From: Europe
Registered: January 15, 2001 Posts: 1842
posted
OK folks I'd like to (re)start the definitive thread on Normalising.
I did a search and this has come up a couple of times before in this thread and this thread, and whilst people made some good points in those I wanted to ask a couple more questions, and sort of 'update' the discussion.
My partner and I have a 'gentleman's disagreement'(!) that Normalising changes the sound in some way. I say, no way, it just makes it louder and that deceives us, he says the quality of the sound changes. I find it much easier to normalise files, because it brings them all into the same ballpark volume-wise and makes it easier to judge relative levels and settings over time. I do it automatically after recording takes, after of course getting as good an original signal level as possible. (BTW I'm using Logic/Mac.)
So a list of questions that might help settle this dispute once and for all are:
1. Does Normalising change the sound in any way other than simply raise the level? And is this in theory or practice?
2. Following on from (1), Is it true that different software will Normalise in different ways and that a file Normalised in Logic might sound different to one Normalised in Pro-Tools?
3. Is it true that digital plug-ins (VST and the like) work better with some digital headroom? (ie do not Normalise to 0dB). What happens in a chain of plug-ins?
4. Following on from (3), What then is a sensible level to Normalise to?
From: Southsea
Registered: June 13, 2001 Posts: 3200
posted
If the software is working properly then normalising shouldn't change the sound. However, changing the sound level, even by a fraction of a dB, can change the way we percieve sound.
Normalising to less than full scale is a good idea to allow any subsequent reconstruction filters some headroom. You'll find reconstruction filters in D-A convertors and in any processing that might interpolate between samples like sample rate conversion.
If you are worried about different software sounding different then try doing exactly the same processing in each program. Then see if pasting an inverted version of one file over the other gives all zeros (or just one or two bits worth of noise due to dither). If you don't get all zeros then there may be problems.
it really changes the sound when it dynamicly compresses it. i dont like normalising the hole track prefer to use something like L2, . but if i want a sample or something to stcik out i might process that.
From: Lat. 52 19' Lng. 4 41'
Registered: October 24, 2001 Posts: 2176
posted
quote:Originally posted by ally: it really changes the sound when it dynamicly compresses it.
Normalising in its standard form doesn't involve any dynamic processing, it's just about setting (increasing) the overal volume of a track to a certain level, in such a way that the loudest signals peak at a certain maximum level; for example -0.5dBFS.
As soon as any dynamic processing is applied the sound indeed changes; as James Perrett already has pointed out, louder signals are perceived differently than lower volume signals so normalised signals without additional compression added may seem different to the original.
From: Essex Fells, NJ now in Sydney, NSW, Australia
Registered: May 07, 2002 Posts: 4557
posted
Most mastering engineers I've worked with or talked to are completely against normalizing as it leaves them with less room to maneuver.
You have to realize that when you normalize you're basically maximizing the signal gain, raising the noise floor and eventually reducing the perceived RMS Power.
From: Europe
Registered: January 15, 2001 Posts: 1842
posted
quote:Originally posted by Zoesch: Most mastering engineers I've worked with or talked to are completely against normalizing as it leaves them with less room to maneuver.
Yes - I wouldn't Normalise a stereo pre-master file going to the Mastering House. I'm thinking more of individual audio tracks.
From: CA, USA
Registered: September 29, 2003 Posts: 595
posted
Just record as "hot" as possible...no need to normilize. More signal, less noise.
Ally - I don't understand what you just wrote. "...when clipping occurs." If you make settings to normilize to 0db or less, why do you have to worry about compression on clipping?
From: Europe
Registered: January 15, 2001 Posts: 1842
posted
quote:Originally posted by James Perrett: Normalising to less than full scale is a good idea to allow any subsequent reconstruction filters some headroom.
So... what to stop at? I think Logic defaults to -0.5dB, is this OK?
And in view of the above, does it make any difference to the stereo mix file if I have 24 audio tracks all normalised to -0.5dB all running through 4 plug-ins?
From: Southsea
Registered: June 13, 2001 Posts: 3200
posted
I wouldn't normalise individual tracks in a multitrack session - there's absolutely no point as it almost certainly means that you'll have to reduce their level again at a later stage.
Ideally I'd want the highest peak in each track to be within 6dB of full scale but once you've recorded the tracks already then there is no point in changing the level outside the final mix.
From: peoplesrepublicofcork
Registered: December 06, 2000 Posts: 444
posted
1. Does Normalising change the sound in any way other than simply raise the level? And is this in theory or practice?
Answer: Can you hear a difference? You should. There IS no theory or practice, because there IS a difference, fact. It's in the numbers.
