naturerecordists
[Top] [All Lists]

Re: noise reduction techniques

Subject: Re: noise reduction techniques
From: Walter Knapp <>
Date: Sun, 04 Jan 2004 17:36:07 -0500
From: Dan Dugan <>
> 
> Rich Peet wrote,
> 
> 
>> > 1. When there is no audio that is important above 15,000 cycles I
>>
>>> drop the gain in this region by 50% to 75% (6db to 10db) on a slope. 
>>> It does not effect the overall recording and hides the ragged
>>> compression of this area that can be seen on a spectral display which
>>> is the signature to many that the recording was a MD.  It really does
>>> not change the heard audio any, but for those that pay attention it
>>> fools a few of them into thinking they may be looking at a 24 bit
>>> recording with >100db s/n.  I use it with restraint but hey if it
>>> makes someone think that I am better and have higher tech equipment
>>> then I do then it was worth the effort.  This part of the edit is
>>> vanity.
> 
> 
> And Walt commented:
> 
> 
>>Unless they use sonograms, few would find the MD effect you speak of.
>>Nor is it consistently there. If it is in your equipment, then there may
>>be some other fault with the equipment. It is certainly not consistently
>>there in all the sonograms I've done on MD recordings.
> 
> 
> I monitor my work with Spectrafoo instruments, and the effect that 
> Rich mentions is quite apparent in spectrum analyzer displays of 
> perceptual coded stuff. It doesn't show on sonagrams. It's a 
> sure-fire indicator that you're listening to a perceptually-coded 
> recording, even though you usually can't hear anything strange. I'm 
> going to play with the rolloffs as Rich suggests, but there isn't any 
> audible need for it--as he says, he does it just to clean up the 
> appearance on an analyzer.

One could hardly call such equipment common. Pretty rare person who will 
have access to such.

If you could hear the perceptual coded stuff, that is a failure. A good 
coder will do the work and not be heard. I don't think they ever claimed 
to do nothing, only to not be heard.

I suppose it depends on your sonogram software. I do see the effect of 
coding in the high frequency end of some MD recordings. Mostly if there 
is something of fairly high intensity in that area. You do have to have 
sonograms that are fairly well resolved, and have to understand what the 
sonogram software itself puts in there.

I would be more interested in what specifically you are seeing in your 
spectrum analyzer. I've got some spectrum analysis software, but it does 
not appear to give me as good a look as a good sonogram.

I note in my reading of Ken Pohlmann's book that he's gone to triple 
blind listening tests, preferably with trained listeners. He's pretty 
down on hardware type tests, saying they miss the point. He also notes 
it's getting harder and harder, and that the unencoded sample does not 
necessarily come out on top in that work. They are doing the tests to 
improve encoders and evaluate one against another. They realize we will 
be working with such things.

He also has some ideas in how perceptual encoding can make systems 
better than without the encoding.

> Because I work in theater, I often have to do noise removal 
> processing, and I keep buying expensive software that does that. Even 
> the best is problematic, and I rarely do more than plain vanilla 
> equalization/filtering to my nature recordings. I would only use 
> dynamic noise filters on nature sounds if I was composing, mixing 
> cleaned-up specific sounds over a natural bed.
> 
> -Dan Dugan

I mostly don't filter either, the scientists I deal with could care less 
as long as they can make out the calls. I did use considerable filtering 
in the ID clips for the CD. Particularly in some cases where our choice 
of material was very limited. There you are not really trying to 
preserve the other sounds. You are trying to get the call isolated so 
people can learn it.

If I use a dynamic noise filter, it's the last thing in line. After all 
other cleanup is done. Even there it can only be used lightly. Very 
annoying filter to use.

I cannot afford to continue to buy different software, though SparkXL is 
a pretty expensive package in my thinking. Got lucky and got it off 
ebay. So only have to keep it upgraded. Having the support to experiment 
with expensive software would be nice.

Walt









________________________________________________________________________
________________________________________________________________________

<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