I normalised...is all lost?

[roosterino]

I'm mixing a 40 track project for a synagogue, (some cantorial pieces but mostly short liturgical songs with guitar, bass and vocals). Being as this is basically volunteer work, they gave it to someone completely unqualified.

I finished the mixing using Adobe Audition 3.0 but I normalised all premasters. I normalised it to 90% to give myself some room in mastering. I thought most premasters were too low in volume and wanted to get them to a pretty consistent level for mastering. Some premasters were ok and I left them alone.

I mastered using uad plugs - Neve 33609, 88rs, Pultec Pro, and Ozone Multi-band compression (native to Audition).

It seems, after reading the posts on this forum, I committed a cardinal sin and need to be shot, or even worse....have to run all the mixes again without normalising, and re-master all the tracks.

Do you mastering people think it's wothwhile to re-do it? I'm not sure I understand why boosting the volume totally in mastering using the compression plugs would make such a huge difference.

I'm not even sure how I would get the volume hot enough using those plugs except just cranking the output volume.

thanks...

[Fluxpod]

Give it to a guy who knows what he is doing.

EDIT: Btw..how does it sound? Do you think it is ok? Good? If so its maybe just alright...

[t_d]

all normalizing does is calculate the difference in dB from your highest peak and zero and then raises the gain by that amount.

it dies no "damage" just raises the volume by a uniform amount.

the reason people scream in horror here about normalization is because it traditionally brings your peak's up to zero thereby leaving no room for the mastering engineer... and THAT is a bad thing.

the fact that you are not using an outside mastering engineer means you should be fine and by only going to 90% you have left yourself some headroom.

the only suggestion I would make is to always save multiple copies of your work. don't erase over the in-normalized file. that way if you DO need to process it further or send it to a ME you have that original starting point.

[Ghostwheel]

Audition uses 32 bit audio by default so who cares. Just amplify it down by however many db's your heart desires. Are you going to lose some quality? Yeah, possibly, if you have alien ears... I wouldn't get too upset about it...

[EvilE]

Originally Posted by t_d

the only suggestion I would make is to always save multiple copies of your work.

ALWAYS DO THIS!!! I always save as a new session before getting into any potentially irreversable behaviors. i also make duplicate playlists (PT's duh) of anything before i consolidate, audio suite or any other type of function that makes a new audio file.

[Lagerfeldt]

Originally Posted by t_d

it does no "damage" just raises the volume by a uniform amount.

Not quite. It depends.

Normalizing a 16 bit source in many DAWs will add extra noise or errors as it truncates or re-dithers to the original bit format of the file, i.e. 16 bit. Raising and lowering the fader afterwards or processing the sound will then expand into 32 bit float since it's not dependent on the output file format. So you've added an unnecessary degrading step to your audio by doing destructive normalization.

Some people won't find it very audible but doing it on multiple files - especially 16 bit - in a mix will eventually lead to some clearly audible extra noise and/or midrange protuding.

Originally Posted by Ghostwheel

Audition uses 32 bit audio by default so who cares. Just amplify it down by however many db's your heart desires. Are you going to lose some quality? Yeah, possibly, if you have alien ears... I wouldn't get too upset about it...

See above.

[peeder]

Yeah if the file is 32-bit float and you normalized to 32-bit float you are OK. But most normalize functions take a fixed-point format (24 or 16bit) back into a fixed point format and thus do damage the audio. Normalization to peak levels doesn't provide equal loudness, and to RMS levels only vaguely provides equal loudness.

Most importantly, your mastering engineer hasn't been given any help at all by your normalization. They could have adjusted the gain themselves, and will ALWAYS have to anyway, while doing so in a minimally damaging way.

[roosterino]

thanks for the replies.

so....since Audition is at 32 bits and my mix is 24bit, I should be ok. My non-alien (yet somewhat alien-looking) ears tells me most the master are ok.

I'm still curious as to why when the mastering engineer boosts the volume, he can do so with impunity but boosing it with a normalisation (which I'm understanding as no different that just boosting the volume on master faders when running the mix) runs risk of degradation.

