Effect of Using Individually Simulated HRTFs on the Outcome of Tournament Selection Procedures in a Virtual Environment.
Individually fitting head related transfer functions (HRTF) play an increasingly important role in the field of virtual acoustics, especially for the acoustic layer in virtual environments (VE). Various studies have shown that well-fitting HRTFs have a positive effect on the perceived immersion in virtual (VR), augmented (AR) or mixed (MR) reality applications. This is especially relevant for music-related applications, where the acoustic domain is of particular importance. However, shortcomings in the determination of the individually appropriate HRTF may mean that they provide little or no advantage over generic HRTFs. This depends on many factors, such as the specific measurement procedure, the type of validation, or the inclusion of additional sensory modalities.
A central question in this context is often how exactly the individual fit is realized best. Acoustic measurements, the probably most obvious method, are usually very time and resource consuming, although new approaches have now been developed to rapidly measure HRTFs in ordinary home environments. Another technique is numerical simulation, e.g., using Fast-Multipole-accelerated Boundary Element Method (FM-BEM). In addition to the approach of indirect individualization based on anthropometric data, individualization based on perceptual feedback seems promising for practical everyday use, since HRTF selection in this way is relatively time-saving and possible without much technical effort.
In the current exploratory study, a tournament task is combined with numerically simulated HRTFs. The two main objectives are (a) to analyze the impact of numerically simulated individual HRTFs on the outcome of a tournament task that uses perceptual categories such as externalization, envelopment, etc., and (b) to investigate whether this correlates with the performance in a localization task in which the participants have to spatially localize sounds using the different HRTFs from the tournament task. Another general objective is to orient the methods used in the study towards a future application-oriented suitability for everyday use (e.g., with low cost 3D scanners comparable to scanners of future smartphones).
The study is divided into 3 consecutive parts. First, the participants' individual HRTFs are simulated based on a 3D scan, then each participant takes part in the tournament task to identify the best fitting HRTF, then the best (and worst) fitting HRTFs are validated in a localization task along with the simulated HRTF.
Although only a relatively small number of subjects participated in the exploratory study, the results of the tournament task show that the numerically simulated individual HRTFs had little effect on the outcome. In the localization task, however, the simulated HRTFs produced the best results, on average.
Factors that could explain why numerically simulated custom HRTFs do not perform overwhelmingly well in the tournament task are the random selection of HRTFs from different databases, the specific tournament mode, the rating categories or the visual/acoustic experimental environment within the VE. Future studies may probably need to pay more attention the specific design of the acoustic environment used in the tournament task. However, numerically simulated HRTFs created with a methodological approach oriented to everyday usability seem to be a promising approach for use in (acoustic) VE.