Visualising Chord Progressions in Music Collections: a Big Data Approach

The analysis of large datasets of music audio and other representations entails the need for techniques that support musicologists and other users in interpreting extracted data. We explore and develop visualisation techniques of chord sequence patterns mined from a corpus of over one million tracks. The visualisations use different representations of root movements and chord qualities with geometrical representations, and mostly colour mappings for pattern support. The presented visualisations are being developed in close collaboration with musicologists and can help gain insights into the differences of musical genres and styles as well as support the development of new classification methods. c © Alexander Kachkaev, Daniel Wolff, Mathieu Barthet, Mark Plumbley, Jason Dykes, Tillman Weyde. Licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). Attribution: Alexander Kachkaev, Daniel Wolff, Mathieu Barthet, Mark Plumbley, Jason Dykes, Tillman Weyde. “Visualising Chord Progressions in Music Collections: A Big Data Approach”, 15th International Society for Music Information Retrieval Conference, 2014. 1. THE NEED FOR LARGE-SCALE MUSIC DATA VISUALISATION In the Digital Music Lab 1 project we work on the automatic analysis of large audio databases that results in rich annotations for large corpora of music. The musicological interpretation of this detailed data from thousands of pieces is a challenging task that can benefit greatly from specifically designed configurable visualisation. Most existing big music data visualisation focuses on cultural attributes, mood, or listener behaviour whereas we are focusing on attributes characterising the music content itself (e.g. chords, keys, transcriptions, etc.).