IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
normalizing - something already peaking above 0.0dB
bomber
post Jan 14 2010, 02:25
Post #1





Group: Members
Posts: 81
Joined: 13-October 08
Member No.: 60003



Hey guys,

The other day I compiled a CD of different stuff and normalized each file/song to peak at 100% volume without clipping.
After I had done this out of curiosity I checked what the peak volume of some of the original files were, some were very low which is the reason I normalized each track I was putting on the CD, however some of the tracks I was compiling were peaking at values like +0.02dB from their original source.

So when I normalized the +0.02dB file it changed to something just under 0.0dB.

I'm having trouble getting my head around what happens to a file when you have an original file that peaks at +0.02dB (already over 0.0dB), and you then normalize it to 100% (which takes it just under 0.0db).

I know a human is not going to hear a difference of something like .2dB, but does the process I described above cause some of the audio to be chopped off ?, any audible damage to the file ?
I know it's a simple answer, I just can't get my head around it at the moment !

Thanks for any help.

Cheers

This post has been edited by bomber: Jan 14 2010, 02:28
Go to the top of the page
+Quote Post
MedO
post Jan 14 2010, 13:08
Post #2





Group: Members
Posts: 341
Joined: 24-August 05
Member No.: 24095



Assuming that you did something like mp3 --decode-> wav --normalize-> wav --burn-> CD and you used normal 16-bit PCM wav, you might have clipped peaks on your CD. The reason is that formats like mp3 can represent levels above 0db when decoded. The lossy encoding process can create peaks that high even if the source data for the mp3 never exceeded that level.

So when decoding an mp3 with peaks above 0db to 16 bit wav which can't represent these levels, the samples will be clipped. Normalizing the wav again after the fact won't help, since the clipping already happened.

However, if you use an intermediate wav format that can represent the peaks and normalize that before going to 16 bits, or do something like mp3 --mp3gain-> mp3 --decode-> wav or mp3 --decode with replaygain-> wav, then no clipping should occur.

Either way, clipping of such a low intensity as you describe is in all probability inaudible.

This post has been edited by MedO: Jan 14 2010, 13:09
Go to the top of the page
+Quote Post
bomber
post Jan 14 2010, 14:05
Post #3





Group: Members
Posts: 81
Joined: 13-October 08
Member No.: 60003



QUOTE (MedO @ Jan 14 2010, 23:08) *
Assuming that you did something like mp3 --decode-> wav --normalize-> wav --burn-> CD and you used normal 16-bit PCM wav, you might have clipped peaks on your CD. The reason is that formats like mp3 can represent levels above 0db when decoded. The lossy encoding process can create peaks that high even if the source data for the mp3 never exceeded that level.

So when decoding an mp3 with peaks above 0db to 16 bit wav which can't represent these levels, the samples will be clipped. Normalizing the wav again after the fact won't help, since the clipping already happened.

However, if you use an intermediate wav format that can represent the peaks and normalize that before going to 16 bits, or do something like mp3 --mp3gain-> mp3 --decode-> wav or mp3 --decode with replaygain-> wav, then no clipping should occur.

Either way, clipping of such a low intensity as you describe is in all probability inaudible.


Some were MP3 and some were WAV.
The process I used was opening the original files with Adobe Audition 1.5 and Normalizing them to 100%
Thanks for the info.
Go to the top of the page
+Quote Post
pdq
post Jan 14 2010, 14:22
Post #4





Group: Members
Posts: 3305
Joined: 1-September 05
From: SE Pennsylvania
Member No.: 24233



It is possible for a digital file to have no data points greater than full scale, but when converted to analog and lowpassed, to exceed full scale. Perhaps that is what is going on, i.e. the software that is calculating peak value is calculating true analog peak value, rather than maximum digital value?

BTW, normalizing to the highest peak value is a terible way to adjust your files.
Go to the top of the page
+Quote Post
Notat
post Jan 14 2010, 16:37
Post #5





Group: Members
Posts: 581
Joined: 17-August 09
Member No.: 72373



It would be nice if software would do the calculations pdq is talking about. Most don't - they just look for the largest sample value in the file and scale everything up so that the largest sample is at full scale. This is wrong but not hugely wrong. I don't know what Adobe Audition does. bombers explanation is probably more likely. The peak value of an encoded file is more difficult to predict.

