ï~~ARCSYN: AN EXPRESSIVE AND EFFICIENT ADDITIVE
SYNTHESIS ARCHITECTURE
David Romblom
Media Arts and Technology
University of California, Santa Barbara
romblom @gmail.com
ABSTRACT
Arcsyn is an architecture and control system for
expressive additive synthesis; it provides satisfying
dynamic behavior, compelling transients, and non-static
tones. The complexity of additive synthesis is
encapsulated within a musically sensitive control system
inspired by instrument physics. Timbral information is
represented in the frequency domain, and can be
specified parametrically or made to use popular formats
such as SDIF.
1. INTRODUCTION
Arcsyn is a software synthesis technique that addresses
the shortcomings of electronic instruments when
compared to acoustic instruments of the classical canon.
It is conventional in the fact that it provides the
performer with tools to create expressive melodic
content, and that it in some ways acts like an acoustic
instrument. It is more modern in the fact that the timbre
can be arbitrarily defined, that timbre can "morph" to
other arbitrary timbres, that it can be sequenced using
"analog" control signals, and that it has novel and exotic
modulation capabilities. Arcsyn is meant as a general
synthesis method that begins where the subtractive
paradigm meets its limitations. As opposed to more
exotic synthesis techniques, it is an elegant
representation of complexity and uses computational
resources efficiently.
Using an Analysis/Synthesis package such as SPEAR [2]
one can see a number of interesting aspects in the
recording of a given note. In addition to expected
features such as vibrato and weaker high frequency
partials, one also observes that all partials have random
motion in amplitude and frequency, highly varied attack
and release times, and are generally inharmonic during
attacks. When thinking of the many notes that might
make up a musical phrase, we are assisted by Strawn's
[6] summary of note to note pitch transitions.
Arcsyn uses steady-state spectral information and models
the transitions and temporal variation. Synthesis is
implemented as a bank of bandwidth-enhanced
sinusoidal oscillators [1]. The steady-state spectral
information can be specified parametrically, taken from
an Analysis/Synthesis package such as SPEAR or Loris,
or imported from an SDIF [9] library. This allows for
both acoustic instrument emulation and instrument-like
electronic tones.
For a given timbre, Arcsyn uses both ff and pp steadystate spectral information. One can interpolate freely
between these two layers, allowing tones to swell or
decay regardless of the onset dynamic. The ability to
arbitrarily define the spectral envelope of the pp and ff
layers allows non-trivial evolution in playing dynamic.
The architecture can be easily extended to include the
additional detail of multiple dynamic layers. Spectral
morphing is achieved by adding a second, entirely
distinct timbre with its own dynamic information. Here
again, it is easy to extended the architecture to include
multiple timbres. The third dimension of interpolation is
pitch; new spectral information is specified at each welltempered note. For acoustic emulation, pitch
interpolation preserves instrument formants and yields
"acoustically-correct" vibrato and glissandi. For
electronic timbres, we gain instrument-like playing
characteristics.
2. BEYOND SUBTRACTIVE SYNTHESIS
Subtractive synthesis is a pervasive synthesis
architecture, where varied harmonic content is achieved
by moving the cutoff frequency and resonance of a
resonant lowpass filter. There are a small set of standard
waveforms, the harmonic vocabulary of each waveform
is predetermined by the filter's sweep range. This
vocabulary is not sufficiently large to be musically
satisfying, though non-linear distortion [3] and waveform
modulation do help the situation.
Samplers are a close relative of the subtractive
synthesizer that operate by playing back recordings
"frozen" at a given pitch and dynamic. For acoustic
instrument emulation, glissandos and vibrato are
modeled by varying the playback speed; this method
shifts the instrument's formant resonance up and down
resulting in a very unnatural sound. A preferable
method, one closer to the physics of acoustic
instruments, would preserve the instrument formants
while the individual partials of the tone move under
them. [7]
Different recordings are played back depending on the
note-on velocity. This presents problems for sustained
instruments where a performer may wish to start softly
and then swell: they are limited to the harmonic content
of the sample they first played. To address this, modern
samplers cross-fade to a second recording, though this
can sound unnatural due to phase misalignment and
requires detailed, case-by-case control programming.
Stasis in the tone is addressed by modulating the filter
cutoff or oscillator pitch with a control signal. These
478
0