Syntrillium Backup Forums :: View topic

ajfaull · Posted - Sat Dec 22, 2001 3:29 pm

For the first time in a long time I am tempted to Normalize recorded audio. Normaly when I record I increase or decrease my input levels so that the peak tends to be about -3 to -5 db. On a couple of other boards people have been telling me that I should normalize to about 96-98%. Is normalizing better or worse when compaired to manually changing the imput level?

If normalizing is the same or better than adjusting the input level, should Normalizing be done on a song by song basis, or accross the entire albumn, thus maintaining the song to song difference that was intentionally used on the record?

beetle · Location: USA

As long as you do it in higher than 24 or 32-bit, there's nothing to worry about.

sk · Location: USA

Hey, aj-

I think part of the confusion surrounding 'normalizing' (aside from the fact the in the US we spell it 'normalizing' and abroad they spell it 'normalising') is that there is, actually, either no such thing as normalizing, or no one way to define normalizing. You can normalize according to PEAK value, which, without additional compression, will place a ceiling on that song based upon the first peak value to reach 0 dB, or whatever you choose to set for the peak value limit.

You can also normalize using various RMS methods. For example, some or most programs build in dynamic compression as part of the RMS normalization routine, to overcome clipping, which allows the final 'normalized' song to sound much louder than the same song normalized using the peak method. But those same programs also offer different algorithms to choose from, so that even within RMS and the same program, you can get widely divergent results, depending upon which algorithm you choose.

This topic comes up here rather frequently, and everyone has their own opinions on this subject. Personally, I have purchased a $15 shareware program called Volume Balancer. I first heard about it on this site, tried it, liked it, and bought it. The author of that program is much more suited to explain the particulars, which he does quite well in the documentation accompanying his program. What I like about the program is that it allows you to choose whether or not you want to balance the 'loudness' across the board for all the songs that are to go on a cd, for example, based upon the quietest song, or the loudest song, of the group. The author recommends basing everything on the loudest song and bringing everything else down in line with that. But he also provides you with the option of going the other way. His approach is, in my opinion, honest, accurate, and fair. He says straight out that there's no way you're EVER going to balance quiet classical music and rock music on the same CD, for example. He doesn't offer any 'magic digital hocus pocus'. He provides honest, real world suggestions, and a product that works well and does what it says it does. His program has helped me save a lot of time, get better results, and he responds within 24 hours to any questions and/or suggestions. He offers a demo that will do 1/2 of the song completely, with no bells/bongs/whistles, etc. to interrupt you 5 seconds into listening to a transform. The URL is:
http://homepages.nildram.co.uk/~abcomp/volbal.htm.

The author is Clive Backham, and he's been a long time supporter of CE. He wrote his program because he couldn't find anything to do what he wanted, but he strongly supports CE. His program is a standalone; it's not a plug in. And it's not in competition with CE; it's in cooperation with CE.

Hope this was helpful.

sk

Graeme · Member Location: Spain

What I don't like about Volume Balancer is that it tampers with the dynamic range. This is not such a good idea.

Someone did once publish a means of semi-automatically achieving RMS normalisation using CEP. Yet again, I will quote it here (apologies to the original author - but I don't know who you are);

Instructions follows:

This manuscript lists the steps you have to go in Cool Edit Pro if you want to build a compilation of wavefiles.
The volumes of the wavefiles will be tuned in an average manner to give the listener a harmonic and easy listening feeling overall the compiled wavefiles.

Before you begin, make sure that you have your wavefiles ready mixed in final version
(16-bit or 32-bit, just a format, which can be invoked into Multitrack Mixer View).
These files will not be altered by the procedure.

1.
Begin in Multitrack Mixer View.
If you are in Single Waveform View then switch to Multitrack Mixer View (F12).

2.
Open menu "Edit/Snapping".
Select "Snap to Waves" only.

3.
Open menu "File/Open Waveform" (Ctrl-O).
Dialog "Open a Waveform" comes up.
Select all the wavefiles which you want to compile.
Click or push the button "Open" to open them all.

4.
Open menu "Insert/Waveforms List..." (F9).
Place the dialog window anywhere at the right edge of the screen.
Select the first filename with a left click.
Hold down the left mouse button and drag the filename out of the dialog window.
Place this first wave snippet to the very beginning of track 1.

5.
Select the second filename with a left click.
Drag the second filename resp. wave snippet to track 2 near the end of the first wave snippet.
Left adjust the start position of the second wave snippet with right click dragging to the end position of the first wave snippet.
There comes up a helping snapline to do the exact aligning automagically for you.

6.
Repeat step 5 until you have dragged all of your titles into isolated tracks.
If the view is getting to small then click the button "Zoom Out Full".
At end of your drag work the Multitrack Mixer View will show you a nice staircase down from the upper left to the lower right edge of the multitrack mixer view.
Save this session file. As it is the first saving, give it a wonderful name.

7.
Click double on the first wave snippet in track 1 or right click on it and select "Edit Waveform" from the contextmenu.
Single Waveform View will come up, displaying the first wavefile in entire view.

