Use of General Purpose Graphics Processors for Realtime Sound Synthesis


Marc Sosnick

Oral Defence Date: 



TH 434


Professors William Hsu, Dragutin Petkovic and James Wong


Musicians are always seeking new and interesting ways of making sounds, to supplement or replace instruments in performance. Software synthesizers, computers that make sound in realtime, have been around for decades. A newer synthesis approach, modeling physical objects and obtaining sound from manipulating these objects, has been impractical due to high computational requirements. This report describes the design, development and implementation of our Finite Difference Synthesizer (FDS), a realtime software synthesis package that uses the Finite Difference (FD) method to simulate a two-dimensional membrane instrument on the GPU in realtime. 2-D FD methods can generate realistic percussion sounds, including the natural sonic variations that would come from changes in the instrument's geometry, construction materials, and modes of excitation. Detailed FD-based simulations are very compute-intensive, and do not run in realtime on conventional CPUs. We were able to implement large realtime FD simulations on the GPU for realtime audio generation; this had not previously been reported.


Graphics Processing Unit, GPU, Finite Difference, Sound Synthesis, Physical Music Instrument Modeling


Marc Sosnick