In any case, you have no business normalizing files that are already this close to full scale. Nothing good will come of it.
Go to the top of the page
+Quote Post
DVDdoug
post Jan 14 2010, 19:00
Post #6





Group: Members
Posts: 2441
Joined: 24-August 07
From: Silicon Valley
Member No.: 46454



QUOTE
I'm having trouble getting my head around what happens to a file when you have an original file that peaks at +0.02dB (already over 0.0dB), and you then normalize it to 100% (which takes it just under 0.0db).
It doesn't make any sense to me either! I suppose Audition could be calculating/estimating "inter-sample peaks", but it could also be a calculation or rounding error...

0dBFS is defined as the highest possible sample value (i.e. 100%). With 16-bit WAV files that means either +32,767 or -32,768 (decimal). You simply cannot "count" any higher than that (in binary) without adding another bit!


NOTE - Audio editors normally use floating-point numbers for temporary storage, so it is possible to go over 0dB within Adobe (i.e. if you increase the volume, or do some other processing that boosts the level). But, if you save a file that goes over 0dB (in the standard 16-bit WAV format) it will be clipped to 0dB.

Go to the top of the page
+Quote Post
MedO
post Jan 14 2010, 20:44
Post #7





Group: Members
Posts: 341
Joined: 24-August 05
Member No.: 24095



QUOTE
It would be nice if software would do the calculations pdq is talking about. Most don't - they just look for the largest sample value in the file and scale everything up so that the largest sample is at full scale. This is wrong but not hugely wrong.

That depends what you want to achieve. I think that for most applications it's irrelevant whether the level of the reconstructed analog signal would exceed 0db between sample points because the reconstruction filters should have no problem with that case.

QUOTE (DVDdoug @ Jan 14 2010, 19:00) *
QUOTE
I'm having trouble getting my head around what happens to a file when you have an original file that peaks at +0.02dB (already over 0.0dB), and you then normalize it to 100% (which takes it just under 0.0db).
It doesn't make any sense to me either! I suppose Audition could be calculating/estimating "inter-sample peaks", but it could also be a calculation or rounding error...

0dBFS is defined as the highest possible sample value (i.e. 100%). With 16-bit WAV files that means either +32,767 or -32,768 (decimal). You simply cannot "count" any higher than that (in binary) without adding another bit!


NOTE - Audio editors normally use floating-point numbers for temporary storage, so it is possible to go over 0dB within Adobe (i.e. if you increase the volume, or do some other processing that boosts the level). But, if you save a file that goes over 0dB (in the standard 16-bit WAV format) it will be clipped to 0dB.


I suppose Audition will directly render mp3 files into its internal processing format - which can probably represent sample values above 0db as you already said. Mp3 files can easily contain sample values above 0db even if the material they were encoded from did not, so the result that you get is not surprising to me.
Go to the top of the page
+Quote Post
greynol
post Jan 14 2010, 20:52
Post #8





Group: Super Moderator
Posts: 10000
Joined: 1-April 04
From: San Francisco
Member No.: 13167



I'm pretty sure Audition detects inter-sample overs. I'm also pretty sure that it doesn't make allowances for mp3s with peak values that exceed 0dB FS.


--------------------
Your eyes cannot hear.
Go to the top of the page
+Quote Post
DVDdoug
post Jan 14 2010, 21:03
Post #9





Group: Members
Posts: 2441
Joined: 24-August 07
From: Silicon Valley
Member No.: 46454



QUOTE
Some were MP3 and some were WAV.
...I missed that. Yes, decoded MP3s can have sample values above 0dBFS.
Go to the top of the page
+Quote Post
pdq
post Jan 14 2010, 21:11
Post #10





Group: Members
Posts: 3305
Joined: 1-September 05
From: SE Pennsylvania
Member No.: 24233



QUOTE (MedO @ Jan 14 2010, 15:44) *
QUOTE
It would be nice if software would do the calculations pdq is talking about. Most don't - they just look for the largest sample value in the file and scale everything up so that the largest sample is at full scale. This is wrong but not hugely wrong.

