Abstract

Genre classification has been a hot topic for years now in the field of music information retrieval. Most of the current works study music using song waves as input data. In this work, we present a different approach to genre classification taking into account chord progressions. A full data set has been created for this work: first gathering songs for each genre (pop, indie, rock and reggae) from Spotify and then scraping chord progressions data of that songs from the website Ultimate Guitar. Different models which aim to classify the genre of the songs have been trained using convolutional neural networks for pair comparison between genres classification. Some of those models are used for discerning between two concrete genres given, getting up to a value of 91% for AUC metric classifying songs between pop and rock. Music Information Retrieval Chord Progressions Convolutional Neural Networks Spotify API Genre Classification.

Statistics