Audacity Forum

There is a test signal called "pink noise." I call it "Rain in the Trees." This test signal has all audio tones in it. That's its job. If you try to reduce rain in the trees, microphone hiss or tape hiss from a show, Noise Reduction will try to remove the whole show.

Noise Reduction is not the cure-all that it seems. That's why we object to calling it "Noise Removal." I don't think it's ever removed all noise.

Koz

We basically work in two domains: time and frequency.
The noise removal takes only the latter into account.
How do you make the recordings? Are the files in stereo or made from two recorders in mono? I ask because the actual sound source location can serve as a tracking key too.
Have you any links to such sample dialogs?

As far as source location...I know and have the camera one was taken on. I guess that would be in stereo?
And the other one is from a VHS conversion of a 10 year old tape, made by someone else, although someone in my organization did the converting.
Why does this matter?

Samples:
http://www.teachingdrum.org/adjul/Sample1.mpg
-The one shot with our camera. It has a fire in the middle of a circle of people talking, and some bird/nature noises. I'd like to wean the fire...I kicked out every frequency 10k and above and that seemed to take the edge off, and without dulling the rest of the sound - it seemed. I wonder about amplifying the frequencies that voice is focused on, but I haven't done this successfully yet.

http://www.teachingdrum.org/adjul/Sample2.mpg
-The VHS conversion. This particular segment has a lot of bird and insect noise, and the woman is talking low and not into the central mic. This is the worst of the worst. I'm guessing I can cut the highs and up the mids like above.

I've got an error (404) while opening the links.
The source location is as far helpful as sounds outside the center can be attenuated separetly.

typo:
http://www.teachingdrum.org/adjul/Sample1.mp3
http://www.teachingdrum.org/adjul/Sample2.mp3

How do you mean sounds outside the center? What is the center and how can it be located on the track?

cybrmarc wrote:typo:
http://www.teachingdrum.org/adjul/Sample1.mp3
http://www.teachingdrum.org/adjul/Sample2.mp3

On sample 2 You can notch out the mains hum for a start ...
then applying a wide notch at 5800Hz cuts back the bird noise ...

cybrmarc wrote:How do you mean sounds outside the center? What is the center and how can it be located on the track?

The center is the part of the sound that's directly in front of the stereo microphone.
The device is most often such placed that the speakers are in the center. A bird that sings from the left can therefore be discarded by isolating (or better emphasing) the center.
There's a quick test to check if the sides (hard left or hard right) contain important speech:
Use the "Vocal Remover" from the effect menu. If the resulting sound has less or no speech, then the center isolation will increase the audibility.
For your first sample, the second speaker disappears after this effect. The first speaker is actually on the left side alone. Thus, instead of left + right, we can assembly the stereo track from left + center and delete the right side.

Here's the second sample as it has been modified by Trebor. The last two stages are the stereo center alone and all that has been deleted (i.e. the side channels).

Robert J. H. wrote:... The last two stages are the stereo center alone and all that has been deleted (i.e. the side channels).

sample2 excerpt , 4stages.mp3

That's a very clean centre-isolation : I can't hear any digital artifacts , is that rjh-stereo-tool.ny ?
[ I use kn0ck0ut for that which usually adds some artifacts ]

Trebor wrote:

Robert J. H. wrote:... The last two stages are the stereo center alone and all that has been deleted (i.e. the side channels).

sample2 excerpt , 4stages.mp3

That's a very clean centre-isolation : I can't hear any digital artifacts , is that rjh-stereo-tool.ny ?
[ I use kn0ck0ut for that which usually adds some artifacts ]

Yes, it is indeed.
I've installed the kn0 plug-in too, just to compare with my own tool. Most often, it fails.
The side suppression is sometimes better but the artefacts are unproportionally noticeable.
I think that the tool uses bit masking for the source separation (i.e. either 0 or 1), whereas my tool uses soft masking.
It is in principle possible to create a filter for musical noise. Those artefacts are often isolated bins that have a short duration (window length of 4096 samples for instance).
It is a big field for experimentation. I currently try to attenuate percussive sounds. It works fine for bass and snare drum so far.