I ran into the whole "normalization" thing 10 or 12 years ago. We were working with a major IT company whose expertise we obtained in exchange for our licensing a bunch of rights to them.
We were in the early days of web audio.
As a quality facility, we supplied materials with the expected/required/proper audio and video levels. Nonetheless, we got our first bill - well, it was a deduction from the credit we had for licensing - and it included "Audio Normalizing."
"We sent you abnormal audio?" I asked, imagining Gene Wilder saying it.
"I don't know," was the response.
"But we were charged for normalizing."
"Well, yes, all audio sent through our group is normalized. That way the levels are constant."
"But what if they're already constant?"
"Doesn't matter. Normalizing ensures it." Then he went into at best a poor explanation of digitization."
That exchange told me they were taking us were back to the wild west again. Like a dorm room carrier current station running unbalanced audio that was all over the place. And, sure enough, that's what their group usually saw. Of course, at that time, a lot of untrained folks bought camcorders and jumped into the business and there lay the problem.
But I hear it now, too - a lot! "Don't worry about the levels; I'll normalize them later."
Makes me reel every time. Also sends me back 40+ years to, "Don't worry about the levels. StaLevel® will catch 'em."
OK. I could say to consider normalization an indictment of sloppy operation. But when I hear someone say they want to normalize, I indict their ears.
Huh?
First, let me say, yes, I know, (oxymoron alert!) "good normalization" involves a more complex algorithm than simple gain adjustment or compression and expansion. But think about two things:
1. Once you get to digital, the mathematics of the process can yield numbers that can't be fully quantified. That is, when the process boosts a section by, say, 2.6dB, a lot of those new numbers don't fall into the exact level of the 16 or 24 bits assigned. Hey, McFly. That's distortion. Might be bad, might be inaudible. But it's there. Some ratios are worse than others. But you can be sure something's there.
2. As smart as normalization might be, it can't (yet) decide how to normalize in certain situations. Dialog, for example - dialog where one line steps on another. Record that with one person about 6 dB below the other. Normalize it and listen. Did a pretty good job, eh? Listen again - to the tails of each line - the parts stepped on. Big change. Did you want that? Normalize a single part or thread and, except for number 1 above, it functions fairly well. Normalize multiple parts already mixed and you get, well, number 2.
Recommendation: just watch the doggoned levels. Get it right in the recording. Your mix will be a lot easier, it'll all be audible, and you won't have to normalize.
OK. I could say to consider normalization an indictment of sloppy operation. But when I hear someone say they want to normalize, I indict their ears.
Huh?
First, let me say, yes, I know, (oxymoron alert!) "good normalization" involves a more complex algorithm than simple gain adjustment or compression and expansion. But think about two things:
1. Once you get to digital, the mathematics of the process can yield numbers that can't be fully quantified. That is, when the process boosts a section by, say, 2.6dB, a lot of those new numbers don't fall into the exact level of the 16 or 24 bits assigned. Hey, McFly. That's distortion. Might be bad, might be inaudible. But it's there. Some ratios are worse than others. But you can be sure something's there.
2. As smart as normalization might be, it can't (yet) decide how to normalize in certain situations. Dialog, for example - dialog where one line steps on another. Record that with one person about 6 dB below the other. Normalize it and listen. Did a pretty good job, eh? Listen again - to the tails of each line - the parts stepped on. Big change. Did you want that? Normalize a single part or thread and, except for number 1 above, it functions fairly well. Normalize multiple parts already mixed and you get, well, number 2.
Recommendation: just watch the doggoned levels. Get it right in the recording. Your mix will be a lot easier, it'll all be audible, and you won't have to normalize.
No comments:
Post a Comment