JAVA AUDIO SPATIALIZATION

This project was implemented for MAT 240C - Digital Audio Programming, instructed by Stephen Travis Pope.

introduction

This project is a simple framework for audio spatialization in Java.  The main goal was to support spatialization of audio using an arbitrary number of input and output channels, without being limited to a specific platform, format, or sampling rate.  In addition to this, various math and audio utilities were implemented towards this goal.

The source code used in the project is available here: src.zip (20Kb)

The javadoc documentation can be viewed here.

implementation notes

Constant power panning and inverse square sound falloff was used in the HeadphoneSpace implementation of the AudioSpace interface.

The current implementation of JavaSound is limited to two channel audio output, but this framework can in theory support any number of channels.

All audio samples are converted to signed double values (range -1 to 1) to simplify manipulation by mathematical functions.

Figure 1 - Examples showing Inverse Square Sound Falloff and Constant Power Panning

A simple example of the application's output (spatializing a constant tone) can be heard here: tone_spin_ccw.mp3 (171Kb), as well as here: tone_spiral_away.mp3 (172Kb)

ifs spatialization

As a test application of the framework, a graphical spatializer was developed using iterated function systems (IFS), a type of self-similar fractal structure (see below for further details).

Figure 2 - Graphic IFS Spatializer

The application allows the user to graphically manipulate an IFS, which represents a two dimensional cloud of sound particles around the user's head.  Captured audio is spatialized at random points in the cloud, and played back in real-time.  An example of the application's output can be heard here: ifs.mp3 (84Kb)

appendix: iterated function systems

An iterated function system can be defined as an attractor formed by the union of a finite number of contraction mappings.  This project uses a type of contraction mapping known as an affine transform, wherein each transform in the IFS is a scaled, translated, sheared, and/or rotated image of the entire IFS.  Since this 'self-transformation' occurs an infinite number of times, the IFS possesses a very complex, self-similar structure.

Figure 3 - Simulated generation of IFS with three affine transforms (scale + translation)

To generate an image of the IFS, a random-iteration algorithm can be used.  Since an IFS is an attractor, if a point is repetitively transformed by randomly-chosen transforms in the IFS, the point will gravitate toward only those points in the IFS.   If the point is plotted after each random transformation, an image of the IFS' structure will quickly coalesce.

references and links

Java Sound Resources

Tritonus

God's gift to Java developers

Ptolemy Project

Donald M. Davis, 'The Nature and Power of Mathematics' - Princeton, 1993

Michael F. Barnsley, 'Fractals Everywhere' - Academic Press, 1993