Algoritmisk komposition av popmusik - utformad med markovkedjor, bayesiskt nätverk samt strukturmodellering

Abstract

This study proposes and explains the development of a method for algorithmically composing modern pop music. The method uses machine learning in a top-down strategy to generate a chord progression and a tting melody. In an attempt to capture the structure and self similarity present in pop music the method generates a structure in the form of partial repetitions. A test to see if a listener can tell a genererated song from a human composed song showed that most listeners had trouble telling the di erence. However, the songs still lacked some important properties resulting in some listeners being able to tell the di erence.