Topic: Does normalize affect sound quality?

[SpareRibs]

I recently purchased a Trey Anastasio download of FLAC files and when I used the Nero software to transfer to CD, I used the Normalize option. Was that fine? Did I diminish the sound quality or no differece?

[sygdwm]

try it both ways, decide for yourself and then post your results.

[Brian Skalinder]

Dre offers good advice regarding whether it's fine, or if diminished the sound quality.  Those are both subjective judgements.  As for any difference, it depends in part on how Nero performs normalization.  If Nero normalizes each file individually (which I suspect it does), then the results will produce individual songs that do not maintain their volume level relative to the entire set.  For example, if one song is quite loud, and has very strong levels, and a second song is very quiet, Nero may normalize the second, quiet song so it sounds as loud as the first.  This kind of normalization will impact the relative dynamic range across the entire song set.  Even if Nero takes into account the entire file set when normalizing (which I doubt), if Nero performs RMS normalization, instead of peak normalization, it may also impact dynamic range.  Personally, I'd leave Nero's normalization off unless / until I understood exactly what it does and how it does it.

[SpareRibs]

Thanks

[PH]

(lookout, long complicated answer below)

You do not want to normalize your files using any sort of program such as this. It will adversly affect quality.
If you do need to normalize any audio, you should do so with 24bit software that is designed to transparently perform that function.
(such as soundforge, wavlab, nuendo, protools, etc....)

When in 16 bit, any sort of volume/gain editing requires the program to perform a complicated set of algorithms and the result doesn't add up evenly, leaving artifacts.
This can be percieved as dullness or lifelessness with the recording. The more edits or actions you perform, the more this is compounded.
It's akin to a large number that is divided by a smaller number and leaves an uneven remainder.

When doing this type of editing, you need to import the 16bit into 24bit program and do the edits in that realm. The errors are dramaticlly less and the chance of these nasty artifacts are greatly diminished.
As a general rule, you should be very sparing with volume adjustments. If needed, use trial and error, and the undo function to find the right level/setting, then do it once and leave it alone.
You will need to dither the 24 bit file back down to 16bit afterwards. After the dither, there should be NO MORE "destructive" edits performed on the audio. If you edit after a file has been dithered, it will scramble the bit quantatization, or in other words, it will make it sound worse.

Most of those above things can and will happen if you use a program like itunes, nero, winamp, foobar, flac, or whatever to normalize your files.
Cheers, Phil

[boojum]

foobar allows applying ReplayGain, like normalizing, to FLAC as a one track at a time or as an album, where relative loudness is considered.  foobar is free.

[PH]

foobar allows applying ReplayGain, like normalizing, to FLAC as a one track at a time or as an album, where relative loudness is considered.  foobar is free.

I hope that I have explained why you do not want to use replay gain or normalizing using these type of quickie/on the fly type of gain adjustments.
This includes foobar. If you care about lossless copies, then you wouldn't consider doing this. If it's for your mp3 collection, then go right ahead, but just know that it will adversely affect the sound quality.

[boojum]

foobar allows applying ReplayGain, like normalizing, to FLAC as a one track at a time or as an album, where relative loudness is considered.  foobar is free.

I hope that I have explained why you do not want to use replay gain or normalizing using these type of quickie/on the fly type of gain adjustments.
This includes foobar. If you care about lossless copies, then you wouldn't consider doing this. If it's for your mp3 collection, then go right ahead, but just know that it will adversely affect the sound quality.

And the reason, in your opinion is . . . . . ?

I read your post just now, and understand your opinion.  Other than anecdotal have you any proof, like a series of ABX tests??  I am a little suspicious of sweeping generalizations like this.  You may be right.  But other than what you believe is there any indication elsewhere that this is true?  I am just curious.

Cheers

[PH]

foobar allows applying ReplayGain, like normalizing, to FLAC as a one track at a time or as an album, where relative loudness is considered.  foobar is free.

I hope that I have explained why you do not want to use replay gain or normalizing using these type of quickie/on the fly type of gain adjustments.
This includes foobar. If you care about lossless copies, then you wouldn't consider doing this. If it's for your mp3 collection, then go right ahead, but just know that it will adversely affect the sound quality.

And the reason, in your opinion is . . . . . ?

