Richard Lee wrote:
Just to bring everyone down to earth ..
There are two easily reproduced experiments first carried out by prominent
members of this group which put these effects into perspective. They are
the
Greene/Lee Neckbrace
and
Malham/Van Gogh Experiment
The first shows 'real life' Fixed Head Localisation (which matched HRTFs
address) is TERRIBLE. Many people can't even distinguish back/front with
perfect (measured on their own noggin) HRTFs ... or even in 'real life'
with a Greene/Lee neckbrace. Anyone who has done fixed head localisation
experiments finds this out real quick.
The second shows that even the tiniest amount of head movement improves
localisation immensely and any ambiguity due to mismatched Pinnae etc (and
YES, the pinnae colouration effects are chaotic) are INSTANTLY resolved.
No 'training' is necessary with head movement.
https://mediatum.ub.tum.de/doc/1207048/1207048.pdf
The bottleneck for an immersive binaural sound synthesis is the
acquirement of individual
Head-Related Transfer Functions (HRTF). Manifold HRTF approaches to
circumvent the
obstacle of HRTF measurement are topic in research. The localization
performance of three
more or less individualized HRTF-datasets is related to the
individually measured HRTF. An
intuitive experimental design, using laser pointing for indicating the
perceived sound source
direction, is introduced to evaluate those individualization
approaches. Besides the impact of
head-tracking is investigated. The results show that the azimuth
localization error does not
differ significantly for an generic dummy-head HRTF, the individually
measured HRTF and
the HRTF selected from a certain set of other person’s HRTFs, if
head-tracking is available.
In the case without head-tracking the measured HRTF grants the
smallest reversal rate. This
results can be taken into account for designing an binaural auditory
application
...
In general the mean localization error for all data is comparable to
prior researches.
< Considering the results for the elevation localization, it was found
that the performance is
significantly better for the noise stimuli and also for the individual
HRTF. In general the
elevation directions were underestimated. >
...
For the HRTF-sets it can be stated that head-tracking reduces the
differences in localization
error. < No significant differences between measured HRTF, KEMAR HRTF
and selected
HRTF for the azimuth error could be found. > In contrast to that the
differences in the means
of the azimuth error are significant for all HRTF-datasets if
head-tracking is not available,
whereas the lowest error was obtained using the individually measured
HRTF.
Thanks (very obviously) to the Technical University of Munich, which
still "dares" to publish research which is funded by the tax payer.
(More and more universities seem to be bound by NDA agreements, even if
participating companies don't contribute anywhere to 50% or more to the
costs.)
At 1st reading, R. Lee seems to have some support...
It could be that we still need personalised HRTFs, in spite of HT.
(Elevation cues; spectral/pinna cues needed for localisation in the
median plane.)
Best,
Stefan
Even vertical localisation, for which Fixed Head HRTFs have the most
benefit, require a priori knowledge of the source spectrum. I've done a
small amount of work involving victims ... I mean subjects ... blindfolded
and tied up face down on anechoic chamber floors which show the first pin
drop is impossible to localise. Second and subsequent pin drops are much
easier.
I'll point out that Gerzon had Fixed and Moving Head versions of all his
Localisation Theories. The infamous Energy and Velocity vector
'magnitudes', rE & rV, are in fact a measure of the correspondence of Fixed
with Moving.
His Energy and Velocity models incorporate (give the same results as) ALL
the existing Localisation models except for the HF interaural delay
(Transient) and Colouration ('HRTF') models.
One of Gerzon's most important contributions is that he shows the
equivalence of the full Moving Head models like Makita, which assume the
listener will fully turn to face the sound ... with the models that only
assume small involuntary head movements. See "General Metatheory ... " for
the nitty gritty.
____________________
If you make some B-format recordings with a properly aligned Ambisonic mike
like TetraMic, you can test some of this for yourself. Have lots of things
happening all around including up & down. Use headphones and the crudest
possible Binaural decoding ... slightly hyper cardioids at about 150.
You will find about 10% of the population have difficulty with front/back.
But let these guys twiddle the Azimuth & Elevation controls on VVMic
themselves and they immediately become happy with the scene ... even before
they work out VVMic's slightly quirky interface.
____________________
CONCLUSIONS
If you have Head Tracking (ie Moving Head Localisation), don't bother with
fancy HRTFs.
Eric Benjamin found that you get most of the benefits from just getting
head size right but even this isn't necessary if you have Head Tracking.
Blumlein shuffle probably worth doing as you essentially get it free with
your simple IIR implementation.
Fancier HRTFs will need EVIL FIRs to be of use. You will lose any chance
of 'real time' and muck up the experience for 'mismatched' listeners.
Expect only small (if any) 'improvement' for the huge extra.computing load
to interpolate between HRTFs.
If you haven't got Head Tracking, GOTO Head Tracking
This covers all the Virtual Reality applications. The Video Game people
like Simon Goodwin of Codeworks have been doing it for at leas a decade
with 3rd Order HOA IIRC.
If you insist on fancy HRTFs and Fixed Head ... do you seriously think you
can improve on the listening experience of present & past generations of
listeners, who have listened to 'music' over ear buds for more than a
decade ... with fancy HRTFs ? GOTO Head Tracking.
____________________
SPECIAL OFFER
Send $500 in used bank notes to me at Cooktown Recording and Ambisonic
Productions mentioning Sursound, for a sample Greene/Lee Neckbrace and
Diamond Encrusted Malham/Van Gogh cap. Golden Pinnae are an extra cost
option on the last item. No Confederate money please.
_______________________________________________
_______________________________________________
Sursound mailing list
Sursound@music.vt.edu
https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here, edit
account or options, view archives and so on.