[Ghostwheel]

Originally Posted by roosterino

thanks for the replies.
so....since Audition is at 32 bits and my mix is 24bit, I should be ok. My non-alien (yet somewhat alien-looking) ears tells me most the master are ok.

With 24bit audio (although Audition's 32bit floating is probably even better), you can do whatever amplifying audio operation you want and your audio will be fine. I think people get obsessed over this "normalization degradation" issue a bit too much.

Just try the following:

Get any 24bit audio source and amplify it up or down by 80db (yes eighty) up or down (it won't clip). Now reverse the amplification in the other direction and paste the result inverted over your original audio. It will probably produce silence, and if not the noise floor is going to be at something like -150 dbfs. I can live with that! :)

Now, I do agree about 16 bit though. Never ever record or process in 16 bit!

[peeder]

Originally Posted by Ghostwheel

With 24bit audio (although Audition's 32bit floating is probably even better), you can do whatever amplifying audio operation you want and your audio will be fine. I think people get obsessed over this "normalization degradation" issue a bit too much.

Just try the following:

Get any 24bit audio source and amplify it up or down by 80db (yes eighty) up or down (it won't clip). Now reverse the amplification in the other direction and paste the result inverted over your original audio. It will probably produce silence, and if not the noise floor is going to be at something like -150 dbfs. I can live with that! :)

Now, I do agree about 16 bit though. Never ever record or process in 16 bit!

False. If you trim 80db off of 24-bit fixed point audio, save to a 24-bit file, and then add 80db of gain, you have only 144 - 80 = 64db of dynamic range left. Your audio is then only effectively 10.7bit, and you will have a similar noise floor to cassette tape.

You should know that posting in a mastering forum.

[dkatz42]

Originally Posted by peeder

False. If you trim 80db off of 24-bit fixed point audio, save to a 24-bit file, and then add 80db of gain, you have only 144 - 80 = 64db of dynamic range left. Your audio is then only effectively 10.7bit, and you will have a similar noise floor to cassette tape.

You should know that posting in a mastering forum.

This is getting pretty far afield from the reality that posed the original question. In particular, as long as the normalization was positive (added gain) there is no truncation error--the bit depth is not affected.

There will be rounding errors in the low-order bits, but the same thing happens when you have a fader or any other gain widget in-line in your DAW. Put another way, the impact of normalizing is *exactly* the same (save for its permanence) as adding a submix fader in your DAW. If you don't worry about one multiply there, don't worry about one multiply here.

[UnderTow]

Originally Posted by dkatz42

Put another way, the impact of normalizing is *exactly* the same (save for its permanence) as adding a submix fader in your DAW.

Not entirely. Most DAWs run at 32 or 64 bit float internally. If you normalize (and save) you are back to 24 bits. (Unless you save as a 32 or 64 bit float file). Not *exactly* the same.

In ProTools it indeed makes no difference (as everything goes back to 24 bits before being handed over to the next step).

Alistair

[UnderTow]

Originally Posted by Ghostwheel

Just try the following:

Get any 24bit audio source and amplify it up or down by 80db (yes eighty) up or down (it won't clip).

If you save the intermediary step as a 24 bit file, it most certainly will clip. (Unless your peak level is -80 dBfs in which case you have bigger issues )

My guess is, when you did this test, you used an application that keeps the intermediary step in a floating point format.

Alistair

[dkatz42]

Originally Posted by UnderTow

Not entirely. Most DAWs run at 32 or 64 bit float internally. If you normalize (and save) you are back to 24 bits. (Unless you save as a 32 or 64 bit float file). Not *exactly* the same.

Mea Culpa. But the results coming out of your 24 bit fixed point converters into your ears will be exactly the same (and the results being written to a 16 bit CD will be really exactly the same.)

[roosterino]

Originally Posted by UnderTow

Not entirely. Most DAWs run at 32 or 64 bit float internally. If you normalize (and save) you are back to 24 bits. (Unless you save as a 32 or 64 bit float file). Not *exactly* the same.

In ProTools it indeed makes no difference (as everything goes back to 24 bits before being handed over to the next step).

