ï~~
A MICROCOMPUTER SYSTEM FOR TUE REAL-TIME EXPLORATION OF MUSICAL STRUCTURES
Michael Greenhough
Physics Department
University College
Cardiff, Wales
U.K.
ABSTRACT
The paper describes a system which allows intuitive
control of the melodic and rhythmic properties of up
to sixteen voices and the horizontal, vertical and
diagonal relationships between them. At-present it
runs on an IBM-PC AT microcomputer and generates
MIDI data for a Yamaha DX7 synthesiser. The pitch
and duration of sound events are produced by a
random generator constrained by probability
distributions which are graphically presented and
can be modified directly whilst the system is
running. Thus the user can explore musical
structures from an unconventional, implicit domain
of control.
All the probability distributions used are identical
in structure and are simply interpreted in different
ways to fulfill different musical purposes. For
example, distributions representing a-priori
probabilities of occurrence can be used to impose
context-free control of the pitch or duration of
events. Other distributions may represent the
first- and higher-order differences between the
event parameter being determined and various userselectable reference points. These points are the
pitches and durations of any previously determined
events. In this way context-sensitive control is
imposed.
The use of a difference method for event generation
brings about a dramatic reduction in data and in
processing time which enables the essential realtime operation but involves, inevitably, a
corresponding loss of generality. The results so
far encourage one to believe that some of this
generality is expendable, that is, that the
particular kind of redundancy in the output is
consistent with certain general musical tendencies.
By varying only the data fed to the system (ie
probability values and context reference points) a
wide variety of musical effects can be obtained.
1. INTRODUCTION
Traditional musical instruments allow us to explore
timbre in an intuitive fashion. This is possible
because such instruments (i) operate in real time
and (ii) have been contrived to permit only a
limited range of sounds, of which a relatively high
proportion are of musical significance. The system
described here is an attempt to extend this
principle of intuitive control to a higher level of
musical structure. The ideal system would have a
natural tendency to produce 'musical' structures
(in the way that, say, a guitar does for musical
timbres).
To the extent that these two structural levels are
comparable it is encouraging to note that neither
the developers nor the players of traditional
instruments have much explicit knowledge of how
these systems work or how we perceive their sounds.
Craftsmen and virtuosi have persisted in making
successful instruments and music in blissful
ignorance of musical and psychological acoustics.
Thus though possessing only a very limited
theoretical musical knowledge I am trying to
develop a system which enables one to generate
useful musical material by virtue, merely, of the
ability to recognise it as such when it occurs.
It is essential to the philosophy of the system,
therefore, that the user should be freed from the
necessity to specify pitch and duration information
explicitly. This information must, however, come
from somewhere. It is generated by the software
itself, under indirect user control, using a
collection of probability distributions. These are
simply arrays held in memory and modified
interactively at run time. If musical output
satisfactory for some ourpose is arrived at in this
way then the responsible distributions can be
stored in a disk file, along with other related
data. The probabilities are stored logarithmically
and the generating algorithm, written in 8086
assembler, is so constructed that the process of
calculating the simultaneous effect of the many
controlling distributions reduces to the shifting
of indices and the addition of elements. Thus the
pitch or duration of an event can be calculated in
a few milliseconds although it may have a complex
dependence on several earlier event parameters.
2. THE CONTROL OF PITCH
Each voice at each moment must be being controlled
by at least one distribution which represents the
a priori weighting of different pitch values and
can be tailored to produce scales of a traditional,
or quite arbitrary, form. A flat distribution will
maximise the information content and produce
chromatic chaos. Biasing the distribution to
pitches of, say, a diatonic scale will introduce
redundancy of the context-free variety and impose a
crude tonal character on the series of notes output.
91
ICMC 86 Proceedings
0