That depends what you want to achieve. I think that for most applications it's irrelevant whether the level of the reconstructed analog signal would exceed 0db between sample points because the reconstruction filters should have no problem with that case.

And let's not forget that regardless of how you calculate peak values, normalizing to the highest peak is a lousy way to adjust the loudness of your files. The only case that I can think of where it might make sense to do such normalization is if one is processing audio data and is limited to 16 bit integers, so you want to maximize the available dynamic range.
Go to the top of the page
+Quote Post
bomber
post Jan 17 2010, 03:33
Post #11





Group: Members
Posts: 81
Joined: 13-October 08
Member No.: 60003



QUOTE (pdq @ Jan 15 2010, 07:11) *
QUOTE (MedO @ Jan 14 2010, 15:44) *
QUOTE
It would be nice if software would do the calculations pdq is talking about. Most don't - they just look for the largest sample value in the file and scale everything up so that the largest sample is at full scale. This is wrong but not hugely wrong.

That depends what you want to achieve. I think that for most applications it's irrelevant whether the level of the reconstructed analog signal would exceed 0db between sample points because the reconstruction filters should have no problem with that case.

And let's not forget that regardless of how you calculate peak values, normalizing to the highest peak is a lousy way to adjust the loudness of your files. The only case that I can think of where it might make sense to do such normalization is if one is processing audio data and is limited to 16 bit integers, so you want to maximize the available dynamic range.


Thanks for all the bits of info guys...
Yeah I know there's no point normalizing any file which is originally already so close to 0.0dB, but as I said in one of my first posts those particular files were in a batch of quite a few files which I normalized because a lot of them were far below 0.0dB (eg. so I normalized all the files I wanted to put on a CD as a whole batch), and didn't even notice there were files so close to 0.0dB until after I had done so...

After I noticed that I just wanted to know what the process/circumstance I described was technically doing to the file - now I know at worse it was probably clipping .4dB
I guess it's no big deal, it's just a CD for my personal use, nothing of importance that needed technically correct editing, and as a few guys said in this thread, if it has clipped .4dB I guess no human ear will ever be able to detect it anyay !

Cheers

Go to the top of the page
+Quote Post
bomber
post Jan 17 2010, 03:41
Post #12





Group: Members
Posts: 81
Joined: 13-October 08
Member No.: 60003



QUOTE (pdq @ Jan 15 2010, 00:22) *
It is possible for a digital file to have no data points greater than full scale, but when converted to analog and lowpassed, to exceed full scale. Perhaps that is what is going on, i.e. the software that is calculating peak value is calculating true analog peak value, rather than maximum digital value?

BTW, normalizing to the highest peak value is a terible way to adjust your files.


I don't understand why you think normalizing a song so the file is max volume without clipping is so terrible ?
Maybe you're being extremely technical, because I'm sure apart from the volume being raised nobodys ears will be able to detect any error in sound or notice any missing dynamics...
*I respect that 'technically' it's not the greatest thing, but there's really not going to be any 'audible' negative effect is there?

This post has been edited by bomber: Jan 17 2010, 03:43
Go to the top of the page
+Quote Post
MedO
post Jan 17 2010, 12:40
Post #13





Group: Members
Posts: 341
Joined: 24-August 05
Member No.: 24095



QUOTE (bomber @ Jan 17 2010, 03:41) *
QUOTE (pdq @ Jan 15 2010, 00:22) *
It is possible for a digital file to have no data points greater than full scale, but when converted to analog and lowpassed, to exceed full scale. Perhaps that is what is going on, i.e. the software that is calculating peak value is calculating true analog peak value, rather than maximum digital value?

BTW, normalizing to the highest peak value is a terible way to adjust your files.