Let's look at your method. The way you normalise, the audio data is re-written at the original bit depth. So, unless you're recording at 32 bit floating point, there are rounding errors (e.g. 0.5 gets turned into 1 or 0, depending) and dither is added because the algorithm does its calculations as floating point (more than likely) and has to save the data as fixed. Even 24 bit requires dither.
So, the FACT IS there is a difference. But you may not be able to hear it, because your monitors might be crap, and you don't know what to look for. What rounding and dither do is "veil" the sound in a very subtle way. There are guys who spend hours trying different flavours of dither on EACH SONG just to minimise this problem.
2. Following on from (1), Is it true that different software will Normalise in different ways and that a file Normalised in Logic might sound different to one Normalised in Pro-Tools?
Answer: Yes. It's subtle, but again, if you know what to look for, you'll hear it.
3. Is it true that digital plug-ins (VST and the like) work better with some digital headroom? (ie do not Normalise to 0dB). What happens in a chain of plug-ins?
It depends on who made the plugin. In VST, it's hard NOT to do everything at 32 bit float. So unless you're using very old plugins, headroom is not an issue for sane people.
4. Following on from (3), What then is a sensible level to Normalise to?
This doesn't follow on from 3, because in the context of your question, headroom is not an issue in a current VST system, because all calculations are 32 bit floating point. That's very, very, very, very close to "infinite" headroom.
SO: You have a 32 bit floating point mixer (in Logic). When you turn something up, it doesn't add any dither, so there's no veiling of the sound. You should not normalise any of your recordings automatically. You're just adding an extra layer of mush. As James said, get within 6dB when recording at 20 or 24 bit resolution, and you're getting the best use of all those numbers.
Finally, it seems to me that you're getting distracted by making your meters hit zero. You need to LISTEN, not look.
From: Europe
Registered: January 15, 2001 Posts: 1842
posted
OK, thanks everyone - I'm learning a lot with this. Can we just clear up one thing and that is that I do know perfectly well about maximising my recording levels at source! In an ideal world all our recordings peak just under 0dB, but there are times, when for one reason or another one ends up with a quieter take that needs some boosting, either on its own or to match up with another take.
So moving on, the above explanations throw up the following question...
If I should NOT be Normalising my individual audio files, what should I be using to INCREASE their level??
The choices would seem to be: 1. Raise the level of the fader in Logic 2. Apply a volume plug-in 3. Use a limiter
It seems to me that in all three cases I'm still adding gain to the original audio data using mathematical calculation, not to mention changes in sound quality as soon as a plug-in is inserted in the signal path. So why is this different sonically from just adding gain directly onto the audio data by Normalising?
From: Greece
Registered: December 13, 2000 Posts: 2257
posted
If you are recording everything as hot as you say James, then when you are mixing and summing a whole load of tracks (you mentioned 24) I would have thought you’d have so much level that you’d be able to balance any slightly lower volume takes in the mix; i.e. you’d have the opposite problem of too much overall level and would therefore be pulling lots of faders down in an attempt to stop overloading the main bus. Or am I missing something…
From: Lat. 52 19' Lng. 4 41'
Registered: October 24, 2001 Posts: 2176
posted
quote:Originally posted by Tomás Mulcahy: So, unless you're recording at 32 bit floating point, there are rounding errors (e.g. 0.5 gets turned into 1 or 0, depending) and dither is added because the algorithm does its calculations as floating point (more than likely) and has to save the data as fixed. Even 24 bit requires dither.
An already digital waveform (which we're talking about here) which will be normalised wonn't contain 'halves' (the 0.5 you're referring to), either before the normalising process or thereafter, since in a digital systems only the figures 0 and 1 exist.
I assume you're referring to quantising errors.
quote:Originally posted by Tomás Mulcahy: So, the FACT IS there is a difference.
I suspect that even with high-end monitoring systems, a signal needs to undergo several normalising processes in order to make the quantising errors noticeable. Correct me if I'm wrong, but I never actually noticed a single normalising process affecting the waveform in such a way that it became audible - in real world practise anyway.
Perhaps the likes of Hugh, The Byre or James Perrett are able to explain matters further. If I have some spare time left in the studio tomorrow I will try a few normalising passes to verify whether there's an actual change to the signal or not.
From: Lat. 52 19' Lng. 4 41'
Registered: October 24, 2001 Posts: 2176
posted
You're right Tim. I recall the Pro Tools mix bus being 128-bit, so running several full-scale level tracks into the mix bus would certainly cause it to overload.
