Noise Reduction of background hummng

[LarryLaser]

I am a new user of Audacity 1.3 Beta, and any Audacity.

I have searched and read many articles on the Noise Reduction process, but I can not figure out how to remove the Noise Reduction in the speach section.

I have tried the settings of the Noise Reduction, by selecting the "background" noises and click the "Get Noise Profile" button,(using the standard settings, and tried different settings as well), and it will only remove the noise in just the silent sections. the noise is still in the voice background.

Background humming.PNG (26.59 KiB) Viewed 1245 times

Is there any more settings that can remove the noise throughout the whole files??

background noises-Setting.PNG (147.22 KiB) Viewed 1245 times

I have thousands of files (.mp3) of speeches that I am trying to clean up. So I want to "Batch Process" sets that are similar.

"another Question"
How can the "Plot Spectrum" settings be used to remove the bad noise settings??

Plot Spectrum.PNG (249.61 KiB) Viewed 1245 times

Thanks for any help.

[kozikowski]

Do you know that you can batch Noise Reduction? In Audacity batch is called "Chains" and not all tools and filters can be chained. It's not that someone wrote Chains and all Audacity tools magically batch. Each filter and tool has to be individually programmed to Chain, so each one is gold.

Mess with the "Smoothing" control. That's the control that allows Noise Reduction to act on one spoken word as one "thing" and not try to noise reduce each syllable within the word. Too tight and words get gargly. Too loose and the correction stops working right.

Capturing the profile is a big deal. The profile is a sample of Noise Only with no valuable performance. The tool will try to subtract the profile sound from the show. Sometimes it helps to Effect > Normalize the profile segment, get the profile and then return the segment to normal.

Koz

[kozikowski]

The spectrum view is frequently more valuable in Log rather than Linear. Many people pick 400Hz has the "middle" of all frequencies. In Log view, it actually ends up in the middle. Plus that view gives much better data down at 50Hz and 60Hz, the two hum frequencies.

Also, size can be much higher. I use 8000 or better to see individual interference frequencies.

Koz

[kozikowski]

Noise reduction and the other "rescue" tools work on the idea that you have a really good show with something minor or gentle wrong with it. Many people want the tools to turn garbage into a show. There was a poster several days ago that wanted to "clean up" his performance. When five of us listened to his work, three of us couldn't find any show at all.

That didn't work so well. The show has to be significantly larger and clearer than the trash to have any hope. Sometimes the best we can do is reduce the trash slightly.

Koz

[LarryLaser]

Thanks for your reply "kozikowski"

But I am not an advanced programmer, just some experience.
I am trying to understand the settings, but I do not understand very much of the data and settings yet.

Can you explain "How to set the proper data" so I can test the work?

[kozikowski]

I wondered if there was a tutorial on Noise Removal...and there is.

http://wiki.audacityteam.org/wiki/Noise_Removal

I usually do this by capturing the noise profile, selecting a sample of the speech and apply the tool with various settings and UNDO if I don't like it. I think everybody else does it that way, too.

There are some broad recommendations. Amount of Correction goes from no correction at all to heavy gargling and bubbling. Start in the 6dB or 9dB range. Attack and Decay affect the work before and after a word is spoken. Frequency Smoothing is a little magic. That one affects the quality of the correction inside words. Too little and words get gargly and bubbly, too much and noise appears during words.

The tool is something of a violin. It takes a little experience and touch to make it work. It's certainly not a one-click tool and it's different for each performance. We urge strongly to shoot works that don't need correction in post production.

All that and there was talk of a second reduction tool designed by someone else. I need to look that one up.

You can also try Steve's Noise Gate.

Effect > Noise Gate. That one tries to suppress noise that's more quiet than the show. It has the disadvantage of sometimes working too well -- it's not natural for the room to go dead quiet between words.

Koz

[Enkephalin]

I think I know what you're doing wrong, and it took me a while to figure this out as well.

When doing noise removal:
1. Select the area where it's just background noise, no desired audio
2. Effects -> Noise Removal -> Get Noise Profile
3. This is key - "unselect" the area where it's just background noise; nothing should now be selected.
4. Go to Effects -> Noise Removal -> OK

The missing part in what you were doing may have been step 3.

If you were doing it right all along, my apologies, and it may just be that you have a very noisy and difficult to manage file.

Good Luck.

[kozikowski]

I'm going with Noisy and Difficult. If you have a damaged voice track with broadband hiss, it's almost impossible to remove the noise during each spoken word. The noise signal will appear to pump and swish with the vocals.

You can certainly take the noise out from the words, too. That's the setting that sounds like space aliens. You can't actually Remove Noise -- ever. What the programs try to do is determine where the show is and remove the noise from everything else. This gives you hissy voices over a dead background if you have a bad show.

It appears to be really easy to go out and capture a live performance, but it's not, and you can accidentally burn in problems that nobody can remove.

Koz

[steve]

I have tried the settings of the Noise Reduction, by selecting the "background" noises and click the "Get Noise Profile" button,(using the standard settings, and tried different settings as well), and it will only remove the noise in just the silent sections. the noise is still in the voice background.

I think that your method is correct, but you're expecting more than can be achieved.

Imagine that you have an original charcoal sketch by a famous artist, but someone has scribbled all over it with a crayon. You want to remove the crayon but not damage the original charcoal drawing. Sure you can remove the crayon by rubbing vigorously with an eraser, but it will utterly destroy the original charcoal sketch. In this analogy, the charcoal sketch is the sound that you wanted to record and the crayon scribble is the noise. The Noise Removal effect can remove noise quite easily from the "silent" parts where there is scribble but no drawing, but during the speech section it needs to be a lot more careful so as to avoid destroying the speech along with the noise.

There is a little "trick" that you can do to persuade the Noise Removal effect to rub a bit more vigorously:
Make a copy of the "silent" area (noise only). You can do this by selecting the area then pressing Ctrl +D (or select "Duplicate" from the Edit menu).
Amplify the copied noise by about 8 dB using the Amplify effect.
Run "Noise Removal > Get Profile" on the amplified copy of the noise.
Then run "Noise Removal (step 2)" on the original track.

Because you have told the Noise Removal effect that there is more noise than there really is, the Noise Removal effect will scrub away a bit more vigorously. This will create a bit more damage to the original audio than would otherwise occur, but will remove more noise.

[kozikowski]

Amplify the copied noise by about 8 dB using the Amplify effect.

Why not Amplify at the default settings? Wouldn't you want the largest sample you could get short of clipping?
Amplify will not affect the dynamic range of the sample, so all characteristics would be preserved.

Koz