At 09:48 PM 3/9/02 -0500, Walt wrote:
>Which, of course gets us right back to hunting that perfect environment
>and perfect mic to record it with. Or record in stereo with good mics.
>Then the truck sounds like a truck coming by. A great deal of why it
>does not sound like it is not really sound reproduction as much as that
>in many cases it's mono, all sounds just dumped on top of each other.
>Our brains are used to processing a soundfield. If they have a
>soundfield, they will process the background sounds and locate them
>distant from the bird that they are trying to listen to. We don't hear
>just the bird we are concentrating on, we hear it all, but our brain has
>been told we are only interested in the sound from this one spot in the
>soundfield and is well up to the task of sorting that out and giving us
>what we want.
Excellent point that, as you say, we have discussed before, and certainly
involves the lay listener in the space more, so the brain can attend to the
bird as a point source instead of the speaker location. All my original
recordings for the last two years are stereo (crossed ME-67's at about 40
degrees) and I should try this test on people not used to hearing field
recordings. Thanks for the reminder, Walt.
HOW TO MINIMIZE ROAR COMPLAINTS, in a mono cut.
In my software situations, stereo must be replaced by mono for disc space
reasons. My guideline for making recordings more immediately acceptable to
lay listeners is this: Make all the background noise the same intensity,
so there are no frequencies much louder than any other. In natural sound
situations, this usually reduces to this practical editing: to make the
lower frequency part of the recording all the same intensity (on a log
plot) as the noise intensity "behind" the song. Essentially: make white
noise out of roar.
Say, the spectrum behind the bird sound appears dark pink, but there are
bands of bright pink from traffic, leaves blowing in the wind, etc. You
build a broad filter curve that reduces these higher bands. There is less
noise in the final, relative to the bird, but with flat distribution laymen
hear just a white noise hiss throughout, and that they seem to expect or
understand.
Directions for using Cool 2000 or Pro to do this: (this is a bit detailed,
but not difficult)
Create a curve in your mind from the spectral display (Spectrum set C.E.
2000 or Pro to these settings: [F4], spectrum tab: Windowing:
Blackman-Harris, Resolution 256, window width 100%, Spectral plot style:
(x) Logarithmic, Range 130 dB; ( ) Linear Energy plot Scaling 0.8%; (
check ) Reverse Color Spectrum Direction. [ Ok. ]
On my monitor this produces the colors and visual densities I find most
readable. Then use this spectrum to view all the frequencies BELOW the
lowest frequency the bird makes, call this Fmin. I usually blow up the
image using the ( + ) button in the extreme lower right, so the lowest part
of the spectrum is now 4 times taller. You might also want to switch the
above settings for more vertical resolution (e.g. 1024 or 2048 instead of
256 recommended).
A real, new example here will be best. I am looking at a mono field
recording where Fmin is 1400 Hz, and am using 2048 spectrum resolution and
4x vertical magnification. I can seen the brightness of the pink behind
the songs (1500 - 3000 Hz) is darkish pink. My goal is to reduce all the
noise frequency areas to that same level, "to remove the roar" we are
talking about, perceptually.
So I prepare a mental curve I want to match, by analytically viewing the
spectrum and jotting sown a few notes.
From 1400 - 800, 0 dB; 800 - 500 transition to -15 dB; 500 - 300 flat -
15dB; 300 - 180 Hz transition up to -5 dB; 180 - 60 flat - 5 dB; 60 -
0 Hz transition to -30 dB. Now I made these numbers up without trying
anything, lets see how close I came.
I press Ctrl-A to select the whole cut, then click Transform -> Filters -> FFT.
The setting right now I am using are FFT ( ) Linear ( X )
Logarithmic; [ X ] log scale. Max 20 dB Min -20 dB. This produces a
rectangle window, bottom axis labelled frequency, side axis labelled 20 dB
at top, 0 middle and -20 bottom, with a green line running at 0dB from left
to right.
I start with a flat curve 0 to 22.5 kHz, and click [ o ] Log Scale, giving
me more detail in the low band. Now I click and drag new nodes on the green
line to agree with the above list of transition points:
800 leave at 0 dB; 500 drag down to - 15 dB; 300 also at - 15; 180 up to
- 5 dB; 60 also at - 5; then zero down to -20
Now I click OK and in 3 seconds I have the new spectrum in front of
me. Actually, not too bad, looks near the same brightness pink from bird
song to 0, but the 400 - 600 Hz is too dark (too much removed). No problem.
I save the file, press Ctrl-Z to undo it, hit [F2] to repeat the FFT screen
with all the settings ready to be modified. I move the 500 Hz node up to
-12 dB and OK to see the results. This is as close as we need, so I
re-save it and listen carefully, at full speed and half speed (Alt-E, A,
move cursor from 44100 down to 22050, OK.). Fine. Total time to type it up
30 minutes, to do it 10 minutes.
For many situations, I reduce the whole bass area more greatly, esp. if the
overall noise is quite high. This gives it a low-fidelity kind of sound
that many professionals hate, but when it comes to children or non-techies
trying to just hear and learn a sound this seems preferable.
my very best,
Marty Michener
MIST Software Associates
75 Hannah Drive, Hollis, NH 03049
coming soon : EnjoyBirds bird identification software.
NB I am now getting posts within 1 minute of send time.
PS Just read "Galileo's Daughter", where Pope Urban II ? around 1630, was
so distraught with the 30 year war, he wasn't getting enough sleep, and he
had all the birds in St. Peters Basilica court yard killed so they would
not make nocturnal noises. Such leadership! . . . ;-)
________________________________________________________________________
________________________________________________________________________
|