In practise, far fewer 'hot' channels would cause it to clip, which is why I prefer to keep the levels of the individual channels a bit quieter.
From: Essex Fells, NJ now in Sydney, NSW, Australia
Registered: May 07, 2002 Posts: 4557
posted
IIRC It's 128-bit fixed point (Or used to be, I'm 100% sure that the HD bus is 128 bit float)
Frank about reducing RMS power... simplified verion: if you maximize the peaks you're reducing the RMS power as the average values get closer together. If you normalize to maximize RMS power then your peaks will be averaged.
Don't want to get into higher order maths (Since the forum doesn't like formulas), but if you want a truly detailed explanation let me know... as an experiment take a sample clip into your favorite wave editor and measure its RMS power, then maximize the peaks and measure the RMS power again... undo it and maximize the RMS power, then measure the RMS power and the peak amplitude.
From: St. Louis, Missouri USA
Registered: November 11, 2002 Posts: 2073
posted
quote: ....as an experiment take a sample clip into your favorite wave editor and measure its RMS power, then maximize the peaks and measure the RMS power again... undo it and maximize the RMS power, then measure the RMS power and the peak amplitude.
One of the things that I try to do when doing some quick and dirty home mastering is to make all the tracks sound about the same volume on the volume on the CD. How is this easily done without normalizing? I have been doing it with the L1 limiter, adjusting each track until they sound pretty close to the same...
From: Southsea
Registered: June 13, 2001 Posts: 3200
posted
James - I don't understand why you want to increase the volume of everything - can you explain?
DH - I would do something very similar to your method. I find the quietest track and reduce the levels of the louder tracks until they match the quietest track. I will then feed the whole thing through a limiter that is as transparent as possible to get my final level. I prepare masters for both vinyl and CD. You don't want limiting on a vinyl master so my method means that I produce a vinyl master before adding limiting for the CD master. Your method probably works fine if you are only producing CD masters.
quote: In an ideal world all our recordings peak just under 0dB
I used to think this, a hangover from the analog world for me.
With digital I always used to aim for -6db on the peaks but after some reading and experementation I've found that recording with peaks of -18db leaves a lot more flexibility when it comes to mixing with plugins and just sounds better.
The only normalising I do is in when I'm compiling CD's in Jam - to adjust the relative volumes of the tracks.
From: peoplesrepublicofcork
Registered: December 06, 2000 Posts: 444
posted
Frank, what IS a digital number? In this case it's a binary representation of an "analogue" waveform. A value can be represented as decimal, or hexadecimal, or two's compliment, or whatever you like. The example of 0.5 being rounded to 1 or 0 is EFFECTIVELY what happens during quantising. It's the clearest way to explain the issue, without clouding it with high level mathematics.
The problem with normalising is that the effect is cumulative. Some plugins use dither whether you like it or not, Logic and Cubase dither quite poorly from 32bit float to 24 bit and below, etc.
Furthermore, there are a few more important issues in this case:
1. Normalising every file during a session is a waste of valuable time.
2. If you're within -6dB of full scale in a 16 bit system, then you're using all the bits.
Recording everything at 0dB is NOT ideal. For a start, your meters might not be that acurate. Do you know how many samples get clipped before the overload indicator is activated? If you're trying to achieve this all the time, you probably have little clippings all over the place. No wonder your recording sounds so cold and "digital"!
3. People are applying analogue gain staging and headroom practices to digital systems. This is totally inappropriate, and leads to stupid questions like this one.
Just trust your ears, and stop obsessing over what your peak meters say. Obsess about whether you're gonna reach anyone with your music (or not).
PS Frank- I can hear the difference. I hope you can. PPS- If you require further enlightenment try: http://www.digido.com or buy Bob's book. You might also like to try The Tao of Pooh. This book covers issues many orders of magnitude more important that digital audio.
[This message was edited by Tomás Mulcahy on October 08, 2003 at 11:04 AM.]
There will be a slight difference after normalising due to the rounding errors introduced whenever you change the gain on a digital audio signal.
Why not test the effect of the normalising with your ears, and see if you notice the difference or not.
Normalise some audio to some peak value, then normalise it down to it's previous level. You can then compare this file to an un-normalised version (copy the file before you normalise). You could even reverse the phase of one of them and mix them back together and see how much cancellation you get...
From: peoplesrepublicofcork
Registered: December 06, 2000 Posts: 444
posted
jcat- the phase thing is a good idea. That'll reveal all sorts of nasties.
BTW, the ONLY reason that DAWs have a normalise feature at all is a legacy from Digidesign's Sound Designer. This was THE digital audio app back in the day (1984?), so everyone blindly copied its feature set. Problem is, it evolved from a sample editor, where normalising your multisamples was/ is a good idea. As I have made clear, normalising is not a good idea for multitrack recording purposes.
