Toward human-level tonal and modal melody harmonizations

Abstract

This paper considers the problem of automatic music generation in the form of chord voicings, which is an important part of melody harmonization, with the use of Evolutionary Algorithms (EA), based on the rules derived from music theory and practice. The rules, whose role is to ensure the fulfillment of both the formal requirements of chord voicings and its less-formalized aesthetic requirements are encoded in the fitness function of EA. The fitness function is composed of several modules, each of which consists of smaller parts corresponding to the implemented music rules. The above modular design allows for flexible modification and extension of this function. The way the fitness function is constructed and tuned for better chord voicings quality is discussed in the context of music theory and technical EA implementation. In particular, we show how could generated voices be modeled by means of adjusting the relevance of particular fitness function components or extended by adding new components to the fitness function. The proposed algorithm is tested on two types of music: tonal and modal. Although tonal and modal chord voicings are significantly different, the achieved results (assessed by a human expert) indicate that the obtained solutions are both technically and aesthetically correct (i.e. they adhere to the theoretical rules and are nice to listen to). This study extends our previously published conference paper (Mycka et al., 2022).