|
First you should know what normalizing means. The normal use of normalizing is peak normalizing.
The file is scanned for it's peaks. Then, gain is applied to the total signal, so that the peaks are brought at the level that you asked.
For example, you made a recording, with the loudest peak at -6 dB. Upon scanning the signal (this can be a selection also, but usually you'll do it on the whole file), the program will know that the loudest peak is -6 dB. Now you ask to normalize to -0.2 dB (make it a habit never to normalize at 0 dB flat. Some cd pressing plants reject this, they count 0 dB levels as errors). This means that your loudest peak will be brought up to -0.2 dB. If you can count, you'll see that's 5.8 dB gain needed. So, 5.8 dB of gain is applied on the whole signal. Only gain, no dynamics are changed (so, you can normalize manually too, scan for peak, then calculate the gain needed to bring up to the adequate level and apply).
RMS is a little different. And RMS normalizing is not so common, and frankly not much used. Take a signal with lot's of peaks, but very dynamic, uncompressed. A signal having lot's of peaks doesn't necessarily mean that's it's loud in average. For example, take a recording you did from vinyl with a huge pop in it. Let's say overall, your signal is quite steady around -10 dB. But the huge pop peaks at, let's make it easy, 0dB. In average, the signal is not "0 dB" loud, but more in the ballpark of -10 dB. Would you scan for peaks, then you would find the 0 dB one, which for your purpose isn't useful.
RMS means Root Mean Square. It's a mathematical term, just know it means it's an average. So it defines more or less average level, loudness if you prefer. The more dynamic your sound is, the more difference there will be between peak levels and RMS levels. The more compressed your signal, the smaller the difference.
So you'd say, if I want to normalize on a loudness basis, RMS normalizing would be perfect. One big problem.
Take the vinyl example I gave earlier. You scan the RMS level, and find roughly - 10 dB. You want to raise the loudness to -7 dB. So, a 3 dB gain will be applied to the whole signal. BUT... We had our pop at 0 dB... That will be amplified too. Resulting in clipping.
That's why in Soundforge, when you select RMS normalizing, new options appear.
Apply dynamic compression : it will compress the signal that would go over 0 dB to a certain point, compression that is quite limited. You have basic control over it, but not much.
Normalize peak value to 0 dB : it will revert to peak normalize to 0 dB, in our example, nothing will amplified (our peak is already at 0)
Ignore : it will do the gain, and clipping will result out of it
Stop processing : it won't do the normalize, because it encountered a clip. Clipping is bad.
In Soundforge, if you normalize without doing anything special, it will scan the file, then apply normalisation when necessary.
You can scan manually in the dialog window too. Either, you want to see what your peak and RMS values are, so you can make a decision, or it has another (limited use) and that is, make a selection, scan it's levels, then make another selection and select "use current scan level". It will normalize your new selection, but based on the scan of the first one. Someone still has to explain me a good use for that...
So for "normal" normalizing, using the scan feature is not necessary, is it will be done anyway.
Generally now :
If you want to bring up the level, without changing dynamics, in an automated way, use peak normalizing.
If your level is close to the peak level you want to obtain, don't bother, you'll lose more quality than doing nothing (example, your signal peaks at - 1 dB, and you want to normalize to -0.2 dB. You frankly won't hear much change, and because in the digital domain, even a simple gain is a calculation, you'll lose some info.
If you want to change the dynamics of your signal (you want it louder), use a compressor, or limiter if you're more heavy handed. Forget about the RMS normalize. You'll have much more control with dedicated tools.
|
