Automatic melodic harmonization tackles the assignment of harmony content (musical chords) over a given melody. Probabilistic approaches to melodic harmonization utilize statistical information derived from a training dataset, producing harmonies that encapsulate some harmonic characteristics of the training dataset. Training data is usually annotated symbolic musical notation. In addition to the obvious musicological interest, different machine learning approaches and algorithms have been proposed for such a task, strengthening thus the challenge of efficient and effective music information utilization using probabilistic systems. Consequently, the aim of this chapter is to provide an overview of the specific research domain as well as to shed light on the subtasks that have arisen and since evolved. Finally, new trends and future directions are discussed along with the challenges which still remain unsolved.
TopIntroduction
In music, harmony is the use of simultaneous pitches (tones, notes), or chords accompanying a given melody (Benward & Saker, 2003). However, in order, to understand harmony, it is first necessary to grasp what melody is. Melody is a group of notes played one after the other, the tune that is often the easiest part of music to remember, the part that one may hum. Harmony is also a group of notes, except that these notes are played in the background, beneath and around the melody. The role of harmony role is to accompany the melody and is usually expressed as a sequence of different voices or instruments that play musical chords. A chord, in music, is any harmonic set of three or more notes that is heard as if sounding simultaneously (Benward & Saker, 2003). Chords are typically consisting of four voices ranging from a higher to lower pitch: Soprano, Alto, Tenor, and Bass. The assignment of musical chords in a given melody is called melodic harmonisation, which is the object of study for this chapter.
The task of melody harmonization incorporates the preservation of balanced relations between the melody and all chord-composing sequences. This is achieved by a set of musical “rules” which defines a certain music style, such as classical, rock, jazz etc. The analysis of harmony is normally performed manually by music experts; however, with the advent of computers, research has investigated whether all these rules can be analyzed and simulated by computerized frameworks. The practice of abstracting rules of harmony and placing them in a linguistic framework has been a part of computer science at least since the 1960s (Winograd, 1968; Jackson, 1967). Automatic melodic harmonization is a natural extension of harmonic analysis, and an important component in music information research. Its function is to clarify principles used by composers and musicians, and to capture these rules in an artificial intelligence (AI) framework (Koops, 2012).
Automated melodic harmonization has so far been approached from two different angles: with either the purpose of finding a satisfactory chord sequence for a given melody (performed by the soprano voice) or with the purpose of finding the remaining three voices that complete the harmony for a given bass line (Figure 1). The four-part harmonization is a traditional part of the theoretical education of Western classical musicians and therefore numerous researchers have attempted to generate automatically the four-part harmonization.
Figure 1.
Automated melodic harmonization approach: Finding the remaining three voices that complete the harmony for a given soprano line.
The task of automated melodic harmonization can be considered as a branch of algorithmic musical composition (Jacob, 1996) which is the application of a strict, well-defined artificial intelligent algorithm for the process of composing music. In the current context, music can be considered as a single instrument/voice or a combination of voices and instruments, as clearly shown by both the music industry and common practice. As far as the computational part of the harmonization is concerned, machine learning algorithms and techniques have been used widely in the field. Nevertheless, most existing methods use a context generic approach (HMM), which makes little use of domain specific information (Eddy, 1998).
The aim and key contribution of this chapter is to provide an introduction to the importance and the requirements of automated melodic harmonization research, as well as to present a concise literature review of the main conceptual approaches in this area. The chapter presents newly emerging research directions concerning idiom representations and discusses the need for further research which stimulates a number of open challenges in the field of automated melodic harmonization.