James, now you've really muddied the waters by bringing limiters into it. Do you understand the concept of "perceived volume"? It's all about perception. trust your ears. Taoism.
If you stick with Logic, then you're OK using their gainer plugin (is that the right name?)or a fader to adjust levels. But as Bob Katz says, never turn your back on digital! Trust your ears. If you find that increasing the gain using a Waves eq plug with all bands bypassed sounds more transparent, then do that. I'm told that this is more accurate than using the fader in a PT mix TDM system. Can of worms... In Cubase, I sometimes use the TLA EQ-1 just for gain, because it sounds good and gives 30dB. There's a free VST gain plug out there somewhere that DOES NOT sound good, and seems to have an error margin of +/- 1dB. Cubase's own meters have a similar error margin.
From: Essex Fells, NJ now in Sydney, NSW, Australia
Registered: May 07, 2002 Posts: 4557
posted
Tomas, exactly... and besides perceived volume is also related to the RMS Power contained on the waveform.
Or translating it back to the analog world, how much travel does your tweeter/woofer cones make if the signal is normalized as opposed to the original?
From: London, England
Registered: September 19, 2002 Posts: 549
posted
Ok, what about this scenario:
I'm working with my bass player, he's just tracked a great take on one of our songs, and before he leaves we decide to have a jam over a plain beat. I quickly plug in my keyboard and we jam along for 15 minutes.
The track is fantastic, but my keyboard level was a little low - I'd have trouble mixing with it at that level. Is it better to normalize the audio or bump it up with a plugin?
From: Maple Grove, MN USA
Registered: June 13, 2002 Posts: 382
posted
Excellent question - here's another:
If normalization is done to -1 db to prevent any clipping, and track levels during mix are then more consistent - is this more harmful or beneficial? I've tried both - and to me, my ears say "go with the normalized (-1 db) tracks."
From: Maple Grove, MN USA
Registered: June 13, 2002 Posts: 382
posted
To clarify:
One reason to bring individual tracks up to near normal is so that limiting can be applied more easily.
I am going to try some experiements later today and see if I can hear a difference. For me, normalizing to -1 db not only sounds better, but makes much of the processing easier to manage.
From: Europe
Registered: January 15, 2001 Posts: 1842
posted
quote:Originally posted by James Perrett: James - I don't understand why you want to increase the volume of everything - can you explain?
I don't. All I want is for my audio tracks to be operating in the same sort of range with some sort of consistency. Perhaps if I was a better engineer I would be able to record everything at exactly the volume I want to mix at and only make adjustment by tiny amounts on the fader. But inevitably my audio tracks end up within a range of volumes - example: I record a screaming guitar take but find that I really want to use a tiny accidental 1/2-note or harmonic that is much quieter than the rest of the file, or maybe I want that singer's breath as an effect. The +6dB of digital gain Logic gives you on the fader might not be enough to bring this up to a workeable level, short of cranking up the master outputs and turning down all the other tracks. My solution would be to bounce this region as a new file, normalise it and stick on a new track. I gave another example in an earlier post of piecing together a voice-over from different takes where the base volumes might be subtely different, using normalising to bring all the takes up to the same level before piling on the EQ/compressors etc so as not to have to draw in complex controller curves and/or run them all on discrete tracks.
I'm frustrated because a) I appear to be operating in some sort of parallel universe because no-one seems to need to do anything like what I've outlined above and b) I seem unable to express myself clearly! Bear with me, I'm trying my best, but I've really run into a subject here that I don't fully understand, which is why I posted the topic!
Call me slow, but I have not yet grasped why adding +6dB with a gain plug-in would sound any different to adding +6dB to an audio file by normalising it?
From: Happy Argentina
Registered: October 01, 2003 Posts: 194
posted
[QUOTE]Originally posted by James Lehmann:
I'm frustrated because a) I appear to be operating in some sort of parallel universe because no-one seems to need to do anything like what I've outlined above and b) I seem unable to express myself clearly! Bear with me, I'm trying my best, but I've really run into a subject here that I don't fully understand, which is why I posted the topic!
Fully agree Now: How about samples and mixing? how d'ya possibly make'em get along if not normalizing in some way?
From: Maple Grove, MN USA
Registered: June 13, 2002 Posts: 382
posted
I just finished my experiments with this, and have two 8 bar test files: one with tracks normalized and EQ'ed and limited slightly, the other without normalization and limiting, but with EQ.
