Noise Reduction

Adds odd effect

So I recently discovered that Goldwave, which I use for my external wave editor, has a noise reduction filter. It allows you to copy a portion of the recording with only the noise you want deleted, then apply that section to eliminate or reduce that noise in the entire file.

When I record I noticed that there is a “noise” in the lower band below 200, so instead of just eliminating it with a high pass, I thought I’d use the noise reduction filter. Here’s the problem. When I apply the filter to the file, the resultant file sounds like it is recorded in a coke bottle, just not quite right. Some of the “liveness” seems to get lost. What’s going on here?

You might be taking too much out, I’d decrease the amount of reduction and see what happens.

I did reduce it to 50% and also 30%, but the effect is still there, just less of it.

I don’t know what settings are available in Goldwave, but there should be one labeled “FFT size” or something like that.
If this value is too small, you can get too much “good” audio removed together with the noise.
If the FFT size is too large, the result can start sounding like it has been processed by a cheapo reverb plugin.
A reasonable trade-off should be a value of 4096 or 8192.
Experiment!

Also, see if Goldwave has a way to set a transition margin for the noise removal.
The idea is to get a “soft” onset for the noise removal (as opposed to a rock-solid treshold for each freq. band).

Another thing to check is if your “noise only” snapshot really has only noise in it.
If, by accident, it also has a tiny bit of “good” audio in it, too much will be removed from your entire track.

Hope this helps a bit.

I accidently reposted the Hiss Removal thread. It has the settings I use for NR in Goldwave.

Sorry everyone.

cliff
:O

these are the settings

go to Noise Reduction, selecting “Use Clipboard”, FFT=12, Overlap=95, Scale=100

hth

cliff
:cool:

Quote (valrecorder @ Jan. 17 2006,22:26)
…When I record I noticed that there is a “noise” in the lower band below 200, so instead of just eliminating it with a high pass, I thought I’d use the noise reduction filter.

Noise reduction is a difficult complex process, won’t work 100% in all situations, and as you’ve discovered, has the potential to adversely affect the desired signal. So you shouldn’t really use noise reduction until you’ve tried other less harsh processes first.

I would approach your current noise problem in the following way

1) (Recording chain) Try to determine the source of the sub-200 Hz noise and reduce or eliminate the source. You can try to re-wire the inputs, change or move cables, try different inputs, etc. you can also analyse the noise to see if it contains specific frequencies (eg 60 Hz and multiples) , which would point to a grounding issue, or something else related to AC power.

2) if the noise source can’t be found or reduced when you record (or you have noisy recordings you need to salvage) analyse the noise to determine what it consists of - is it just a set of discrete frequencies, or just a rumble/roar in the low frequencies? If it’s discrete frequencies, I would kill them with a series of sharp high-Q filters; for a rumble, you can try a hi-pass or a broad notch across the problem area.

3) if the above EQ didn’t clean out all the noise, or it’s broadband noise, then this is the right point to try noise reduction. Don’t try to get 100% clean in one pass; this almost always leads to artifacts. Instead, try for a setting that audibly removes some of the noise without altering the desired signal. I found I get much better results by doing several light passes, checking the paramaters each time.

Also remember that you cannot always get 100% clean. The goal is to reduce the noise but not damage the signal.

For analysing and cleanup, these days i mostly use CoolEdit2000, which is similar to GoldWave or SoundForge.

(Industry anecdote: I used to work with the Sonic Solutions No-Noise software many years ago. On one job, we were asked to help transcribe some music from an Arthur Fiedler/Boston Pops LP. After doing the transcription, we used No-Noise to reduce the clicks and other vinyl noise. We got good results but there was an odd low rumble in places. After trying a bit more denoising and doing some intense listening, we were able to determine that the low rumble was trucks going past the studio while the orchestra was recording :D . )

Great Advise Ken!

Just one thing: be sure that in the “noise profile” you take, there isnt any little portion of “good sound”, you can been saying to Goldwave to remove that too.
I agree with Ken that Cooledit has a very nice noise reduction. If you dont have it, google for “Cooledit 96”, and old but functional demo of it that can do the trick (or at least can be nice compare with Goldwave)

Thanks all for the advice.

On further reflection, I believe I may have come on to something. I live 3/4 block from a freeway. Though not audible in the house, I may be picking up a low frequency rumble. On occasion a picture frame will rattle from what I suspect is the freeway. Is it possible that even a vibration would be passed through my mic stand to my LD mic? If so, would a shouckmount help? I am not currently using one.

best approach is to try to record a couple of minutes of “silence” in the house, with the mic pre turned up fairly high, then listen in good headphones to the playback. Try cutting variuos parts of the spectrum (lo’s, mids, hi’s) to see what is revealed.

If it really is freeway noise, you should be able to hear it at some point. You can analyse it to see what frequencies it occupies.

A shock mount will help a bit with super-low frequencies that are transmitted through the structure, but these are so low that they’s be easy to EQ out. You can also try using bass-cut or high-pass settings of your mic preamp, if it has those.

Hansje makes a VERY important point about fft-size. For lower frequencies, it is CRITICAL to use a bigger fft-size. The FFT size must be significantly bigger than the wavelength of the frequencies to be removed.

The best way to visualize this is to open up N’s frequency spectrum display while playing the noise (or the music with noise). Right-click to adjust FFT window size until the low frequencies where the offending noise are displayed clearly and accurately in the display. You’ll note that at the default size, low frequencies are rather smoothed off in the display; and as you increase the window size it gets defined much more clearly. When the display looks right, use the same value for the filter fft-size.

Archimedes makes the excellent point that it’s far better to keep the noise out than to remove it later.

And Marce is absolutely right: if there’s any program in the spot you sample for noise, you’ll lose important stuff when you remove it!