Drummistic Fingerprints: Unique Drummer Identification via the Application of Data Analytics & Machine Learning

This paper investigates the subject of artist recognition in recorded performances, an area of Music Information Retrieval (MIR) called Performer Identification. Focusing on the drum set, it explores the ideas that skilled drummers possess unique 'drummistic fingerprints' that can be extracted using machine learning techniques, and used to identify them in previously unseen recordings, and if it is possible that transfer learning can be successfully executed between music styles based on this technique. Many performer identification studies exist for other melodic and harmonic instruments, but the area of drummer identification has to date not received much coverage. The artefact produced was able to produce a detailed comparative analysis of the contained performances (by drummer, music style, and limb), and utilise classification methods like K-Nearest Neighbours and GradientBoostedClassifier ensembles to make predictions on the identity of the drummers. This study determined that yes, such unique 'drummistic fingerprints' exist, can be extracted from performances and used to successfully identify drummers in unseen performance data much like other studies were able to identify other types of instrumentalists. Additionally, strong evidence was found indicating transfer learning is possible, but results appear variable depending on the individual performers.