Presenter Name:Maya Flannery

School/Affiliation:McMaster University

Co-Authors:Swati Anant, Matthew Woolhouse

Abstract:

Classification of music by genre is highly speculative. Often, industry experts determine a song, album, or artist’s genre when new music is released, but the criteria by which their decisions are made remain vague at best. Furthermore, such expert decisions do not necessarily mean that listeners universally share these views. Our present work aims to establish alternative methods of classification based on objective measures. We tested the viability of the Essentia library, a programming tool which extracts low- and high-level audio features (e.g., loudness, tempo, mode, and genre) from digital audio files, with respect to its ability to distinguish genres. We acquired lists of representative songs by genre (e.g., pop, classical, dance, etc.) based on curated collections by Spotify (“top”, “best”, or “essential” genres). The audio content of each song was analysed with Essentia, then analysed by genre to obtain a descriptive representation of musical features. These descriptive representations allowed us to identify similarities and differences of musical feature use across genres and feature variability within genres. Here, we present this “snapshot” of popular music as determined by Spotify, but we can apply this method to music categorized by period or region to determine how music differs along other dimensions. Such insights will allow us to empirically validate stimuli that represent specific categories (e.g., genre) and features in our experimental studies investigating music-listening behaviour. Thus, we can gain a deeper understanding of the relationship between listeners and musical preference.