ï~~Physically Informed Sonic Modeling (PhISM): Percussive Synthesis
Perry R. Cook
Department of Computer Science,
Department of Music, Princeton University
[email protected]
Abstract
This paper describes methods for analysis/synthesis of percussive sounds, applying
principles from physical modeling, and coupling the results to parametric synthesis
techniques such as modal, sinusoidal, and granular synthesis. Emphasis is placed upon
using physical modeling to provide expressive control of efficient synthesis models.
Examples presented include marimba, police whistle, and maracha.
1 Introduction
As synthesis using physical modeling becomes more
widespread and works its way into mainstream
commercial workstation software and dedicated music
synthesis hardware, large subsets of traditional
instruments have yet to be explored using physical
models. This paper focuses on natural and expressive
synthesis of sounds in the percussion family, and
describes a set of methods for systematically analyzing
and parametrically synthesizing many percussion
instruments. Since percussion instruments arguably
represent a broader spectrum of physical configurations
than any other instrument family (bars, plates,
membranes, cavity and tube resonators, and nonlinearites of countless varieties, all coupled to each
other in varieties of ways), there exist no common
model features which allow the percussion family to be
captured with simple variations in model topology or
variations of the parameters of a meta-model. By
grouping the percussion family into sub-families, some
similarities can be exploited, but there are still large
varieties of configurations and excitation means. The
expressive nature of percussion lies in the ability to
excite (usually strike) objects, with a variety of objects,
in a variety of locations, and in a variety of ways.
Sampling synthesis based on digital recordings lacks
the obvious means for manipulating meaningful
parameters in musical ways. Direct real-time synthesis
of many percussion instruments by physical modeling
is likely to remain beyond practical computing means
for some time to come. Further, the physical
mechanisms of many such systems are still not
completely understood.
The analysis/synthesis approach described here,
called Physically Informed Sonic Modeling (PhISM),
draws upon many of the techniques found in research
in physical modeling, Fourier analysis/synthesis,
wavelet analysis/synthesis, and granular synthesis. The
primary goal is to provide expressive synthesis, with
meaningful control parameters similar to those a
percussionist would use to create subtle expression on
an actual instrument. Secondary goals are efficiency of
the algorithms, and the ability of the system to behave
somewhat as a traditional analysis/synthesis technique.
Motivated by the two distinct families of percussion
instruments modeled in this paper, two PhISM-family
synthesis algorithms will be presented. The first
algorithm, called Physically Informed Spectral
Additive Modeling (PhISAM), is implemented using
modal synthesis (resonant filters). PhISAM is suitable
for bells, bar percussion, and other such instruments
exhibiting exponentially decaying resonant behaviors.
The second synthesis algorithm, called Physically
Informed Stochastic Event Modeling (PhISEM), is
based on pseudo random overlapping and adding of
small grains of sound according to predetermined
physical rules and parameters. The PhISEM algorithm
family is suitable for instruments such as maracha,
guiro, tambourine, and others characterized by random
interactions of sound-producing component objects.
Both model families allow for scaleable amounts of
modeling of the physics of the instrument and
excitation. One of many possible synthesis techniques
can be employed for synthesizing the actual sound,
such as amplitude modulated oscillator synthesis,
frequency modulation, stochastically parameterized
sampling (granular) synthesis, and modal synthesis.
2 PhISAM: Physically Informed
Spectral Additive Modeling
The PhISAM (Physically Informed Spectral Additive
Modeling) algorithm is based on modal synthesis, but
could be realized with sinusoidal oscillators as well.
The modal filter or oscillator control parameters are
driven and controlled by rules derived from
predetermined Fourier boundary methods, and/or from
analysis data extracted from recorded sounds. The
PhISAM algorithm is suitable for resonant percussion
Cook
228
ICMC Proceedings 1996
