrainz project, in which we have been able to annotate over 3 million music recordings. We are challenged to constantly improve feature extraction algorithms and make our tools more useful for researchers. We reflect on the quality of our current data and identify how it can be improved in following versions to be in line with the essential goals of Music Information Retrieval of building final usable systems. We call on the community for help and contribution.