Field Auralization is the process of extracting and displaying meaningful information in the form of sound from data. Through not only visualization but also auralization, users may have better understandings of the volume data, especially when it is visually complicated. Different from visualization, auralization maps volume data to some sound pressure values. In this work, a volume auralization technique is introduced, which objective is at the synthesis of physically meaningful sounds from time-varying volume data. During the preprocessing step, we detect sound sources at each time step, from which the acoustic volume data is computed. From this acoustic volume data, we synthesize and localize sounds as users navigate within the volume space. Our work shows the techniques for source detection, acoustic volume generation, rendering, and localization.

>> >> (a) Synthesis (b) Rendering (c) Localization [ Fig. 1 Steps of Auralization ] As shown above, our auralization method is composed of three major steps: Synthesis, Rendering, and Localization. During the first step, synthesis, acoustic radiation from sound sources are computed, which may result from many physical phenomena. Vibration of object surfaces, for example, results in the vibration of surrounding air which is propagated out to the space. Sounds from this kind of physical phenomena can be synthesized by analyzing the vibrational modes of the objects or the medium and summing up all of them. Sounds can also be generated from turbulent flows such as aircraft jet engine. There is no vibrating boundaries or objects that generate disturbance in the medium. Rather, the turbulence forms a region of quadrupole sources. Accurate analytical solution of sounds from such sources was first given by Lighthill. The most prominent result of his research is that the turbulence shows the quadrupole behaviour. Once generated, sounds radiate into the medium as acoustic waves. During the propagation, sound experiences various physical phenomena such as reflection, refraction, attenuation due to absorption, and diffraction. Therefore, sound heard by users should be quite different from that generated by sound sources, depending on the distance sound traveled, objects that sound reflected or transferred, the type of the medium, etc. Hence, during the second step, rendering, we compute sound heard by the listener, taking all of these effects into account. For example, a beam-tracing based algorithm is used to model sound propagation in a virtual environment \cite{funkhouser:1998, funkhouser:1999}. This method becomes difficult to deploy if the objects in the space are composed of a large number of small surfaces for smooth curves due to the increase of the nodes in the beam tree, i.e. explosive fragmentation of the beam tree, or if the sound sources are moving due to the dynamic updates of the beam tree. The last step is localization. Until upto this step, sound is calculated for users in virtual environments, which is independent of the actual configuration of speakers in the real environment. However, real world speakers generally can't exactly simulate the sounds in virtual environment because they are not exactly located at sound sources in virtual environment. We therefore have the need of a nice mapping of virtual sounds to a distribution of audio signals for real world speakers to give the best approximated spatial aural images to the audience. This is what is done in the this step. However, this has been a long challenging problem in acoustics because multiple loudspeakers do interfere with each other and this makes the process of analyzing their collective effect difficult.