Normalizing is getting the peak point of a file and then the program raises the entire file to that peak level.

Normalizing can make you lose the dynamic feel of the file being processed so it's not recommended in the long run for files within a multi-track

Stereo mixdowns are ok though but I would recommend a gain change though.

Opus :D

Originally posted by Opus2000:

Normalizing can make you lose the dynamic feel of the file being processed so it's not recommended in the long run for files within a multi-track

How so? The ratios and dB levels between the loudest and softest parts remain the same after the process. (unlike compression/limiting, which is a different story)

Normalizing samples is a good way to combat quantization noise that creeps in in every FX process that's applied to a sample (since the noise is proportional to the value of the smallest bit, and not the amplitude of the waveform, more bits = less quantization noise vs signal per process). This was/is very important in 16-bit recording, but becomes MUCH less of an issue in 24-bit.

I make it a habit to bring up the level of my tracks to about -3dB peak value (normalizing to 50%) before starting to mix, unless the track is already hotter than that, or pretty close. Normalizing (as any gain change process) will add quantization noise of it's own, so if it's not really needed, I just skip it.

Edit: To clarify Opus' post - Normalizing is the process of scanning a file for the largest value of an entire waveform, then scaling the entire file so that that peak with the largest value touches 0dB digital headroom. If you were to play the resulting file at a lower volume, it will still sound identical to the source file. (The process is completely transparent, other than quantization noise)

I have an additional question, which perhaps should be posted in the mastering forum, but since the topic is brought up here....

I mix my songs in cubase, but when I export the mix, I often get a not-so-hot file, i.e. quite quiet. If I'd normalize this file in wavelab to, let's say -3dB, are there any reasons why I should or should not do this?

Rob

BTW, it's good to see that a senior member also makes a habit out of normalizing files before starting to mix.... I was getting worried I was the only one ;)

Maybe the fact that I use -6 dB for all tracks explains why my exports are not that hot. Now that I think of it, I see no reason not to normalize to a higher level... The things one just does without thinking...

Rob

Actually, normalizing before tracking doesn't make much of a difference in sound - A gain change anywhere in the signal chain during actual mixing would give you the same effect... it's just that I'm a stickler for details and I like to tell how loud my tracks are in relation to each other just by glancing at the row of pseudo-faders on my DAW.

Anyway, about the question: I think that you'd be better off just exporting the mix at a hot-enough level, then letting the ME do the rest. A gain-change (as any other FX) will always introduce slight quantization errors comparable to the least significant bit. It may be tiny compared to the signal, but it's there, and hey - why pass up a free lunch when it's given to you? :D

Thanks falkon2, that's about what I figured. So much to learn, and so little time to finish this project....

This thing I'm working on is really my first mixing job ever. I couldn't have imagined the number of issues one should take into account... :) (and also takes more time than I expected...)

Cheers, Rob

what about saving your board/daw fader moves and raising the master fader a little bit more?
I have lost counting the number of times I advised folks not to normalize the final mixes they were suposed to deliver to me.
Things do not sound natural, let the ME do it. The normalization will be a pure no-soul math process.
A nice reference, if you are mixing at 24 bits, use a ceiling of -0.8 db.

That's right - if your final mix is so small that you have difficulty seeing the waveform, you need to mix louder - normalizing won't help. It's just like zooming in on a tiny .gif file won't help you see details - all you get is a pixelated mass.

one beer for the road]
:)

What average RMS and/or what peak level should one be aiming for when mixing an original master?

In other words, what is about optimal to work with for a mastering engineer?

I know that it all depends on music style and the type of song and stuff, but a rough indication?

Thanks, Rob

Rob, I do junior mastering here. I generally end up cutting out lots of messed bottom, but peaking at -0.5dB or less is fine under a 24 bit file.

When mixing: peaks (as in, the entire mix, not a single track) should be as high as you can get without it ever passing clip, but taking additional effects into account that could make the peaks jump a little, -3dB is a pretty safe margin for your loudest peak.

When mixing RMS will fall somewhere between -20dB and -15dB for me typically, though if there aren't that many percussive transients, it can go up to -12dB.

It's not such an issue anymore if you bounce to a 24-bit two-track to send out for mastering though... 24-bits is about 144dB of level between your signal and quantization noise, and practical levels in recording gear rarely go above 90-100dB SNR (If I'm wrong here, let me know!). That means your final bounce can be at like -35dB RMS at 24-bit and still be more or less safe from any ill-effects of quantization.