I read your post just now, and understand your opinion.  Other than anecdotal have you any proof, like a series of ABX tests??  I am a little suspicious of sweeping generalizations like this.  You may be right.  But other than what you believe is there any indication elsewhere that this is true?  I am just curious.

Cheers

It's audio basics. You can find plenty of information on the subject with a google search.

[StuStu]

I would not personally recommend using the normalize affect in Nero. Normalization (IMO) can improve or degrade audio quality. If you decide to normalize, I would use a respectable "audio" software program such as Sound Forge, Wavelab, etc. Nero is basically an affordable all-in-one program. I run it as well, I'm not knocking it for what it is.

For audio, I run Sound Forge 9 and depending on the specific recording, I either normalize OR adjust the volume.
Generic explanation...
Normalize: The levels are fairly consistent but the recording will sound better with levels raised.
Volume$$$: The recording is inconsisent in its recording levels and to normalize will not accurately improve or accurately reproduce the true sound of the performance. It will sound "unnatural."

$$$: A. This process is a much more time consuming pain in the ass B. The cause of the level indifference can be FOH, equipment related, the music itself, a heavily buzzed level-adjusting tapir , a heavily buzzed level-killing-clapping-wooing-neighbor...etc.

Brian and Phil's above posts are very well said.

YMMV

Stu
         

[bobbygeeWOW]

Just a note on replay-gain mentioned above - applying flac's replay-gain does not affect or alter the actual audio content of a flac file.
Its just a metatag (header) that music playing software can be set up to read. (or not)

The idea is that if you're listening in "random mode" with tracks randomly selected from lots of different sources, volume levels will jump around a lot. To avoid having to dive for your volume knob when a really loud -1db 16-bit track comes on right after that -18db 24-bit tune, the software will read this replay-gain tag and make the adjustment for you on the fly - essentially trying to keep each tune more or less at the same volume level.

I think most flac-supporting players (including my rockboxed ipod) have an option like "use replay-gain when playing randomly, otherwise don't bother".

When listening to a complete show, you want the loud parts loud and the quiet parts quiet, so you don't use replay-gain; but I find it pretty handy when I tell my SB3 to just play random background tracks while I fool around in the house doing stuff.

 -Cheers!

[PH]

I stand corrected on the Flac reply-gain setting since it doesn't alter the actual file, but it would essentially do the same thing as normalization but only for temporary listening.
I used to use this setting occasionly on my Rio Karma, but it's long dead.

What I'm specifically referring to when I talk about normalization being destructive is altering the actual file at the time of ripping or burning, or just to raise the levels to make it louder.
All of those things are ok to do, providing you upsample the audio before doing them to prevent any artifacts as were listed already.

Now, all of that being said, it's really not going to make that much a difference in the overall sound to make quick 16 bit edits, but you should know that it can and will leave artifacts.
If you are striving to preserve the integrity of the files you are ripping, such as Commercial CD's, it would not be advised to alter them in any way until after you have archived them on the hard drive, then alter a copy of the fileset.
Compare the two and there you have it.
Cheers,Phil

[boojum]

foobar allows applying ReplayGain, like normalizing, to FLAC as a one track at a time or as an album, where relative loudness is considered.  foobar is free.

I hope that I have explained why you do not want to use replay gain or normalizing using these type of quickie/on the fly type of gain adjustments.
This includes foobar. If you care about lossless copies, then you wouldn't consider doing this. If it's for your mp3 collection, then go right ahead, but just know that it will adversely affect the sound quality.

And the reason, in your opinion is . . . . . ?

I read your post just now, and understand your opinion.  Other than anecdotal have you any proof, like a series of ABX tests??  I am a little suspicious of sweeping generalizations like this.  You may be right.  But other than what you believe is there any indication elsewhere that this is true?  I am just curious.

Cheers

It's audio basics. You can find plenty of information on the subject with a google search.

Maybe you should have done that search yourself before you suggested it.     The results are mixed with the majority for normalizing.  The untilmate test, outlined in one article is to normalize up ndB and then normalize it back down the same amount and compare it to the original.  Tne author says they are bit-for-bit identical.  So unless there is a mysterious "something" that digital cannot capture - this can be used as an out - it would seem that it process is non-destructive.  In my software, Samplitude Master, it is within the project and applied only at burning time.

