One standard method for rendering an auditory display is playback of previously generated audio files. An advantage of this method is that it allows for precisely controlled studio engineering involving recording, synthesis, and processing techniques that would be computationally prohibitive for real-time rendering. In addition, tools for editing and playback of audio files are readily available under virtually any operating system. A disadvantage of this method is that multiple playback of the same audio file, over and over, can become tiresome to the ear. Moreover, while disk storage generally costs little, the management of large numbers of audio files is unruly and inevitably error-prone. Finally, no matter how many different audio files are used, there will always be holes in the mapping of data features to acoustic features if the data or process to be mapped is even slightly complex. This becomes more and more the case as we move toward analogical displays.
One way to deal with this situation is to employ already existing (and for the most part, free) audio tools which allow for real-time manipulation of stored sound files. This can be a viable solution for "symbolic" auditory displays representing discrete events and process states. Through the manipulation of a single sound file, one could represent many different states of a display object such as an icon or mouse cursor. However, this approach still suffers under the requirements called for with analogic displays. In order to deliver effective analogic representations, an audio rendering system is needed which can be deeply mapped to the model being represented.
At first glance, MIDI-based synthesizers offer an appealing alternative to the use of sound files. Using MIDI, a synthesizer can be triggered by mapping data to particular MIDI messages [8]. The problem with MIDI, however, is two-fold. First, one must accept the sounds that are given by the particular synthesizer technology at ones disposal. These sounds may differ, depending on the particular device being used. Second, MIDI is limited by its narrow control bandwidth and thus cannot be used for the display of highly multivariate and rapidly evolving data [10]. This has a secondary consequence which is that while one might solve the problem of mapping multivariate data to sound through the use of system exclusive messages (messages that are manufacturer-specific), these messages exacerbate the problem of narrow bandwidth since they usually require higher bandwidth per event.
What is needed is a way of rendering audio that is to auditory display
what a system like OpenGL is to graphical display. Moreover, tools
are needed with which auditory display designers can test hypotheses and
experiment with different methods by which a data model might be rendered
acoustically.