Page 20 - Spring2020

Scott H. Hawley
Address:
Department of Chemistry & Physics Belmont University 1900 Belmont Boulevard Nashville, Tennessee 37211 USA
Email:
scott.hawley@belmont.edu
Vasileios Chatziioannou
Address:
Department of Music Acoustics (IWK) University of Music and Performing Arts Vienna Anton-von-Webern-Platz 1 1030 Vienna Austria
Email:
chatziioannou@mdw.ac.at
Andrew Morrison
Address:
Department of Natural Sciences Joliet Junior College 1215 Houbolt Road
Joliet, Illinois 60431 USA
Email:
amorriso@jjc.edu
Synthesis of Musical Instrument
Sounds: Physics-Based Modeling or
Machine Learning?
Physics-based modeling provides insight into sound production processes, whereas machine learning generates increasingly realistic imitations from recordings alone.
Introduction
Music (and sound) synthesis has become widespread and is found across many musical genres. Many members of the Acoustical Society of America (ASA) also use various types of synthesis to produce the sounds used in research projects, ranging from measurements of hearing to studies of sound propagation in the oceans. Although there is a long history of attempts to synthesize music and other sounds (e.g., Pejrolo and Metcalfe, 2017), this paper focuses on the most recent, and very exciting, approaches now being used to synthesize the sounds of musi- cal instruments as well as for a wide range of other acoustic phenomena. Earlier methods for musical instrument sound synthesis used signal-processing effects such as frequency-modulated (FM) synthesis or wavetables (i.e., sampling), whereas the applications of physics-based modeling and machine learning have only been applied relatively recently because of their higher computational cost.
Musical acoustics is a diverse scientific field that deals with research and applications ranging from making musical instruments to the perception of sound. One of the driving questions for musical acoustics is “How does this musical instrument pro- duce its characteristic sound?” Well-known scientists throughout history, including the ancient Greek mathematician and philosopher Pythagoras, Galileo Galilei, Ernst Chladni, Herman von Helmholtz, and Chandrashekhara Venkata Raman (Nobel Laureate in Physics), increased the overall scientific understanding of their eras by working on music and musical instruments. The formulation of reliable physi- cal models of musical instruments was pursued after developments in the field of differential equations, whereas digital sound synthesis was only possible following advances in numerical analysis and computer science.
The ability to digitally model musical instruments provides music creators with multiple desirable capabilities (Smith, 2011), among which are
(1) portability: virtual instruments and software effects require no space or weight; (2) flexibility: many such instruments can be stored and accessed together and
quickly modified;
(3) signal to noise: often can be higher with digital instruments;
(4) centralized, automated control;
(5) repeatability: simulated instruments can be exactly the same as opposed to
physical systems that may involve variations due to, for example, construction,
humidity, and temperature; and
(6) extension: the development of digital instruments involves fewer constraints
than their real-world counterparts.
20 | Acoustics Today | Spring 2020 | volume 16, issue 1 ©2020 Acoustical Society of America. All rights reserved. https://doi.org/10.1121/AT.2020.16.1.20