Alistair

This is over my head. Had I skipped the normalization, and just boosted the volume via compressors and eq in mastering, are you saying that's different than boosting the volume through normalizing and then mastering it through the same compressors? Is it the extra step of saving the file? Is it something about the normalizing process? Do I lose something everytime I process and re-save the file (like with mp3's)?

[UnderTow]

Originally Posted by roosterino

This is over my head. Had I skipped the normalization, and just boosted the volume via compressors and eq in mastering, are you saying that's different than boosting the volume through normalizing and then mastering it through the same compressors? Is it the extra step of saving the file? Is it something about the normalizing process? Do I lose something everytime I process and re-save the file (like with mp3's)?

I was just nitpicking on the word "*exactly*" in dkatz42's post. :) There is a mathematical difference but it is extremely unlikely that this is audible.

The bit about saving the file has to do with how a lot of applications work internally. They often use floating point maths. Floating point formats have a mantissa and an exponent. 32 bit float uses a 24 bit mantissa and a 8 bit exponent. You can look at the mantissa as being the actual resolution of your audio and the exponent as being the level.

Thanks to this exponent figure, you can increase or lower the level of a signal by huge amounts (below and above 0 dBfs) without really affecting the resolution of the signal. (If my memory serves me right, 32 bit float gives you 1538 dB of dynamic range).

In the end, for the signal to actually be audible, it needs to go back to a 16 or 24 bit fixed point format for which all the usual rules apply (anything above 0 dBfs clips and anything too low disappears in the (dither) noise floor).

When working in these floating point applications, the intermediary steps will be kept in floating point formats. Hence the comment about saving the intermediary file in 24 bit.

Back on topic: It depends a bit on your set-up but most DAWs have various points where you can boost the level if needed. Also most plug-ins (if you use any) have output level controls which can be used to boost the signal. In light of this, it is unusual to need normalisation to get the signals at the right level.

Alistair

[dkatz42]

Originally Posted by roosterino

This is over my head. Had I skipped the normalization, and just boosted the volume via compressors and eq in mastering, are you saying that's different than boosting the volume through normalizing and then mastering it through the same compressors? Is it the extra step of saving the file? Is it something about the normalizing process? Do I lose something everytime I process and re-save the file (like with mp3's)?

Using compressors and EQ to bring up level is a very different beast (and far more destructive, though hopefully in a good way) than normalizing or any simple gain reduction. If your question was "is it different than bringing up the input gain on my compressor plug" or "is it any different than putting a gainer plug in front of the compressor" the answer would be "no."

The minor quibbling we're having is over whether sticking a gain change in while the numbers are in the floating point realm is any different than normalizing the file (which does it in the fixed point realm.) For all intents and purposes, the answer is "no." While it is true that the floating point representation has much higher dynamic range (making it essentially unclippable) it does *not* add any bit depth; the data part of the representation is still 24 bits.

And given that the point of the question was to raise the level without clipping, the wider range of the floating point representation does exactly nothing for you.

But regardless of all this complexity, the bottom line is this: If you're *boosting* levels with normalizing (which is normally what happens) and you're not clipping (which is easy to control with normalization) then the amount of "damage" is the same as putting a gain plug into the track or boosting the fader--a single multiplication that has no audible impact.

So no, not only is all not lost, but essentially nothing is.

FWIW, my default template in Logic when mixing has a gain plugin inserted into each channel. This lets me bring the levels to where I want them and still get the sweet spot of the fader. The net effect is the same as normalizing each track, but it's faster and impermanent.

[peeder]

Originally Posted by dkatz42

Using compressors and EQ to bring up level is a very different beast (and far more destructive, though hopefully in a good way) than normalizing or any simple gain reduction. If your question was "is it different than bringing up the input gain on my compressor plug" or "is it any different than putting a gainer plug in front of the compressor" the answer would be "no."

The minor quibbling we're having is over whether sticking a gain change in while the numbers are in the floating point realm is any different than normalizing the file (which does it in the fixed point realm.) For all intents and purposes, the answer is "no." While it is true that the floating point representation has much higher dynamic range (making it essentially unclippable) it does *not* add any bit depth; the data part of the representation is still 24 bits.

