Subject:RMS vs. Peak Normalization

Posted by: Sound Dog

Date:9/1/2005 6:53:30 PM

Greetings Fellow SoundForgers!

I've been searching for postings about this, and have found surprisingly few--so I thought I'd ask some of you about your experiences.

I have had decent luck with both RMS and Peak Normalization techniques--but "LUCK" is the operative word!

Do any of you have recommendations about when to use which method--or when NOT to? Any favorite settings out there? Types of music (e.g., classical, live, folk, jazz, etc.) where one method yields superior results?

If it's any help, a good deal of my work (but not all) of my work is to MP3--so I do prefer to leave at least minimal headroom.

ALSO: if you have any preferences regarding when in your process to do normalization (e.g., BEFORE or AFTER other editing), I'd appreciate your thoughts on that as well. And if we get some discussion about any of these topics, so much the better.

Your ideas are greatly appreciated; thank you in advance!
Sound Dog
COLORADO, USA

Subject:RE: RMS vs. Peak Normalization

Reply by: ghova

Date:9/6/2005 9:43:34 AM

Our company does most of our sound work using behind-the-scenes automation that I've coded. I have a normalization step in there that's an RMS -20 dB, with dynamic compression upon clipping.

Peak normalization is useless for what we do. If we get one sound file that has a bunch of quiet sounds at -25 and one transient shrill sound at -2, we'll wind up with something too quiet. RMS normalization ensures that everything comes out at a relatively uniform volume.

~Gil

Subject:RE: RMS vs. Peak Normalization

Reply by: Sound Dog

Date:9/6/2005 9:59:47 AM

Thanks Gil for responding! Interesting response, too!

Interesting that you use RMS -20db. I've tried several settings (including having SF scan for max). I've found myself slowly ratcheting down the RMS setting (now I'm down somewhere between -16 and -20db). Have you given up on having SF scan for max? Or is it just that a straight -20db works best in your automation scenario?

What you describe (RMS at -20db) actually sounds like a pretty decent match for a lot of the music I find myself dealing with--such as quieter classical passages. At times, I've ended up with too much headroom across the board. (Yes, sometimes it's necessary to re-record.)

I also find it curious that you find Peak Normalization "useless." I, too, will probably phase out this method. It's great to hear that someone else has reached this conclusion. Thanks again!

Regards,
Sound Dog
Colorado

Subject:RE: RMS vs. Peak Normalization

Reply by: wymondham

Date:9/6/2005 1:03:09 PM

Hi Doug,

I can only speak from my own experiance but I always normalize last and I use Peak Normalizing. I mainly record classical music and it seems fine to me. I set it to as high as I can get to 0dB without overload. I Normalize last because if you do any processing after there is no headroom and you can go over max.

David.H

Subject:RE: RMS vs. Peak Normalization

Reply by: Sound Dog

Date:9/6/2005 1:50:52 PM

Thanks David!

I've tried RMS normalization with classical music, and (as I mentioned) have had generally good luck with it--but sometimes I wonder if it sounds kind of 'fake.' It (RMS) can pump up the quiet passages quite a bit. Probably something the younger generation would prefer--but I still can't decide if I like it all that much.

Any other opinions out there on RMS vs. Peak? Thanks in advance!

Sound Dog
Colorado USA

Subject:RE: RMS vs. Peak Normalization

Reply by: jumbuk

Date:9/6/2005 4:47:46 PM

I normally record in 24-bit floating point, so there is not much point in normalising. I still do it sometimes, just so I can see the waveform! I think theoretically that is a bad thing to do, since any processing can only degrade your recordings.

I use a limiter (Waves L2) on the output as part of the mastering process. That is probably equivalent to RMS normalisation.

One comment: I don't think you can adopt any kind of formula like -20dB RMS and just assume that will produce sensible levels. I think you need to listen in a reproducible environment (same room, moitors etc) to work out what is the right overall level and degree of limiting that makes sense to you.

If you are only recording classical, and only recording the same orchestra/ensemble in the same room all the time, you might get away with a fixed method of normalisation, but my guess is that most of us rarely if ever record under exactly repeatable conditions.

Subject:RE: RMS vs. Peak Normalization

Reply by: Sound Dog

Date:9/7/2005 7:04:14 AM

Hello Jumbuk,

Thanks for your comments. I typically record in good-ol' 44.1K but have toyed with the idea of using some higher sampling rates. Ahhh, one of these days! (But just an aside, why would 24-bit recording obviate normalizing? Also, sorry to say, I'm not familiar with Waves L2.)

