Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!wuarchive!uunet!mcsun!ukc!icdoc!qmw-cs!gn!tmn
From: tmn@gn.UUCP
Newsgroups: comp.music
Subject: Re: bi-neural audio
Message-ID: <1179000002@gn>
Date: 26 Jul 90 13:48:00 GMT
References: <1497658397@ERICM.90Jul25114931@hendrix.ftco>
Lines: 75
Nf-ID: #R:ERICM.90Jul25114931@hendrix.ftco:1497658397:gn:1179000002:000:4354
Nf-From: gn.UUCP!tmn    Jul 26 14:48:00 1990


"Bi-Neural"? What an amazing name! Sounds like a stereo version of Pat 
Flanagan's Neurophone! 

But seriously, folks, do you mean "binaural"? If so, the technique is to 
record actuality with a 'dummy head' fitted with mics in the artificial
ear canals. The two-channel signal is recorded normally and played back via
headphones. 

There are advantages and disadvantages with the technique.

The more accurately you model the dummy head the more individual-specific
the effect is. A basic head will give a reasonable effect for anyone, while
a highly-tuned head will give wonderful effects for some people and gross
errors for others (generally the person whose head was modeled gets the best
effect).

The system is almost useless on loudspeakers, unless you face the speakers in
to each other and sit between them (in which case you might as well have
cans on). Binaural can, however, be transcoded into a speaker-based surround
system, notably Ambisonics (a British-designed surround system based on
Blumlein's original stereo research) with great success, the simplest method
being to place the two channels at the east and west edges of the soundfield
and then reduce the radius vector until best results are obtained. However
this is not as accurate as a purpose-build transcoder. The resulting Ambisonic
B-format signal can then be UHJ 2-channel encoded and treated as stereo if 
desired.

The main problems encountered in dummy head work is that good modeling 
requires some special components: some researchers believe that internal
head density is important and have simulated this with fluid bags inside the
dummy head. Ear canal modeling has always been important and it is pinna 
differences that are generally regarded as making the system person-specific
(see above). 

Another consideration is that the signal actually passes through the ear canal
twice: once through the artificial one and once through your own (if you are
wearing cans). This can be compensated for in binaural-to-ambisonic transcoding
but in a traditional headphone-listening environment an equalizer is 
required along with a phase-compensation network which can adjust phase
with frequency. The curves for both of these are available in the literature
and notably in the Sony and Matsushita patents of the mid-Seventies.

Interestingly, neither company has exploited these discoveries commercially.
The best commercial exploitation of binaural techniques to date was that
by Hugo Zuccarelli, an Argentinian based temporarily in Europe during the 
early Eighties. Zuccarelli called his system "Holophony" and was not 
prepared to discuss what went into the system in detail: what he did say
is generally regarded as meaningless pseudoscience designed to cover up
what he had actually achieved to avoid it being ripped off. However when
he split up with his former collaborators and, after extensive legal difficulties
moved to LA, his former partners confirmed the view that Zuccarelli's system
was based on fully-implementing head modeling, phase and EQ compensation,
internal densities, etc etc as described above. It has been suggested that
another reason Zuccarelli was never forthcoming on what was _really_ involved
in his system was that he may have infringed the Sony and Matsushita patents
inadvertently in the course of his own research (it is not believed that he
ripped them off).

For some reason every project that he was involved in became the subject of
litigation, and ultimately the system faded from view. It was, however, the
best binaural system to have been used commercially to date.

There are some distinct possibilities in combining binaural and ambisonics,
as binaural is particularly good on nearby (<1m) moving sources while
the ambisonic Soundfield mic tends to de-localize when close to the capsules.
Ambisonics, on the other hand, can be used with conventional multitrack and
mixdown systems via Ambisonic panpots and other devices which can be 
computer-controlled and simulated (viz the work done by Dave Malham at the 
University of York, UK) while binaural requires actuality recording and 
is difficult to synthesize or use in a multitrack environment (binaurally
recording the sound of speakers in a room is generally disastrous: this is
better achieved with a Soundfield mic).

-Richard Elen at The Music Network (tmn@gn.UUCP)