I don't understand why you think normalizing a song so the file is max volume without clipping is so terrible ?
Maybe you're being extremely technical, because I'm sure apart from the volume being raised nobodys ears will be able to detect any error in sound or notice any missing dynamics...
*I respect that 'technically' it's not the greatest thing, but there's really not going to be any 'audible' negative effect is there?


I think pdq means that you normally want to adjust the volume of the songs on a compilation CD so that they have a similar perceived loudness, and normalizing is not good at achieving that. Songs with low dynamic range will end up much louder than those with high dynamic range if you just go by peak value.
Go to the top of the page
+Quote Post
bomber
post Jan 17 2010, 14:56
Post #14





Group: Members
Posts: 81
Joined: 13-October 08
Member No.: 60003



QUOTE (MedO @ Jan 17 2010, 22:40) *
QUOTE (bomber @ Jan 17 2010, 03:41) *
QUOTE (pdq @ Jan 15 2010, 00:22) *
It is possible for a digital file to have no data points greater than full scale, but when converted to analog and lowpassed, to exceed full scale. Perhaps that is what is going on, i.e. the software that is calculating peak value is calculating true analog peak value, rather than maximum digital value?

BTW, normalizing to the highest peak value is a terible way to adjust your files.


I don't understand why you think normalizing a song so the file is max volume without clipping is so terrible ?
Maybe you're being extremely technical, because I'm sure apart from the volume being raised nobodys ears will be able to detect any error in sound or notice any missing dynamics...
*I respect that 'technically' it's not the greatest thing, but there's really not going to be any 'audible' negative effect is there?


I think pdq means that you normally want to adjust the volume of the songs on a compilation CD so that they have a similar perceived loudness, and normalizing is not good at achieving that. Songs with low dynamic range will end up much louder than those with high dynamic range if you just go by peak value.


Yeah that's true, it would be a big problem if all the songs were from the same album which had been mixed / produced like that, but it's just a compilation I've made of all sorts of bits and pieces so that's not really a problem.
Go to the top of the page
+Quote Post
greynol
post Jan 17 2010, 18:35
Post #15





Group: Super Moderator
Posts: 10000
Joined: 1-April 04
From: San Francisco
Member No.: 13167



...and that's precisely why it is not a good idea. Of all those bits and pieces, how do you know they were recorded/mixed/mastered with the same EQ settings and dynamic range compression in order to provide equal loudness through peak normalization?


--------------------
Your eyes cannot hear.
Go to the top of the page
+Quote Post
bomber
post Jan 18 2010, 10:36
Post #16





Group: Members
Posts: 81
Joined: 13-October 08
Member No.: 60003



QUOTE (greynol @ Jan 18 2010, 04:35) *
...and that's precisely why it is not a good idea. Of all those bits and pieces, how do you know they were recorded/mixed/mastered with the same EQ settings and dynamic range compression in order to provide equal loudness through peak normalization?


If they are from different sources they are obviously not recorded / mixed / mastered / equalized the same way, therefore leaving them at the original volume they will never have equal loudness anyway...
By normalizing all the files individually I'm not trying to achieve "equal" loudness across all tracks, I'm doing it so each track is as loud as it can be without clipping, the only thing that is doing is raising the volume (just as a volume knob does).
Reason for doing so is the CD I've made is being played on a very low powered / volume system and some of the original files were extremely low volume, so the normalizing process is for nothing other than to achieve the highest volume of each individual track withouth clipping.
Go to the top of the page
+Quote Post
greynol
post Jan 18 2010, 16:52
Post #17





Group: Super Moderator
Posts: 10000
Joined: 1-April 04
From: San Francisco
Member No.: 13167



So you're left adjusting volume between tracks, whereas with a replaygain-based approach you won't have this problem.


--------------------
Your eyes cannot hear.
Go to the top of the page
+Quote Post
bomber
post Jan 19 2010, 00:35
Post #18





Group: Members
Posts: 81
Joined: 13-October 08
Member No.: 60003



QUOTE (greynol @ Jan 19 2010, 02:52) *
So you're left adjusting volume between tracks, whereas with a replaygain-based approach you won't have this problem.


