An Automatic Chord Progression Generator Based On Reinforcement Learning

Rapidly developing technology has led convincing improvements in the music production domain and generation of chord progression is one of the affected areas. Chords are the most significant part of music to create harmony. An arrangement of two or more chords in a music portion makes chord progressions. Generation of chord progression is not just a random selection of notes; it requires as much knowledge of grammatical rules as we need for selecting letters to write a sentence. Some note combinations may sound good together while others may sound harsh. A novice musician has to spend significant amount of time to achieve perfection in music composition. To make the task of musicians less time consuming, we delve into the usefulness of automatic chord progression generation through RL algorithm. RL has already been used in numerous fields, but researchers are still investigating its performance in creative tasks. The proposed work uses music theory concepts to define the rewards and Q-learning algorithm to train the RL agent. The fundamental objective of this paper is twofold. Firstly, the goal is to test this approach as an alternative to generate chord progression. Secondly, the aim is to generate tuneful chord progression that the composers can utilize in their work. Results are validated by comparing it with the chords used in the database of 1300 songs.