naturerecordists
[Top] [All Lists]

Re: Re: again about compression, actual hearing

Subject: Re: Re: again about compression, actual hearing
From: Marty Michener <>
Date: Sat, 06 Apr 2002 15:01:12 -0500
At 12:18 PM 4/6/02 -0500, you wrote:
>This is exactly the sort of problem I was describing. I'll have to see
>what happens with the UA 30. I know it gives nothing that's audible, but
>the difference test is much more sensitive.
>
>This area is one of the problems of digital audio. Unlike the popular
>belief digital audio does not guarantee that your signal will be
>unchanged. I happen to believe that we got rid of more problems (that
>were in analog recording methods) than we gained from digital. Making
>digital worth it.
>
>Walt

Excellent points, Walt, and can certainly be argued either way.  The one 
thing that digital has always screwed up royally is arrival times!

compare this crude signal:

- - - - - - /= = = =\_______

sampled at these points:
  |      |      |     |     |    |     |
and then "restored to analog" as:
                _ _ _ _
- - - - - - =/           \ - - -______
the point here, is the arriving times are shifted ahead and back at random, 
as it all depends on the exact sampling time, and not on the actual sound 
arrival times.

Much of the animal acoustic research since the 50's has repeatedly 
discovered that arrival times meant everything to many biological 
analytical systems.  Especially, bats, whose published nerve analyses of 
echoes is repleat with references to sloped-loudness-form-matching. But 
that is another story (D.Griffin, N. Suga).

In 1965, Dwight Wayne Batteau at Tufts showed to my satisfaction (on me) 
that the human ear, although limited to upper sine wave detection around 20 
kHz (ah, to be young again) actually analyzes sharp pulses using much 
shorter arrival times than the 20 - 20k range would ever predict (down to 
25 microseconds).

Specifically, he used a human head model, with pinnae molded to exactly fit 
a single person.  He then recorded, using near-video quality techniques, 
from two tiny eXpenSive mics embedded inside the artificial head.  Using 
rich sound sources, such a jingling a ring of keys, he moved the sound 
source toward and away from, and all around, up and down, the head 
location, while recording and keeping a video record of his actual planar 
coordinates of the key locations in time.  If you think your recording 
system is better than mine, record some jingling keys and listen to the 
result.  Don't worry, they all sound that bad!

The person would later listen on headphones in another room, while the 
recordings were played back, binaurally.  Not only could the person point 
nearly exactly to the direction in which the key jingling had come, they 
could pinpoint the distance as well, with some accuracy.  This, as far as I 
know, has never been explained to my satisfaction by hearing theory, before 
or since.

Wayne told me (now here comes the interesting part):  He could impose or 
bypass line filters on the headphone output circuit.
If people "can only hear to 20 kHz" (the PREMISE that all later digital 
sound HiFi sampling is based on) then he should do fine imposing a sharp 
filter in the headphone line of 25 kHz.  Right?  NO.  NOT. The position 
sensing went away, as I remember his explaining it, when the line filter 
came below about 80 kHz, and was very, very crude when he got into the 20 
kHz range.

He also manipulated the pinnae shape, (it didn't hurt, it was only a rubber 
model) and it changed the recordings he made adequately to also ruin much 
of the subject's position sensing.  He interpreted this as meaning: the 
very short reflections on the major and minor features of each person's 
pinnae contribute echo arrival information that allows near 3 dimensional 
direction decoding.

But, alas, all this is gone, even at the famous new 24 bit 96 kHz 
"standard".  Why? because the filters that are imposed after the digital 
signal is turned back into voltage and before it is turned into sound 
pressure, can at best represent a sample every 11 microsec.  The echoes 
Batteau was talking about, created by pinnal features with dimensions of 2 
to 8 mm, create sound echo delays on the order of 5 to 10 
microseconds*.  At 96 kHz sampling rate, these are all lost in the filter 
ringing.

But, luckily for me, I can no longer hear any of this.  Good luck youngsters!

my very best,

Marty Michener
MIST Software Associates
75 Hannah Drive, Hollis, NH 03049

coming soon : EnjoyBirds bird identification software.
* these calculations all come out of the fact that a wave of 30kHz, going 
about 1000 ft per second in air, will have a wavelength of 1 cm, and an 
inter-peak interval of about 33 microseconds.  So 5 mm travel or bounce 
distance, becomes an interval of 16 microseconds, etc.  These are worst 
case calculations, the analytic system having to do much better than this, 
about 1/10 of a cycle, to come up with beam-forming information 
analysis.  But, after all, that it has been doing for millions of years, 
before engineers came along to measure and mis-understand it all.




________________________________________________________________________
________________________________________________________________________

<Prev in Thread] Current Thread [Next in Thread>
Admin

The University of NSW School of Computer and Engineering takes no responsibility for the contents of this archive. It is purely a compilation of material sent by many people to the naturerecordists mailing list. It has not been checked for accuracy nor its content verified in any way. If you wish to get material removed from the archive or have other queries about the archive e-mail Andrew Taylor at this address: andrewt@cse.unsw.EDU.AU