Links:

http://en.wikipedia.org/wiki/Audio_normalization

http://en.wikipedia.org/wiki/Replay_Gain

http://audioengineering.tribe.net/thread/a657da50-93d3-4c01-b356-6705c19f97e5

http://www.gearslutz.com/board/music-computers/31828-digital-audio-myths-never-normalize-etc.html

http://www.gearslutz.com/board/music-computers/150128-when-should-you-normalize-if-ever.html

http://www.gearslutz.com/board/low-end-theory/78763-normalize-not-normalize.html

http://www.gearslutz.com/board/music-computers/143280-normalize-function.html

http://recforums.prosoundweb.com/index.php/m/237663/17312/?srch=normalize#msg_237663

[Javier Cinakowski]

I would think that increasing the bit depth to 24, then back down to 16 would be more destructive than simple normailzation of the original 16bit file....

[boojum]

I want to say that I am not trying to be argumentative but to find out what is true and what is not.  Audio, as in other professions and hobbies, has some built in "ideas" that are not always true when subjected to reason and analysis.  This may or may not be the case with normalization.  I would like to know as I use it myself.  I use it in FLAC files (non-destructive there) and MP3's, same as FLAC.  The question, I think, that we are working on here is does Normalizing degrade WAV/PCM files when applied?  If it does, is it noticeable or immeasurable and/or undetectable by ear in ABX testing?

This is a pretty important question.  I, for one, would like to get a definitive answer.  Hopefully we can find it here or on a search of the data on the Internet.

Cheers   

[PH]

Boojum and others, these are facts, they are not opinions and it's important that you distinguish betweeen the two here.
You can believe whatever you like, but it's a fact that if you edit the gain of any track in any bit realm that you will alter the makeup of the track.
The suggestions that they are identical is ludicrous and misinformed. Try that test for yourself and show us the results.

Any gain editing is called "destructive" editing if it alters the actual file.
The point is to minimize the amount of errors or artifacts that are introduced when doing destructive editing. The best way to achieve this is to a 16 bit file is to upsample, edit, then downsample.
These are facts people.

Ask yourself, if this were folly or just some random opinion, then why are programs like Nuendo and Pro-Tools designed and created to work/edit in a 24 or 32 bit realm with non-destructive editing?
The information and knowledge is out there, so I don't need to do any AB tests, or look up silly articles on google to find out what I already know from years of personal experience working with digital audio to be fact.
and...yes I can hear the difference. That's what I get paid for.

[datbrad]

I want to say that I am not trying to be argumentative but to find out what is true and what is not.  Audio, as in other professions and hobbies, has some built in "ideas" that are not always true when subjected to reason and analysis. 

Some would say this is true of many of the conventions of audio. For example, the theoretical 144 db dynamic range performance of 24bit versus the actual realized dynamic range as limited by the analog front end performance in real world recorders.

[PH]

I would think that increasing the bit depth to 24, then back down to 16 would be more destructive than simple normailzation of the original 16bit file....

Another misnomer. Upsampling a 16 bit file to any other larger bit depth will only add zeros to the end, making the file have a much larger bandwidth to work with for editing, thus minimizing the aritifacts caused by destructive editing of a narrow 16 bit file. Downsampling applies dither, which effectively widens the scope of the 16 bit file with various techniques to hide or mask any artifacts caused by the downsample. In other words, it moves the artifacts to a non audible realm in most cases. What this means is that any artifacts caused by the gain adjustment or downsample are smoothed away and leave virtually no trace of ever happened, which is not the case with 16 bit editing. There is no such built in noise or artifact cover up, leaving the artifacts in the audible realm.

This is really the entire point of this thread.

[Javier Cinakowski]

I understand that nashphil, my point is that normalizing a 16bit file might be less destructive than upsampling, normaizing, dithering and downsampling....   If you are doing serious editing such as eq, effects or filters than upsampling is certainly going to help the situation.  I just think upsampling a 16bit file for the simple task of normalizing is a bit much....   

[PH]

I understand that nashphil, my point is that normalizing a 16bit file might be less destructive than upsampling, normaizing, dithering and downsampling....   If you are doing serious editing such as eq, effects or filters than upsampling is certainly going to help the situation.  I just think upsampling a 16bit file for the simple task of normalizing is a bit much....   