Open menu "Analyze/Statistics...".
Waveform Statistics starts immediately.
Make sure that there is value=0 in editfield "Window Width".
If there is yet another value, then cancel the gathering process.
Go to editfield "Window Width" and fill in value 0.
Restart the gathering process with the button "Recalculate RMS".

If gathering process has stopped, take a look at the line "Average RMS Power".
There will be displayed the values for left and right channel in units minus dB.

Right click with your mouse into the field onto the digits with the _lower_ db value
(that is the greater absolute value) (value=max(abs(left),abs(right))
From the contextmenu select "Copy".

Close the dialog "Waveform Statistics".
Switch to Multitrack Mixer View.

Right click into the little Volume editfield at the left side of the track 1.
The dialog "Volumefader" opens, editfield "dB" will be highlighted.

Right click into the editfield "dB".
From the contextmenu select "Paste".
Delete the minus sign (that's important).

Press "OK" to close the volume fader.

8.
Repeat step 7 until you have gathered all the minimum Average RMS Power values and copied them into the allocated tracks's volume control editfields.
All values must be positiv (greater or equal zero).

9.
Find out, which value is the greatest.
(value=max(trackvol1, trackvol2, ..., trackvoln))

Right click into the editfield with the greatest value.
(that should be a volume fader editfield)
From contextmenu select "Copy".

Right click into the "Master" editfield.
The dialog "Main Volume Fader" opens, editfield "dB" will be highlighted.

Right click into the editfield "dB".
From the contextmenu select "Paste".
Put a minus sign in front of the value (that's important).
Save this session file.

10.
Now you have to set some cue points.
The first cue point must set to the hard left
(click on the button "Go To Beginning or Previous Cue"),
press F8 or use the right click contextmenu.

The next cue point must set at the end of the first snippet
(click near the end, exact snapping will be done automagically), press F8.
Repeat this setting of cue points until you have set all needed cue points (e.g. 9 cuepoints for 8 snippets)
Save this session file.

11.
Make sure that in "Options/Settings" (F4),
on tab "Multitrack",
in editfield "Mixdowns" the entry "16-bit" is selected (CD-quality).

Make sure that in Single Waform View,
menu "Edit/Convert Sample Type" (F11),
the appropriate values are set
(stereo, 16-bit, maybe dithering enabled)
(that's important because Mixdown will "auto-dither" to 16-bit,
using the selections from the dialog "Convert Sample Type")

12.
Back in Multitrack Mixer View.
Open menu "Edit/Mixdown/All Waves".
(now you have some time to wait, depending on your hardware power,
don't smoke, better drink a cup of tea ...)

13.
When Mixdown is done, the compiled wave is displayed in Single Waveform View.
You can easily see the averaged tuned shapes of the wave snippets.

Open menu "View/Cue List".
Highlight all the cue point entries and "Merge" them.
If button "Batch" is greyed then highlight the displayed cue ranges again.
Use button "Batch".
Save the wavefiles as Windows PCM *.wav file using a numbered template into any folder.

The compiled wavefiles are now ready to burn.

14. Close window "Cue List"
Discard the MixDown file.
Save the session file for later use.
Close the session.

15. End of RMS Averagesizing.
Burn!

All I will say about this technique is that it does work, although I find it quicker to do this sort of thing by ear.

ajfaull also asked a couple of questions which seem to have got a little buiried in the resultant posts.

HairWasHere · Location: USA

beetle · Location: USA

It was our own Younglove who came up with that method, Graeme. It's just that it is too long a process to really be bothered with. It basically just gives you a manual way of doing what the programs do.

Graeme, as you know, the downloadable program will not screw with the dynamics IF you RMS normalize low enough, say, about -16 or -17db. If you do experience compression you can set the release time to help compensate for it in some types of music.

The idea with Younglove's method is that you do this in 32-bit float so when you're done you can unity peak normalize down to around 98%. No compression, no limiting!

Of course, with these programs, you can also just RMS normalize (normalise for you Brits out there!) to -18 for good headroom ang unity peak normalize to 98%, or so.

Edited by - beetle on 12/23/2001 07:32:43 AM

beetle · Location: USA

Oh, BTW, I differ slightly with Graeme in that I preserve the levels on the LP ONLY if the disk was well mastered AND it is fairly short in duration. Sometimes the first cut or two is much louder than the ones on the inner groove. If I encounter this, AND the music is consistient, I will RMS normalize the LP. If I have a ballad with a wide dynamic range, I may still RMS normalize it along with the rest.

But, since tracks can vary with EQ, compression, ect, it pays to make a run through the sequence by EAR just to ensure that it all works to your liking.

Sometimes, you can play god here by boosting levels a bit, EQing, and/or using limiting and/or compression. Some albums (if you are doing restoration) are benifitted by a touch of compression. People have varying opinions about this. I tend to do what feels right for the projeact in question. I used to be a hardline purist about this but i've mellowed a bit. Some people NEVER do it. and some always do it. That, you will have to decide for yourself after listening/reading the opinions and evaluating your work and considering the music.

This is all called mastering. Fun, huh?

