~ICMC 2015 - Sept. 25 - Oct. 1, 2015 - CEMI, University of North Texas
Improving and Adapting Finite State Transducer Methods
for Musical Accompaniment
Jonathan P. Forsyth
Music and Audio Research Lab (MARL)
New York University
[email protected]
Michael Musick
Music and Audio Research Lab (MARL)
New York University
[email protected]
ABSTRACT
A common area of research within music technology is the
automatic generation of musical accompaniment. Of particular interest is the generation of harmonic accompaniment
to melody. While there are many approaches to solving this
problem, the use of various types of finite state machines is
popular One such finite state machine, the finite state transducer (FST), has been used in a few accompaniment generation systems. Although the FST is widely used in speech
recognition, its application in accompaniment generation systems has been relatively unsophisticated. In this paper; we
introduce an improved approach to generating harmonic accompaniment to melody based on techniques used in speech
recognition. In addition, we describe how we extend this
method to the generation of rhythmic accompaniment. Finally, we discuss the integration of these offline accompaniment systems into a real-time system.
1. INTRODUCTION
Composers, performers, and researchers have been developing a wide variety of interactive music systems for over 40
years. Some systems are designed for use in a live improvisational context, while others are organized around a "scorefollowing" paradigm in which a system attempts to align a
live performance with a pre-existing score. A few of these
systems are specifically designed to generate some form of
accompaniment in real time. Other, closely related research
has focused on the development of offline solutions to the
problem of generating musical accompaniment, usually in
the form of harmonic accompaniment to melody. Some of
these automatic accompaniment generation systems [1,2] use
Copyright: ~2015 Jonathan P. Forsyth et al. This is an open-access article
distributed under the terms of the Creative Commons Attribution License 3.0
Unported, which permits unrestricted use, distribution, and reproduction in
any medium, provided the original author and source are credited.
Rachel M. Bittner
Music and Audio Research Lab (MARL)
New York University
[email protected]
Juan P. Bello
Music and Audio Research Lab (MARL)
New York University
[email protected]
a particular type of finite state machine, a finite state transducer (FST), to generate harmonic accompaniment. While
FSTs are well-suited to the task of accompaniment generation, these systems use a fairly simple transducer topology,
and are designed for offline use.
In this paper, we propose an improved FST topology inspired by techniques used in speech recognition. In addition,
we describe how this approach is extended to generate rhythmic as well as harmonic accompaniment. We also discuss
adapting our approach for a real-time context, in the form of
a collaborative music-creation installation entitled The Harmonically Ecosystemic Machine; Sonic Space No. 7, and describe the various challenges and compromises involved in
implementing the new FST topology and adapting the system
to a real-time context.
The remainder of this paper is organized as follows. In the
next section, we provide background on relevant interactive
music and automatic accompaniment generation systems. In
section 3, we provide background information on finite state
transducers. In section 4, we discuss our current approach
to harmonic accompaniment generation and an approach to
rhythmic accompaniment generation, and in section 5, we describe the application of these systems to a real-time context.
Finally, in section 6 we discuss our conclusions and provide
directions for future work.
2. RELATED WORK
In this section, we provide some of the context for the current
work within the fields of interactive music systems and offline
accompaniment generation systems.
2.1 Interactive Music Systems
Interactive music systems, defined by Rowe as systems "whose
behavior changes in response to musical input" [3], include a
great many different types of systems, and cover a variety of
musical applications. A number of systems, such as the Music
Plus One [4] and Antescofo systems [5], are score-followers,
- 290 -