As I have already stated, it doesn't work that way if you are looking to PRESERVE the audio in a lossless and (minimally) non destructive way. 
I don't understand why that isn't clear to anyone reading this thread. It's a really simple concept based on fact, not myth. 

As has been stated already, it depends on what you want to normalize it for as to what you should do.
If you don't care about artifacts, then go for it. If you do, then try using a professional audio editor to edit your files.
Soundforge, Wavlab, and Nuendo all get plenty of use in my house. They are designed to handle these type of issues for you.
Programs like Nero are not.

Judge for yourself. Your ears will answer these questions for you.
If you can't hear it, then you can't hear it and it doesn't matter, what else can I say.

<<talk amongst yourselves>>

[datbrad]

Phil, What is your opinion when recording at 16 bit regarding 48khz sampling, then resampling to 44.1, versus just mastering at 44.1? I have heard mixed opinions on that. Some say that the analog anti-alaising filters in ADs at 44.1 impact the sound quality more than doing 48khz, since the filters are kicking in at a higher frequency. Then, the thinking goes, resampling to 44.1 delivers a better end result than if mastered at 44.1 originally, supposedly. I have always stayed at 48khz, but now wonder if I am really wasting time with that step.

I ask you this because at one time, resampling was thought to be a very destructive thing to do to audio files. I suppose with today's higher end software, this is not as much an issue, but you are the studio guy and should know.

[PH]

Phil, What is your opinion when recording at 16 bit regarding 48khz sampling, then resampling to 44.1, versus just mastering at 44.1? I have heard mixed opinions on that. Some say that the analog anti-alaising filters in ADs at 44.1 impact the sound quality more than doing 48khz, since the filters are kicking in at a higher frequency. Then, the thinking goes, resampling to 44.1 delivers a better end result than if mastered at 44.1 originally, supposedly. I have always stayed at 48khz, but now wonder if I am really wasting time with that step.

I ask you this because at one time, resampling was thought to be a very destructive thing to do to audio files. I suppose with today's higher end software, this is not as much an issue, but you are the studio guy and should know.

This is a very good question!
From 2000 to up until about a year ago I was recording everything at 24bit/48k, both in the studio and for live stuff too. A well known Nashville engineer and I were talking about this very subject and after listening to his thoughts and then following up with some of my own tests, I changed my mind. I now record at 24bit/44.1k.

Since 99% of what I do is to wind up at 16/44, why resample. I don't ever mix anything down to 16/48.
The differences in quality are just NOT great enough to warrant the extra time, space, and potential artifacts from the resample.
I decided a long time ago that 96k is a ridiculous waste of space and really more about gear hype than actual improvements. The real difference comes at the 16/24 bit level.

Now, if I were still recording at 16bit in the field, I would use 48k. I have always sensed a difference in that at 16bit, even going back to the DAT days in the 90's.
Once you reach the 24 bit realm, that difference is virtually impercievable from 44k to 96k. Some would argue this point I'm sure, but my experience tells me otherwise.
I guess the differences are probably pretty small here, so I don't waste time with the extra step. Makes my life easier.

[Josephine]

I am so happy to read your comments above, Phil.  I've recorded two shows now at 96k (just because I could) and have come home wondering why.  The files are massive and will be a pain in the ass to store. 

So despite my "ability" to record at 16 or 24, at 44.1, 48 or 96, you'd recommend running @ 24/44.1?  It indeed would make life so much easier. 

[boojum]

Boojum and others, these are facts, they are not opinions and it's important that you distinguish betweeen the two here.
You can believe whatever you like, but it's a fact that if you edit the gain of any track in any bit realm that you will alter the makeup of the track.
The suggestions that they are identical is ludicrous and misinformed. Try that test for yourself and show us the results.

Any gain editing is called "destructive" editing if it alters the actual file.
The point is to minimize the amount of errors or artifacts that are introduced when doing destructive editing. The best way to achieve this is to a 16 bit file is to upsample, edit, then downsample.
These are facts people.

Ask yourself, if this were folly or just some random opinion, then why are programs like Nuendo and Pro-Tools designed and created to work/edit in a 24 or 32 bit realm with non-destructive editing?
The information and knowledge is out there, so I don't need to do any AB tests, or look up silly articles on google to find out what I already know from years of personal experience working with digital audio to be fact.
and...yes I can hear the difference. That's what I get paid for.

