[Im_Beta_00[_2_]]

Richard Crowley wrote:
Im_Beta_00 wrote: I just want to erase the roar of the car,

You simply cannot remove wideband noise without doing
significant damage to the wideband signal of interest. Dunno
how many different ways we have to say this before you can
understand it?

I'm used to mixing lots of stuff in with the Boolean XOR function;
everything that can be mixed in, can be mixed out, you just have
to remember which sequence you did it. But that's video data.
The same thing ought to apply to sound, too. But your adamant
opposition to this, implies I am missing something really obvious.
(As for losing bits of precision, that's another thing altogether;
it's probably really annoying to people with perfect hearing.)

or at worst, detect the
moment of the roar, and replace the relevant snip with a snip from a
different microphone (one that is not pointed in the direction of the
roar.)

Lot of work for absolutely nothing. But apparently you will
need to do the experiment for yourself since you refuse to
learn from the mistakes of others. Good luck.

If you have another mic picking up the signal of interest, but
without the noise, then why not just use that one and dump
the intrigue?

Well, that *is* something I have been wondering about- Whether
the microphones (I have many more than I mentioned previously)
are better used for recording the "signa"l as opposed to the
"noise."

It is not simple subtraction, and it isn't simple ANDing off the noise
and ending up with some miraculously "clean" track. If you thought
that was what I was getting at, I apologize.

You have not mentioned any viable methodology of doing what
you say you want to do.

I was planning on using the Boolean XOR function to extract the noise;
performing the function twice, just happens to return the sound to its
original state:

A=Input#1 (mic pointed north) is the cast's dialog, up close from a boom
B=Input#2 (mic pointed directly east) is the cast's dialog from a
cross-direction, about 10 feet away
C=Input#3 (mic pointed directly west) is the noise of the traffic

I am relying on the following formula:

D = B XOR C

and once you have done that, the following formulas will work:

B = C OR D
C = B OR D

One of the problems I am facing, is the sampling rate. That cheap recorder
only samples at 48K. The other problem is the amplitude (or "loudness").
None of this bit-twiddling will be automatic. Attentive listening to each
of the files is in order. I want A to be the loudest, B the second loudest,
and C the quietest. Unfortunately, a car driving by with a really bad
muffler is going to wash out everything, and there is nothing that can be
done about that. But otherwise, if it is just a reasonably well-tuned car
driving in an ordinary manner, the noise recorded at 'C' is going to be a
whole lot quieter than A and B.

If you have any experience normalizing 80 bit integers so they can be juggled
around in floating point operations, you'll understand what I was talking
about, when I said I had to adjust their volumes (amplitudes) before I
attempt performing the B XOR C function..... The bits have to be the
same "widths" before you do Boolean stuff with them.