I believe that Thomas is correct: individual tracks should not be normalized or limited. The difference is subtle, but I think that the trained ear can hear the difference.
I am going to continue to test this out, but I think that I have learned a valuable lesson here today.
From: Europe
Registered: January 15, 2001 Posts: 1842
posted
quote:Originally posted by PD: I just finished my experiments with this, and have two 8 bar test files: one with tracks normalized and EQ'ed and limited slightly, the other without normalization and limiting, but with EQ.
I think you've got too many variables in your test. What we are discussing are purely the sonic differences between a normalised file and one that hasn't been normalised - the moment you throw in a limiter you're adding a whole plug-in meaning the sound is bound to change anyway, and the differences will surely be much more significant than anything one might be able to hear in the normalising process.
The listening test should be:
Audio File A - raw audio file Audio File B - normalised audio file, but reduced in volume manually on playback to match level of A
As James P pointed out it will be critical to line the two volumes up exactly as a slight discrepancy could fool the ear into thinking one is different, when it's only a matter of perceived loudness.
I am interested solely in the question of whether normalising an audio file creates audible artifacts, aside obviously from simply raising the level.
Actually that's not totally true because I'm also interested in how a digital mix bus works, because I don't understand this fully yet either, but this is a separate although related(!) question.
quote: James Lehmann: I am interested solely in the question of whether normalising an audio file creates audible artifacts, aside obviously from simply raising the level.
Yes it does
Taking your original questions:
quote: 1. Does Normalising change the sound in any way other than simply raise the level? And is this in theory or practice?
Yes. Mathematical errors in the process lead to distortion of the original.
Both.
quote: 2. Following on from (1), Is it true that different software will Normalise in different ways and that a file Normalised in Logic might sound different to one Normalised in Pro-Tools?
Not sure how it follows from 1 but yes, it's true. In essence the theory is the same but how each algorhythm approaches it is not necessarily the same.
quote: 3. Is it true that digital plug-ins (VST and the like) work better with some digital headroom? (ie do not Normalise to 0dB). What happens in a chain of plug-ins?
4. Following on from (3), What then is a sensible level to Normalise to?
Sorry, no idea - I don't use them often enough to know.
Ignoring the rights or wrongs of normalising individual tracks or files - which I'd suggest changes according to the content and context of the track - it's a mathematical process which creates errors. These errors represent a distortion of the original signal. How big a problem it is depends upon how sensitive one is to the finer details of sonic quality and whether or not the rest of your equipment is capable of resolving it. If the change is bigger than subtle you need to look at your software; it should be quite hard to detect. In practice it can be quite easy to hear the difference.
If you're getting into listening to these kinds of small differences, try making a copy of an audio file and ABing against the original. Then try copying an audio file from one drive to another and listening to it, or if you have some kind of multi drive rack, try physically moving a drive from one slot to another and listening to it. Some of these changes are very subtle - some aren't. Some are audible on pretty ordinary gear, some need really high end equipment before they show up in any real way.
To the people who berate you for even thinking about doing this I'd say if you're happy normalising individual tracks or files, it's giving you the sound you like and want and it's not causing you any problems elsewhere in the recording process, (e.g. complaining mastering engineers, crap sounding masters, bad reviews or complaints from your writing partner ), then carry on. Does it matter whether it's "right" or not?
From: Southsea
Registered: June 13, 2001 Posts: 3200
posted
James - I think your biggest problem is using the term normalising when all you want to do is change the level of something. Normalising has a very specific meaning whereas the examples you give simply involve changing the level of something without relating them to full scale. I think that you are probably getting a little too hung up on terminology here.
If you really want to take a purist approach then try to think about how many level changes you are doing in your whole recording process and try to minimise the number of changes.
From: peoplesrepublicofcork
Registered: December 06, 2000 Posts: 444
posted
OVU- "thanks" for the unintended synopsis. I get the impression that you sped read the thread...
I will attempt to present my points in a different manner:
1. It isn't James' problem that not only the term but the practice of "normalising" exists in DAWs because they evolved from sample editors. It's a legacy feature, that has its uses:
(i) Normalising samples for multisampling or mixing purposes makes sense. If you're gonna be transposing, filtering and chopping them, rounding errors and dither are the least of your worries!
(ii) Normalising takes in a multitrack recording is a bad idea. Your adding dither noise to each track, then adding those tracks together, and dithering again for the final mix. That's a big veil over your sound.
2. James- just take my advice about bit depth when bouncing, and you will retain clarity in your sound.
[This message was edited by Tomás Mulcahy on October 09, 2003 at 12:19 PM.]
Topic Closed
Next