It's audio basics. You can find plenty of information on the subject with a google search.

Phil, I am sure you are absolutely convinced of what you say being the actual fact.  I am trying to ferret out proof of this as I do not know what you know.  Some folks in the business still swear analog is better.  It may sound nicer but that is a euphonious distortion.  It is not as accurate because it is distorted.  I am trying to get to where I, we, can determine whether this is the case in normalizing or not.  What you say may be true.

But when you first say "google it" and then say the googled stuff is silly you weaken you argument a bit.  That does not make your argument wrong, though.  I am sure you are way better a ME than I will ever be.  But I am just trying to find a definitive proof of this, that normalizing is bad as it "adversely affects sound quality."   I have not tried the experiment suggested with normalizing a file and then restoring it and comparing it to the original.  I am not sure I can do it in SAM, maybe in Cool Edit.  If it can do this the argument is done.  If it cannot it is still up in the air as we are looking for "adverse effects" and maybe we should define our terms.  

On gearlsutz and PSW they are divided on this.  There is another "pro" board out there I have forgotten, but I will try there if I can remember it.  

The search continues.

Cheers

[PH]

I am so happy to read your comments above, Phil.  I've recorded two shows now at 96k (just because I could) and have come home wondering why.  The files are massive and will be a pain in the ass to store. 

So despite my "ability" to record at 16 or 24, at 44.1, 48 or 96, you'd recommend running @ 24/44.1?  It indeed would make life so much easier. 

Yes, running at 24/44 would be the best. There would be virtually no benefit to running at any sample rate above that.

[Gutbucket]

...Now, if I were still recording at 16bit in the field, I would use 48k. I have always sensed a difference in that at 16bit, even going back to the DAT days in the 90's.
Once you reach the 24 bit realm, that difference is virtually impercievable from 44k to 96k..

I totally agree on NOT using an algorithm that is not understood and found in some inexpensive or free software (mostly because it is probably not as well designed or thought out) to adjust the level of one's master recordings.

I also totally agree on 96kHz being overkill and that the true quality jump is adding those 4 or 8 bits over 16 to leave headroom for real world recording (I think its harder to justify that those bits are necessarily the delivery medium).  I'll take your word on the 48 vs 44.1 difference since I've never really tested it.  I doubt I'd hear much of a difference since I don't in very casual comparisons.  However, I don't follow the part above.  Why would choice of bit depth have any influence on bit rate? Specifically in this case, why would a lower bit depth recording benefit from a slightly higher sampling rate?  Conversely why would the increased dynamic range of a 24 bit format effect what should be frequency artifact issues with sample rate changes?


As for the original question- From the theory side, if you analyze a WAV file you can find the highest and lowest sample values. There is no information stored above or below those values.  Assuming you use good tools and understand the implications and real world issues that those tools and their use may have on the file and if you use the correct practice to do so, there is no information lost when shifting the entire range of recorded values in the file up or down within the larger range of the allowable values determined by the bit depth of the file. Doing a simple gain addition to all sample values of the file (which is what we are talking about 'in theory') doesn't loose any information as long as that addition keeps the highest sample value beneath the maximum value allowable.  Experienced users would suggest using good tools to do so and leaving some room at the top.  Doing so shifts the range of values up into the unused portion at the top and adds useless zeros to the bottom.  Same goes for adjusting the other way.

In the real world nothing is simple or straight forward.  If you are doing any other manipulations of the file there are implications and complications.  You need enough 'padding' around the edges of the information to allow for those changes.  Here's a good question, "what do they really mean by normalizing?"

Does normalizing affect sound quality? Like everything the answer is "it depends".  If you are a realist who understands the complications of real world software and equipment because you work with it for a living everyday like Phil, the answer is "Yes usually, unless you are sure of what's going on and use good tools".  If you are a theorist the answer is, "No, with a lot of qualifications".  If you understand the complications, you'll understand those answers approach the question from opposite directions, but are both correct and communicate the same concerns and qualifications. 

[jerryfreak]

for what its worth, soundforge and most other decent wave editors do all the calcs in 32-bit floating point anyway.

normalization done *once* with software like this should essentially sound as good as the original

I would think that increasing the bit depth to 24, then back down to 16 would be more destructive than simple normailzation of the original 16bit file....