If you EQ or compress bass, do it BEFORE any RMS normalization!

Edited by - beetle on 12/23/2001 07:38:35 AM

sk · Location: USA

Graeme · Member Location: Spain

sk · Location: USA

Ok, Graeme.
I think I understand what you're saying and where you're coming from.
However, I think that either you or I might just have a fundamental misconception with regard to how VB works. Particularly, when I said it gives you the option of taking the quietest song and work from there, the basis for not only determining the quietest song, but also for balancing all of the other songs, is based on actual 'loudness', which the author does not specifically define, but in running some RMS 'tests' on 'before and after' songs, it seems as if he's using 'loudness' according to some scale that is neither RMS nor peak. And unless I'm completely not understanding the terminology, lowering a louder song to match the level of a quieter song is NOT the same as compressing that song, or in any other way fundamentally altering its inherent dynamic qualities/structure. That's the reason the author suggests THAT method, because it does NOT necessitate compression. It DOES, by definition, lower the overall volume or loudness of all of the 'louder' songs, but it does not 'take away' anything. It does not 'compress' anything to avoid clipping. It naturally avoids clipping by bringing the songs to a naturally occuring common ground. If there's a 'problem' with this, it's more along the lines that most of the songs that had their volume reduced COULD be played louder - because they have adequate headroom to be played louder without clipping. In essence, then, they end up relegated, so to speak, to a quieter place in that particular compilation. But by the same token, you can still turn up the volume on those songs without running into any serious clipping, because they have retained their full range of headroom and inherent dynamic properties because nothing had to be compressed to make them fit. Maybe I don't understand what you're saying, but just reducing the volume to match up with a quieter song should not result in any lossy compression. If I've misunderstood your point, I apologize, but I wanted to share this because it sounded as if we might be talking about two completely different things.

sk

younglove · Posted - Sun Dec 23, 2001 7:47 pm

Beetle and Graeme, the procedure reposted above was by Detlev Dalitz, not me. I had
a different procedure for RMS normalization. In Detlev's, you use the multitrack, and set up
track faders. It is long and fiddly and it's easy to make a mistake with the plusses and minuses, but once set up, you just push a button and it normalizes the whole album automatically.

My procedure is less automatic (you press a button for each track instead of for the whole album), but it's easier and not so prone to mistakes. Also, the results of the two procedures
are not exactly the same!

The Volume Balancer program spoken of balances by RMS normalization. It differs from mine in how you get the RMS value to normalize to for a track.
It can refrain from hard limiting, but I don't recall if that is chosen with a hard prohibition or
if that is just the case if your chosen level is low enough.

The CD Spin Doctor track balancer does not do straight RMS normalization, but uses some
other or additional algorithm. I don't find its results to be better than RMS normalization
and can produce track levels that are wacky.

And then there's the implementation of 2BDecided's work, called Wavgain. It takes the
Fletcher-Munsen curve into account, but makes some rather arbitrary level choices. The
resulting respective levels are balanced well enough (it's hard to decide if they are better
than RMS normalization), but the overall level recommendation is funkily low.

Graeme · Member Location: Spain

younglove · Posted - Mon Dec 24, 2001 3:31 pm

OK, here's my procedure for RMS Nomalization:

-------------------------------------------------------

The Normalize function normalizes to peak. To get the output
levels sounding all the same (close, anyway) you should
normalize to RMS, instead. Here's the manual procedure in
CEP and CE2000:

1. [steps 1-4 were relevant for earlier versions of Cool Edit]

2.

3.

4.

5. Starting with a New file, Open Append (not Open) all the tracks (now the
whole compilation is opened as a single wave file and each
track will have its own cue range). Warning: if you select
multiple files in a directory to Open Append at once, Windows
may not Open Append these in the same order as your selected
list.

6. Edit\Convert Sample Type the compilation wave to 32 bit,
if not already (must proceed in 32 bit).

7. For each cue range (track), do:

....a. select the cue range so that it is highlighted

....b. Analyze Statistics. Note the Average RMS Power of the louder of
the left and right channels (for example, if the two channels show
-14.7 dB and -17.8 dB, the -14.7 dB one is louder).

....c. Transform\Amplitude\Amplify by *minus* that many dB
(the value to amplify by will normally be a positive value,
for example
"minus -24 dB" = 24 dB; make sure that "View all settings
in dB" and "Lock Left/Right" are checked). This will
temporarily grossly amplify your track--don't worry about
it: it's 32-bit float!

8. Having done step 7 for each track, the tracks are now
RMS-normalized to avg RMS of 0 dB, rather than
to peak. Now Edit\Select Entire Wave.

9. Transform\Amplitude\Normalize to whatever value you
want, say 96%. This brings the levels back down without
clipping.

You can now use the Batch function on
the Cue List (highlight all the cues first) to automatically
save the tracks to separate files, if you want. If you do,
then you will probably want to audition them to see which is
which and rename them once saved.
The procedure reads complicated, but if you try it it's
really easy, just a bit tedious because it's a manual
procedure.

Edited by - younglove on 12/24/2001 3:36:56 PM