Too true--about every process potentially degrading the end result. It's something I struggle with; trying to process the waveform as few times as possible. I do make good use of the "Undo" feature! :)

I also wonder about a "blanket approach" using -20db RMS. I think the previous poster must have considerable automation in place--and he must be secure in the knowledge that his technique will be adequate and clips (if introduced) will be compressed. I, however, am "not there yet" myself. And, as I mentioned, my opinion is that quieter passages can be "OK" in music--and that I've had good luck with RMS but I'm still not 100% sold on it...

Thanks for your comments! Others out there? Your thoughts, please?

Regards,
Sound Dog
Colorado USA

Subject:RE: RMS vs. Peak Normalization

Reply by: wymondham

Date:9/7/2005 1:32:05 PM

Hi SoundForgers,

As I mentioned earlier I record a lot of classical music, this does not rule out a tiny bit of compression and anything else I might think is fitting, when I get into CD Architect a adjust the levels as well wih the volume envelope. So I often cut down audiance applause and bring things up if the volume is low from perhaps a solo singer. This is a sound picture of the event and probably the listener will hear more than if he/she was in the audiance at the time. So although I normaize I suppose I'm playing with other elements in the sound files.

David.H

Subject:RE: RMS vs. Peak Normalization

Reply by: jumbuk

Date:9/7/2005 6:16:25 PM

"Hello Jumbuk,

Thanks for your comments. I typically record in good-ol' 44.1K but have toyed with the idea of using some higher sampling rates. Ahhh, one of these days! (But just an aside, why would 24-bit recording obviate normalizing?)"

I was talking about 24-bit instead of 16-bit, not changing the sampling rate.

Sorry, I think my comment about normalising was a red herring. What I was thinking was that, as 24-bit or 32-bit uses floating-point arithmetic, rather than integers, it doesn't matter what your initial recording level happens to be. With 16-bit integer, you are getting less resolution if you record at low levels.

In fact, there is no justification for normalising in either case, unless you want to avoid having to crank up the analogue component of your system (typically the monitor amp and speakers).

I think (??) what this thread is about is getting the relative levels of different tracks adjusted to a consistent level - "consistent" among themselves (the tracks) and also consistent with other productions on the market.

In either case, I don't think normalisation of any kind is much help. Like someone else suggested, you have to do it manually at the mastering stage, preferably in CDA while compiling your album, and using a couple of reference CDs from a similar genre as a guide to what you are aiming at.

Subject:RE: RMS vs. Peak Normalization

Reply by: wymondham

Date:9/8/2005 11:34:14 AM

Hi Jumbuk,

I'm new to 24 bit recording and am using a new AlesisHD 24 for the recordings. What I'm doing at the moment is sending the tracks via Eithernet to the computer then mix, and down to 2 track. As I now have SF I can play with individul tracks, like compression on a voice track? SF will deal with 24 bit so I can edit the two track as mentioned before in 24 bit not 16. At the end I still have to go to 16 bit CD standard. It's a pity SF lacks behind Wavelab regarding the sound montage feature. AND Surround Sound. Hopefully they are working on it. Regarding higher sample rates, O.K but who will pay you for it? It uses a lot of space on hard disk not to mention all the other complications of high sample rates. There are not many people who will be able to tell the differance anyway I think.
David.H

Subject:RE: RMS vs. Peak Normalization

Reply by: ghova

Date:9/8/2005 3:22:06 PM

Have you given up on having SF scan for max? Or is it just that a straight -20db works best in your automation scenario?

-20db works best for me. I'm at the opposite end of the spectrum from the classical folk; I have a lot of little bits of spoken word from all kinds of sources that need to come out at roughly the same volume. When I say a lot, I mean A LOT. :) Hence the need for automation. The audio is ultimately put in an audio format with lossy compression and an extremely low bit rate, so I'm not as concerned about fidelity as I am about legibility.

What you describe (RMS at -20db) actually sounds like a pretty decent match for a lot of the music I find myself dealing with--such as quieter classical passages. At times, I've ended up with too much headroom across the board. (Yes, sometimes it's necessary to re-record.)

I can understand the argument for peak normalization in classical recording, as RMS normalization might result in overcompression of the loud stuff. I can also understand the argument for no normalization. But at the low bit rates I work with, that's the least of my concerns. :)

~Gil

Subject:RE: RMS vs. Peak Normalization

Reply by: jumbuk

Date:9/8/2005 9:56:38 PM

"Regarding higher sample rates, O.K but who will pay you for it? It uses a lot of space on hard disk not to mention all the other complications of high sample rates. There are not many people who will be able to tell the differance anyway I think."

I don't know where this came from - I certainly wasn't advocating higher sample rates. I use 24-bit 44.1 KHz.

Subject:RE: RMS vs. Peak Normalization

Reply by: wymondham

Date:9/12/2005 12:42:08 PM

Sorry Jumbuk.,

It was Sound Dog that mentioned it.

Quote:I typically record in good-ol' 44.1K but have toyed with the idea of using some higher sampling rates.

David.H