MUSE: A Genetic Algorithm for Musical Chord Progression Generation
Location
Loosemore Auditorium
Description
PURPOSE: Foundational to our understanding and enjoyment of music is the intersection of harmony and movement. This intersection manifests as chord progressions which themselves underscore the rhythm and melody of a piece. In musical compositions, these progressions often follow a set of rules and patterns which are themselves frequently broken for the sake of novelty. PROCEDURES: In this work, we developed a genetic algorithm which learns these rules and patterns (and how to break them) from a dataset of 890 songs from various periods of the Billboard Top 100 rankings. OUTCOME: The algorithm learned to generate increasingly valid, yet interesting chord progressions via penalties based on both conditional probabilities extracted from the aforementioned dataset and weights applied to the characteristics from which the penalty is derived. Additionally, the beginning and end of a progression may be seeded (either in totality or for a percentage of generated patterns) such that the algorithm will generate a bridging progression to connect the seeded points. IMPACT: To this end, the algorithm proposed chord progressions and supplied vectors of computer-aided algorithmic composition. To demonstrate the validity of the system, we present a subset of generated progressions that both conform to known musical patterns and contain interesting deviations.
MUSE: A Genetic Algorithm for Musical Chord Progression Generation
Loosemore Auditorium
PURPOSE: Foundational to our understanding and enjoyment of music is the intersection of harmony and movement. This intersection manifests as chord progressions which themselves underscore the rhythm and melody of a piece. In musical compositions, these progressions often follow a set of rules and patterns which are themselves frequently broken for the sake of novelty. PROCEDURES: In this work, we developed a genetic algorithm which learns these rules and patterns (and how to break them) from a dataset of 890 songs from various periods of the Billboard Top 100 rankings. OUTCOME: The algorithm learned to generate increasingly valid, yet interesting chord progressions via penalties based on both conditional probabilities extracted from the aforementioned dataset and weights applied to the characteristics from which the penalty is derived. Additionally, the beginning and end of a progression may be seeded (either in totality or for a percentage of generated patterns) such that the algorithm will generate a bridging progression to connect the seeded points. IMPACT: To this end, the algorithm proposed chord progressions and supplied vectors of computer-aided algorithmic composition. To demonstrate the validity of the system, we present a subset of generated progressions that both conform to known musical patterns and contain interesting deviations.