Automatic Composition of Music Using a Genetic Algorithm, Emotional Musical Theory and Machine Learning

Adriana Lara, Giovanni Guzmán, Natan Vilchis


This work proposes a mathematical model for computer-aided music composition as a multi-objective optimization problem. This work aims to create a framework to automatically generate a set of songs with two melodies by combining a genetic algorithm with machine learning. Musical patterns were studied [16, 6, 18, 2] to simplify them and apply them for the construction of the optimization model. This work uses recent emotional music theory to construct the optimization problem [11]. Three conflicting objective functions represent the desired characteristics of the melody to be created: (1) song happiness, (2) song minimalism, and (3) song genre. Two of these objectives are analytically designed, fulfilling well-studied features like those in [14, 25, 11]. The third objective function was developed using a machine learning model like in [5, 8, 27]. The software JSymbolic is used [15] for extracting features in real-time and getting the score with the machine learning model trainer in the present work. The results obtained by this work can be listened to by test examples presented in a video format


Music composition, multiobjective optimization, evolutionary music

Full Text: PDF