Forums | Search | Archives

All Forums  Cool Edit  Spectral Decay: what is it exactly?

Author	Topic
MrGerbik Posts: 34	Posted - Wed Jun 11, 2003 11:58 pm  Hi, I am about to run Noise Reduction in CEP 2.1 (to get rid of some vinyl noise) and I noticed the spectral decay option... "help" didn't explain much, and some searches in the forums left me a bit confused. What exactly does the spectral decay percentage affect, and how is this such a useful tool for staying away from artifacts (which I gather from other threads)? What are some good settings to try in my case? In CEP 2.0, I simply got a print from a section of pure noise (between tracks) and applied NR to the whole wav at about a level of 60. Thanks :)
SteveG Location: United Kingdom Posts: 6695	Posted - Thu Jun 12, 2003 2:15 am  In the manual, it says the following: Spectral Decay Rate When audio above the noise floor is encountered, more audio in the same frequency band is assumed to follow. With low Spectral Decay Rates (measured in percent), less audio is assumed to follow, and the carving function will cut more closely in time to the frequencies being kept. • With Spectral Decay Rates too low, background bubbly effects might be heard, and music may start to sound artificial. • With Spectral Decay Rates too high (above 90%), unnaturally long tails and reverbs might be added. Values of 40% to 75% work best.[/list:9abb0fdc4c] So I suppose that you could attempt to simplify this by saying that it's a bit like the speed at which the 'knife' doing the noise cutting is being withdrawn. That's not a very good analogy - but the real explanation isn't that simple, as far as I can tell. So you might not like the following very much either! It helps if you consider that the noise profile has to be subtracted from the signal being processed, and when there is no wanted signal, then any difference in the instantaneous noise and your noise sample (which there inevitably will be, because they are not correlated) will result in the 'bubbly' sound. The problem is that you have to predict what is going to happen next, and since the way that the spectrum is treated appears to be adaptive, you have some choice in how you present the result at any given instant. And since the computed result is based on the audio present, as well as the noise, you can see how having a high rate set could lead to what soundslike additional reverberation being added - it's the information from the decaying part of the audio signal being used as a partial predictor. So rather than leave this as a fixed level, which it used to be in the NR, Synt decided to let the predictive information effect die away when nothing above it was detected, because it makes for less noise under somecircumstances. And since everything else about this is computed, the only parameter you can vary is the rate at which it decays. But the results from this vary, and you haveto experiment with it. There are some circumstances where its use turns out to be rather undesirable, and others where using slightly less NR, but a carefully controlled decay rate, is a considerable improvement. The interaction between the overall amount of NR and the spectral decay rate is the crucial bit - and unfortunately there is no one setting that is going to work for everything. In the past, you didn't have this choice, but now you do. And more choice isn't alwaysa good thing... So my best shot for setting this up is like this - YMMV: Leave the spectral decay control set at the default 65%, and set up the NR slider so that you have what sounds like an acceptable amount of NR. Only now should you consider altering the decay rate, and you may find at this point that you can possibly get a better combined result by juggling these two settings against each other. And just to add a bit of fun to the proceedings, the Transition Width Factor will also alter the results in some circumstances - this rather depends on the noise itself. And of course the Smoothing Amount can make a huge difference - increasing the amount of this whilst altering the spectral decay rate could easilygive you a more acceptable result. As with alleffects, the default settings are only a starting point - you have to experiment each time you use it. I don't know if that's any help - it's the best I can do with toothache. _________________
Mark T Location: Norway Posts: 890	Posted - Thu Jun 12, 2003 5:08 am  Sorry to hear about the toothache Steve, it's a killer. But I was just reading your post and thinking "Oh so that's what it is about" Good one! _________________ Mark nil desperandum - nunc est bibendum
ozpeter Location: Australia Posts: 3200	Posted - Thu Jun 12, 2003 6:29 am  I wouldn't dream of trying to explain it following SteveG's post (and anyway he's done it!), but a practical example might be useful - which I came across the first time I tried this new parameter. I was dealing with a string quartet recording made in a large reverberant hall, which unfortunately had noisy heating - a constant rumbly shush (that's a technical term). Using the NR to remove this sound, I found that if I had the spectral decay rate not quite right, as the reverberation at the end of movement died away, I could hear the heating noise behind it just audible, with the noise dying away just after the reverb, as the processing adapted slightly late to the change. Increasing (in this case) the SDR just a couple of percent caused the noise to be removed just ahead of the reverb, so that the reverb covered it. It's a sensitive adjustment, but one that can provide remarkable results if you get it right on the particular material that you have. - Ozpeter
MrGerbik Posts: 34	Posted - Thu Jun 12, 2003 2:32 pm  Thanks guys Obviously there is not clear cut way of going about this; I need to play with the settings. As I am still pretty new to CEP, alot of this is quite intimidating... NR makes me nervous Oh well, at least I figured out how to use the graph so that no low end is taken out with the NR... that was quite helpful. For this particular vinyl recording I did, I ended up using a NR level of 60 with spectral decay set to 55%; got pretty nice results I think.
	Topic

Page: