Re: HRTF failure

[Brian FG Katz <brian.katz@xxxxxxxx>]

Christian,

While there now seems to be a number of comments responding to your
question, I though that I would add a few more, as I think there are some
things which need to be considered. 

First, there has been some discussion now about headphone choice and
equalization. While this is important for coloration considerations when
trying to create a true-to-life recreation of a given sound source, for
strict localization it is less important. In addition, you test seems to
concentrate on the horizontal plane, and not vertical, where spectral cures
are more crucial. We have recently presented a paper examining the
importance of headphone choice/equalization and it is not the as important
as some may state. As you are already using a generic HRTF, there will
already be coloration by the pinnae which differs from the actual
individual. 

D. Schonstein, L. Ferré and B. Katz, ?Comparison of headphones and
equalization for virtual auditory source localization.? Proceedings of the
155th ASA, 5th Forum Austicum, & 2nd ASA-EAA Joint Conference, Paris,
29Jun-6Jul 2008. 

Second, are the comments regarding head-tracking. It is sure that
localization improves greatly with head tracking, and the lack of can pose
some errors, with front/back confusions being a serious effect. While adding
tracking is good for rendering, it rarely falls into the possible realm for
many studies due to the context, let alone the price. What I do note is that
you are using very long sound signals (6 sec noise burst). Localization of
noise bursts is predominantly dominated by the onset part of the signal,
rather than the duration. Brungart I think has done some studies showing
comparing pulse length and repetition number for optimal localization. We
typically use 150msec bursts. As relating to head tracking, if the pulses
are sufficiently short then head-tracking adds little benefit as
unintentional head-movements are not a problem. This assumes that the
subject is instructed not to move of course. 

Third, is the question of the importance of the HRTF itself in your test.
Again, as you seem to be performing a test in the horizontal plane, it is
the ITD which is of great importance. You should be aware of the fact that
any difference in head size between your subjects and KEMAR will result in
distortions of the responses relative to this disparity. Methods exist for
ITD individualization, even if you do not have an individual complete HRTF.
Selecting the best HRTF for each individual from a database is a good
approach for improving spectral cues, but the best spectral cues are not
necessarily in any way correlated to the best ITD cues. 

I think much has to do with the goal of your tests, to determine where you
should place you efforts. 

Regards,

-Brian FG Katz

---
Brian FG Katz, Ph.D
Perception Située
LIMSI-CNRS
BP 133
F91403 Orsay
France
tel. (+33) 01.69.85.81.23
fax. (+33) 01.69.85.80.88
direct.(+33) 06.27.30.67.60
e-mail Brian.Katz@xxxxxxxx
web_theme:http://www.limsi.fr/Scientifique/ps/thmsonesp/SonEspace
web_group:http://www.limsi.fr/Scientifique/ps/