On the Use of Eye Movements in Acoustic Source Localization Experiments

The ability of virtual auditory environments and sound reproduction techniques to provide well localized virtual sources is a key feature. Experiments that judge the localization are hence amongst the first ones performed in evaluation. One difficulty in such experiments lies in reporting the perceived direction by the test subjects. Various exoand egocentric methods have been developed and used in the past for this purpose. See e.g. [1, 2] for an overview. Applied exocentric methods are for instance marking the perceived direction on paper or using a graphical user interface. Egocentric methods are for instance hand-pointing, head-pointing [3] and eyepointing [4]. Also combined headand eye-tracking has been used to obtain the absolute gaze direction [2, 5]. We aim at a method that allows to perform source localization experiments with mostly untrained subjects that are restrained as little as possible in order to obtain results under natural conditions. Highly accurate and unobtrusive techniques for head tracking are commercially available. Most of the published studies based on eyepointing used the scleral search coil technique [2, 4, 5] for eye movement recording. Besides its high accuracy this method requires to insert a contact lense containing the coil into the subjects eye which can create quite some discomfort. In this study we used a commercially available camera-based head-mounted eye tracker. It is hence less obtrusive than the scleral search coil techniques. The eye tracker provides the relative binocular gaze direction at high spatial accuracy and sampling rate after a short calibration. We combined the relative gaze direction of the eye tracker with a head tracker in order to obtain the absolute gaze direction. In order to evaluate the accuracy and applicability of this setup, a basic localization experiment was conducted where the users had to report the perceived direction by fixation of the presumed source direction. This paper will present the general setup, as well as results concerning the accuracy of the method.