Well, it seems that it is a good thing that I do use 24 bit files... But still I should try to get my mixes a bit hotter.

The problem is that some of the individual tracks are pretty low in volume, too low at least to crank up using the headroom of the fader, so I am more or less forced to normalize. Like I said, I used to normalize to -6 dB, but -3 dB might be a better idea.

Anyway, falkon2, Alecio, thanks for the help, now I have something to at least aim for when mixing!

Regards, Rob

Rob, do not track too hot also!!!
I myself always try to be around -6.
Seems things sound better, ya still have comps/eqs, subs etc etc to do later

Ya wellcome!

Normalization is (as the word itself indicates) a BAD thing. It makes your music NORMAL which is not good. If you want an exaple listen to anything by Lionel Ritchie. This music is so normal that it just makes me wanna puke :) Why? because it was NORMALIZED! heheheh:)

I was just kidding...for real I'm thinking that normalization was created as a self-destruction feature. It supposed to be pressed when you have enough of listening to the crappy engineering of yours when you can't even set the levels up properly upon recording :) Press it and go to hell hehehe :)

On the second thought...isn't normalization created for musicians so they can feel better after pressing it? :)

practical levels in recording gear rarely go above 90-100dB SNR (If I'm wrong here, let me know!).

Hey falkon , you may want to raise that bar a little bit. Try 90-105db SNR or so now a days. You were'nt wrong , just off a bit. :cool:

[ August 25, 2003, 02:40 PM: Message edited by: dabmeister17 ]

I make it a habit to bring up the level of my tracks to about -3dB peak value (normalizing to 50%) before starting to mix, unless the track is already hotter than that, or pretty close. Normalizing (as any gain change process) will add quantization noise of it's own, so if it's not really needed, I just skip it.

Sir, You realise you are bringing up to a more audible level the quant noise/errors that already exist on your files. As well as putting these digital files thru an unnecessary digital process that will introduce another level of distortion
?
I understand exactly what you are saying in the paragraph I have quoted, and I also suspect that you have miswritten what you meant to say.

Perhaps?

At 24 bits, lets say my signal peaks at about 18 bits (-36dB from 0), and the noise floor from the analog recording equipment is -72dB (noise covers more or less the 6 least significant bits).

Normalizing to 23 bits (about -3dB peak) will introduce quantization noise (or "distortion" as you understand it) roughly on the equivalent to 1 bit. That is akin to adding in a tiny bit of noise that's 36dB even lower than the analog noise floor that's already been recorded.

If my levels are too low on hard disk, I don't normalize. I tweak, then rerecord.

Like I said, normalizing doesn't give any audible benefit over a gain boost in the mixing chain. It's just that I like a neat pseudo-fader panel that will tell me volume levels off the bat, and I use the waveform displays to spot peaks, transients, and such for mixing troubleshooting. Small waveforms = hard to see.

what are you using to measure your mixes with accurate RMS levels?

If I have to, I'd use Cooledit, (it does RMS measurements), but take note I was referring to peak values in my post.

Also, I'd like to make a slight amendment - 1 bit of quantization noise would be *MUCH* lower in ratio compared to signal in my example...

When you normalize, (assuming the program is doing it's work properly) your signal is converted to 32 or 64 bits, gain-changed, then dithered back down to original wordlength (24 in this case), giving you the equivalent of quantization noise along the least significant bit.

Back to the example: From 18 bits to 23 bits used for the signal is roughly 30dB boost (5 bits x 6dB). That means the analog noise goes from covering 6 least significant bits to 11 LSB. Quantization noise introduced during dither covers 1 LSB. That means there is a 10 bit difference, or 60dB difference between analog noise and quantization noise introduced.

Simple test - in 24-bit, take a track with moderate volume, duplicate it, normalize the duplicate and invert it's polarity, then play with the gain on the original track. There'll be a spot where the sound completely disappears. Whatever you hear at this point is quantization noise - at 24-bit, it's practically inaudible.

Again, I'd like to reiterate: In no way do I harbor the fantasy that normalizing somehow makes my tracks "clearer" or "warmer" or "punchier". Neither do I feel the need to normalize just to "maximise the use of bits." I normalize because I'm a visual geek. I've always looked beyond senses and tried to understand how things that work, work. Therefore I rely a LOT on visual input about what should strictly be audio topics. This is NOT a dodge for me, as it is for many people. My eyes don't guide me. My eyes confirm or deny what my ears tell me. Subtle but important difference.