Ingeniero Electrónico

Real-Time Audio Spatialization with Inexpensive Hardware
David A. Burgess Graphics Visualization and Usability Center - Multimedia Group Georgia Institute of Technology Atlanta, Georgia 30332 burgess@cc.gatech.edu

Abstract

There are a variety of potential uses for interactive spatial sound in human-machine interfaces, but tremendous computational costs have made most of these applicationsimpractical. Recently, however, single-chip digital signal processors (DSP’s) have made real-time spatial audio an affordable possibility for many workstations. This paper describes a spatialization technique based on empirically derived FIR filters. The fundamental performance and quality limits for this technique are discussed as well the minimum bandwidth required for the associated audiochannels. It is shown that current single-chip DSP’s may be expected to spatialize several sources to different positions in real time. Techniques for improving spatial audio quality and performance are described. As an example application, the spatial sound system of an all-acoustic computer interface is described.
Introduction

step is to convert monaural sounds to binaural sounds by artificiallyspatializing them. This goal has lead research interest in the subject by the military, by NASA (Wenzel et al 1988), and by user interface designers (Ludwig et al 1990, 1991). The work described in this paper is part of the Mercator2 project (Mynatt & Edwards, 1992). The goal of Mercator is to allow visually-impaired computer users to have access to application software packages with graphical userinterfaces under the X Window System3. Mercator will accomplish this task by mapping the structures and behaviors of X applications to an auditory and tactile space. An important feature of Mercator will be the use of spatial sound as a primary organizational cue. For Mercator to be useful, this feature must be implemented at a reasonable cost with off-the-shelf hardware. This paper is written intwo sections. The first section describes the general spatialization technique. The second section describes spatial sound as implemented in Mercator.
Section I: Background and Spatialization Technique How We Localize Sounds

In recent years there has been research into what is called binaural1 recording. Binaural recordings are intended to be reproduced through headphones, and can give thelistener a very realistic sense of sound sources being located in the space outside of the head. These virtual sound sources can be in front of, behind, or even above or below the listener. This effect is achieved by using a suitably accurate model of the human acoustic system, such as a dummy head with microphones embedded in the ears (Plenge, 1974). Because of the better model, the sound waves thatarrive at the eardrums during playback are a close approximation of what would have actually arrived at a listener’s eardrums during the original performance. Along with greater realism, binaural sound conveys spatial information about each sound source to the listener. Furthermore, when sounds are spatially separated, a listener can easily distinguish different sources, and focus on those sourceswhich are of interest while ignoring others (Cherry, 1953). If sounds can be recorded in this manner, an obvious next
1. In strict terms, “binaural” and “stereo” mean exactly the same thing—two channels of sound. However, in the music recording field, stereo generally means recorded with multiple, spaced microphones and binaural generally means recorded through a dummy head.

Much of ourinformation about the position of a sound source is based on the effects of resonances inside the ear and refraction in the vicinity of the head and upper body. The dominant cues provided by these head and upper body effects are interaural delay time (IDT) (Rayleigh, 1907), head shadow (Mills, 1972), pinna and ear canal response (Gardener, 1973), and shoulder echoes (Searle, et al, 1976). Together,...