Repetative Tasks in Cool Edit

Through experience and digesting some of the information available here on the forum I realise that with noise reduction it is far preferable to make many small dB reductions rather than one or two big steps. For example it would probably sound better to reduce noise by 2dB ten times rather than 20dB in one step.

Is there a way in Cool Edit of programming the software so that you can enter noise reduction settings and then basically tell it to do it "10 times" "20 times" etc?

If Cool Edit doesn't do this can users recommend other software which does and how to use it? I have thought about getting Soundbooth, Wavelab, Cubase etc but I don't know if they have this useful function.

Thanks very much
David

Not going to comment on the basic concept that 10 2 dB reductions being superior to 1 20 dB. 

In CEP the basic task outlined can be done via a script (In CE I think scripting was located under the 'Options' menu)

In my experience scripts seldom address a complex succession of tasks in as straightforward manner as one might expect/hope.  This simply means that setting up a script to do what you want can be a non trivial task with unexpected collateral artifacts.  Merely saying that while scripts can be beneficial for all sorts of repetitive tasks, trying to set one up to execute without adult supervision (hit the button walk away, come back and cake is baked) can require some time (to troubleshoot)

A script, generally speaking, is created by assigning a title then manually executing a series of tasks.  Upon closing and saving the script those task can be repeated by running the script.  Simple.  A NR script in CE looks something like this:

Title: nr sldjy 01
Description:
Mode: 2

Selected: none at 170568 scaled 9153792 SR 44100
cmd: Channel Both

Selected: 0 to 9153791 scaled 9153792 SR 44100
cmd: Noise Reduction\Noise Reduction
1: 0.44
2: 0
3: C:\Soundwav\tools\SldJy2009 NP01 24k.fft
4: 24000
5: 1000
6: 8
7: 44100
8: 2
9: 32
10: 8
11: 0
12: 0
13: 0
14: 0
15: 0
16: 0
17: 1
18: 1
19: 5
20: 0
21: 1339
22: 873
23: 1566
24: 12233
25: 1566
26: 13325
27: 1584
28: 16384
29: 1547
30: 0
31: 1
32: 1
33: 0
34: 0
35: 0

End:

______________________

one can of course simply repeat the meta task (NR) 10 times to achieve the 10X2 dB reduction.  A script can be assigned a 'hot key' and a 'script' can be embedded in a script so it is relatively simple to press the hot keys nine more times and complete the script (the above script uses the default 40 dB reduction)

What was nice about scripts in CE was that they were editable with any common text editor.  The commands were not buried in the registry so for what the scripts could do they were pretty straightforward to modify.  For example rather then pressing the hotkey nine more times (or  running NR from the menu a total of 10 times) while creating the script you could highlight and copy the appropriate section(s) and paste them nine times to complete the script.  Or if one wanted to use the same basic parameters but change the fft profile that is very easy to do.  You can create the entire script in the text editor.  The down side is that there is no documentation on how to edit a CE script so it's all trial and error and it can be frustrating.

Some things are pretty obvious.  'cmd' reference CE internal commands and do not require a number preceding them.  If a command has variable parameters those will be presented with a numerical listing with line numbers corresponding to the parameter.  Some of those are also fairly obvious.  in NR line #1 is the amount of  NR applied, where 0 is none, .44 is 44% and 1.00 is 100%, etc.  Line 3 references the fft noise profile, 4 & 5 list snapshots and fft window size.  Line # 6 is the precision factor, 16 & 17 are transition width and smoothing amount.  (& this is all more or less from memory so the caveat is: trial & error)  The execution of the script completes with the 'End:' command. So you can terminate script execution at any point simply by inserting the 'End:' command. The frustrating thing with CE scripts in general (besides the length of time it can take to polish one) is that not all that infrequently the thing one most wants to automate simple lacks an assignable parameter.  In the OP's query I'm not sure that the dBs of NR is controlled by a single line item parameter.  Again: for example.  It might have been calculated sample by sample as a function of noise profile, percentage of reduction, window size, number of profiles, etc.

That said, the basic task of repeating the steps applied to one file on another file is relatively easy and more or less straightforward.  You just do them once and save the steps.

good luck

It is certainly better to do noise reduction in more than one pass. However many passes of identical settings with , say, 2dB of noise reduction won't necessarily produce better results. At each pass the amount of noise in the file is less therefore the settings will tend to have less effect each time. It also depends on the nature and frequency spread of the noise. The best way is with a couple of passes at perhaps 6dB with different FFT window sizes to effect the noise reduction differently.

The Noise Reduction in subsequent versions of CEP/Audition have improved somewhat. So if you are looking at spending money for another program to do noise reduction better than your version of Cool Edit perhaps you should think about upgrading to Audition 3 (or 4 when it comes out next year).

You may want to experiment, doing several passes manually, then compare the result with a single, greater reduction pass, to see if there is really any benefit before going to much trouble trying to automate the process.

For the work I do most often, cleaning up recordings made from phonograph records, I have not observed any benefit to doing several smaller reductions using the same place in the recording as the noise sample for each. If there is indeed a difference between approaches, it has been too subtle to interest me.

The suggestion to use different FFT sizes for different passes may be valid but, again, while the result of doing so are different, I haven’t found anything I considered worthwhile. It might be quite different for some different kind of recording, however.

For what I do, there is some basic surface noise that tends to be prevalent across the entire disk, and thus a NR profile made from a sample of it can be applied to the entire recording. Other noise is rather localized. For best results, the noise sample must be taken from within the effected area and the resulting profile applied only to the appropriate area. Automating that doesn’t seem likely.

You should bear in mind that Andy is using CE2000, and therefore his comments about noise reduction only relate to that product, and are still somewhat suspect. Subsequent versions are considerably improved. The product moves on; unfortunately Andy doesn't.

So if RX2 uses varying FFT sizes over the frequency spectrum, why not do this in AA3?  The trick is, which ones would sound best.  Then the question becomes: is 20dB of NR done one-pass over a limited frequency range better than 6dB of NR done 3x with varying FFTs?

Here's what I mean: take the offending audio, run it through the Frequency Band Splitter in MV.  Cut it up into, say, 5 files, a couple of octaves each. Run NR on each file, tailoring the FFT size to each (but what would that be???).  When NR is all done, mix the 5 files down into one again.

That sounds really time consuming.  I've actually done this for radical pitch shifting of music, but never for NR.  Has to be REALLY worth it!  Definitely a poor man's RX2 approach, at best.

(but what would that be???)

When the wavelength of the signal you are trying to process won't fit in the window length, the processing rapidly becomes less effective - but the artefact level increases. In general, you need to use less processing percentages when you use longer (lower number) FFT windows because of this, but since most of the noise that the majority of people are concerned with is higher in frequency, then using smaller windows means less artefacts therefore bigger processing percentage before the artefacts become offensive.

So to work out what you need, you'd need a profile of the noise you're trying to remove (ie, from the sample). In general it means that the biggest NR pass is the highest FFT size one - and this one you'd always do. What you'd have to add to this is the only thing that there's any real question over, I'd say.