And given that the point of the question was to raise the level without clipping, the wider range of the floating point representation does exactly nothing for you.

But regardless of all this complexity, the bottom line is this: If you're *boosting* levels with normalizing (which is normally what happens) and you're not clipping (which is easy to control with normalization) then the amount of "damage" is the same as putting a gain plug into the track or boosting the fader--a single multiplication that has no audible impact.

So no, not only is all not lost, but essentially nothing is.

FWIW, my default template in Logic when mixing has a gain plugin inserted into each channel. This lets me bring the levels to where I want them and still get the sweet spot of the fader. The net effect is the same as normalizing each track, but it's faster and impermanent.

False! For one thing, I understand most of these systems actually use a 25 bit mantissa. But you are confusing dynamic range and level/headroom. In the floating point realm, you can alter level without loss of dynamic range. YOU CANNOT DO SO IN FIXED POINT. Nossir, you cannot.

Do I need to piece it apart as to why?

[Ghostwheel]

Originally Posted by peeder

False. If you trim 80db off of 24-bit fixed point audio, save to a 24-bit file, and then add 80db of gain, you have only 144 - 80 = 64db of dynamic range left. Your audio is then only effectively 10.7bit, and you will have a similar noise floor to cassette tape.

You should know that posting in a mastering forum.

First of all, no need for this tone, alright!

So, sure if I save the file as 24 bit *after* the 80db reduction it would destroy the dynamic range. What I meant is that in Audition, it doesn't matter what you start with, you can apply dramatic amplification operations and the internal 32 bit floating point will take care of it for you.

In fact, I just repeated the test and here is what I got (see attachments).

Now back on topic - if you work in Audition, the default saved format is actually 32 bit floating - unless you explicitly tell it to save as 24 bit interger, it won't. So in this case, you are absolutely fine!

Even if you did save as 24 bit - don't sweat over it - I am sure your Normalization operation had absolutely no practical implication about the quality of your audio.

[peeder]

Originally Posted by Ghostwheel

First of all, no need for this tone, alright!

So, sure if I save the file as 24 bit *after* the 80db reduction it would destroy the dynamic range. What I meant is that in Audition, it doesn't matter what you start with, you can apply dramatic amplification operations and the internal 32 bit floating point will take care of it for you.

In fact, I just repeated the test and here is what I got (see attachments).

Now back on topic - if you work in Audition, the default saved format is actually 32 bit floating - unless you explicitly tell it to save as 24 bit interger, it won't. So in this case, you are absolutely fine!

Even if you did save as 24 bit - don't sweat over it - I am sure your Normalization operation had absolutely no practical implication about the quality of your audio.

Well if you knew that, then you would know it was also true for 16 bit, which you said it wasn't. In fact it would be true for 1 bit audio, provided all of your operations (gain and trim) were done within 32-bit floating point with no fixed intermediary.

The correct advice was provided already in the thread before that post.

[Ghostwheel]

Originally Posted by peeder

Well if you knew that, then you would know it was also true for 16 bit, which you said it wasn't. In fact it would be true for 1 bit audio, provided all of your operations (gain and trim) were done within 32-bit floating point with no fixed intermediary.

The correct advice was provided already in the thread before that post.

You continue to attack me despite my trying to clarify my post. Why don't you cut it out and realize that we both are saying the same thing. Gee some people are just looking for confrontation!

[dkatz42]

Originally Posted by peeder

False! For one thing, I understand most of these systems actually use a 25 bit mantissa. But you are confusing dynamic range and level/headroom. In the floating point realm, you can alter level without loss of dynamic range. YOU CANNOT DO SO IN FIXED POINT. Nossir, you cannot.

Do I need to piece it apart as to why?

No you don't, because you are entirely missing my point.

In the *particular case at hand*, namely normalizing a low signal, there is no difference (other than potential roundoff error down 144dB). The reason is that the source signal is already dynamic-range-limited--it's *quiet* after all. Suppose the peaks are at -24dbFS. This means that the top four bits are zero, and there are (in theory with perfect converters) 20 bits of dynamic range. Boosting this signal by any amount that does not clip (which isn't going to happen with normalizing) only moves those 20 bits higher in the word, either explicitly (in the case of fixed point) or implicitly (by changing the exponent in the case of floating point.)