It worked out without having to adjust volume between tracks, they're all from the same band (but different albums / sources etc), and all pretty loud type songs, so it worked out...
Thanks again for all your opinions / info guys.
Go to the top of the page
+Quote Post
willow
post Jan 21 2010, 14:29
Post #19





Group: Members
Posts: 28
Joined: 15-January 09
From: Russia
Member No.: 65602



QUOTE (Notat @ Jan 14 2010, 18:37) *
It would be nice if software would do the calculations pdq is talking about. Most don't - they just look for the largest sample value in the file and scale everything up so that the largest sample is at full scale. This is wrong but not hugely wrong. I don't know what Adobe Audition does. bombers explanation is probably more likely. The peak value of an encoded file is more difficult to predict.

In any case, you have no business normalizing files that are already this close to full scale. Nothing good will come of it.


+1
Personally i think your post is undervalued. Moreover most of the files around already trade the audio details for raw pressure power. Take those chinese mp3's for the local cellphones market, they do bring 10x power for Real out of nothing, in exchange of zero dynamic range and having details raped off - who care. The communist anthem must dominate in city traffic lol. Ok, take another example, i'll assure you the Russian audio market was ruined because of license media with overpowered audio levels and not because of folks criminal intensions. Ideally no additional processing needed after finalization stage on the record studio. All the normalizing by the user himself is the freaking sin. The main problem is, the equal loudness can't be achived because the effective loudness for some files already above 0.0db! What i'm doing in my own practice is lowering the audio level of the overpowered files down to 60-80% to make the listening experience somewhat uniform and comfortable while leaving the rest of files (the "normal" ones) alone.


--------------------
maiko.elementfx.com
Go to the top of the page
+Quote Post
bomber
post Jan 25 2010, 06:41
Post #20





Group: Members
Posts: 81
Joined: 13-October 08
Member No.: 60003



QUOTE (willow @ Jan 22 2010, 00:29) *
QUOTE (Notat @ Jan 14 2010, 18:37) *
It would be nice if software would do the calculations pdq is talking about. Most don't - they just look for the largest sample value in the file and scale everything up so that the largest sample is at full scale. This is wrong but not hugely wrong. I don't know what Adobe Audition does. bombers explanation is probably more likely. The peak value of an encoded file is more difficult to predict.

In any case, you have no business normalizing files that are already this close to full scale. Nothing good will come of it.


+1
Personally i think your post is undervalued. Moreover most of the files around already trade the audio details for raw pressure power. Take those chinese mp3's for the local cellphones market, they do bring 10x power for Real out of nothing, in exchange of zero dynamic range and having details raped off - who care. The communist anthem must dominate in city traffic lol. Ok, take another example, i'll assure you the Russian audio market was ruined because of license media with overpowered audio levels and not because of folks criminal intensions. Ideally no additional processing needed after finalization stage on the record studio. All the normalizing by the user himself is the freaking sin. The main problem is, the equal loudness can't be achived because the effective loudness for some files already above 0.0db! What i'm doing in my own practice is lowering the audio level of the overpowered files down to 60-80% to make the listening experience somewhat uniform and comfortable while leaving the rest of files (the "normal" ones) alone.


Are we entering here into "loudness war" territory ?, I think the loud volume a lot of bands this day and age are mastering their studio albums at, and the reasons why is inexcusable total garbage, I can't believe some of the so called "professional" artists who have lowered their standards by conforming to that crap.
Go to the top of the page
+Quote Post
neelX
post Jan 25 2010, 21:50
Post #21





Group: Members
Posts: 10
Joined: 17-July 08
Member No.: 55904



QUOTE (greynol @ Jan 14 2010, 21:52) *
... I'm also pretty sure that it doesn't make allowances for mp3s with peak values that exceed 0dB FS.


IIRC from the times of CoolEdit Pro, that entirely depended on actual plugin used for mp3 decoding. Maybe some (default out of the box) plugin cuts them to 16-bit integers (and so clips to 0dBFS).
Go to the top of the page
+Quote Post

Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 20th April 2014 - 12:57