The point is that all 20 bits are preserved in either case, and there is no practical cost (nor, granted, any practical benefit) in having normalized the signal.

I understand perfectly well that floats preserve significant bits over a much wider range of scaling, but that wider range is not at issue here.

You're right about the 25th bit, though once again it has no practical effect.

[peeder]

Right you are.

Originally Posted by dkatz42

No you don't, because you are entirely missing my point.

In the *particular case at hand*, namely normalizing a low signal, there is no difference (other than potential roundoff error down 144dB). The reason is that the source signal is already dynamic-range-limited--it's *quiet* after all. Suppose the peaks are at -24dbFS. This means that the top four bits are zero, and there are (in theory with perfect converters) 20 bits of dynamic range. Boosting this signal by any amount that does not clip (which isn't going to happen with normalizing) only moves those 20 bits higher in the word, either explicitly (in the case of fixed point) or implicitly (by changing the exponent in the case of floating point.)

The point is that all 20 bits are preserved in either case, and there is no practical cost (nor, granted, any practical benefit) in having normalized the signal.

I understand perfectly well that floats preserve significant bits over a much wider range of scaling, but that wider range is not at issue here.

You're right about the 25th bit, though once again it has no practical effect.

Although an important question is what the rendering app does to the replacement bits being shifted in. If it pads them with 0's, this is not a good thing, as there could be quantization distortion that will work its way up from the depths.

[Bob Olhsson]

I think its safe to say the fewer calculations that are performed on digital audio the better the results are likely to be. I've heard some surprising changes caused by different applications' and plug-ins' gain calculations. They are by no means all created equal.

If the gain is going to be changed again, normalizing buys you nothing but additional noise, distortion and more potential for faulty math.

[roosterino]

Originally Posted by Bob Olhsson

I think its safe to say the fewer calculations that are performed on digital audio the better the results are likely to be. I've heard some surprising changes caused by different applications' and plug-ins' gain calculations. They are by no means all created equal.

If the gain is going to be changed again, normalizing buys you nothing but additional noise, distortion and more potential for faulty math.

Interesting.

Working under that assumption, you could argue it makes sense to run a few copies of the master (with the 10 plugins, all doing complex math) and pick the copy with the least faulty math. I wonder if people do this (I wouldn't).

[peeder]

Originally Posted by roosterino

Interesting.

Working under that assumption, you could argue it makes sense to run a few copies of the master (with the 10 plugins, all doing complex math) and pick the copy with the least faulty math. I wonder if people do this (I wouldn't).

It makes sense to 86 any DAW or plugin with bad gain/trim. (and please let us know which ones!)

Most gain/trim will null, math is math.

Some programmers are incompetent of course.

[Bob Olhsson]

It's very foolish to assume programmers get it right. This is why I've always found monitoring is far more critical for digital processing than for analog.

[dkatz42]

Originally Posted by peeder

Right you are.



Although an important question is what the rendering app does to the replacement bits being shifted in. If it pads them with 0's, this is not a good thing, as there could be quantization distortion that will work its way up from the depths.

Well, this takes place at the very beginning of the process when the original fixed-point sample is converted to a float; it will in fact fill with zeroes as it's normalized.

It's not at all clear what *else* you would put down there; you are not going to extract lower-order data like you do with dither and wordlength reduction, because there *is* no lower-order data.

Fact is, if you take a very low gain signal with few bits and boost it, it'll sound crappy and there isn't much that can be done.

If there is going to be some noise treatment to soften this, it's probably best done at the end so as to not accumulate noise in multiple stages. Dither and reduction to 16 bits probably serves the purpose...

[Herb Utsmelz]

So from reading and digesting this discussion, I can choose from these:

A) No one really knows how digital audio works.

B) Record at the highest quality settings possible, and use no dynamics processing whatsoever, for fear of the dreaded dithering problems.


I think I'll choose B, although I suspect that A is also correct.