Teletronix LA-2A or similar vs. RMS normalization

[Ben B]

RMS normalization has nothing to do with compression or limiting. It will simply change the overall gain of the sound file.

[SmoothVibe]

Originally Posted by Ben B

RMS normalization has nothing to do with compression or limiting. It will simply change the overall gain of the sound file.

OK, I'm using the native RMS normalization in Pro Tools 9 and I don't know exactly what that does, but you are describing what to me sounds like peak normalization. Please note I'm trying to understand what gives the most musically pleasant outcome between the two methods RMS normalization vs. hardware limiting with a device such as Teletronix LA-2A, as part of introducing that musical quality through loudness increase.

[drbob1]

They are very different if I'm understanding what the software is doing. It's moving the RMS level as close to the 0 dBFS as it can without any overs? A limiter on the other hand, is a special case of a compressor. It reduces dynamic range by primarily flattening any loud notes that go over the threshold, using a much more aggressive compression ratio than one might use for more gentle compression of more of the mix. As such, it also tends to impart some character or harmonic distortion due to the hardware topology which some people feel helps to "glue" the mix together. As to which is preferable, that's entirely up to your ears. Might be fun to try a limiter on a mix if you haven't done it before (or even just a plug-in to get the idea).

[Thuneau]

RMS can not be close to 0 dBFS* as these are measurements of two different properties. dBFS is a peak value at any given sample, RMS is a statistical product of many samples- usually 50msec worth, but the window can be defined to be longer or shorter.
You take a bunch of samples, square their values, find the average (mean) value of those, then compute the root of that average.

I remember that Sound Forge had an RMS normalizing algorithm, which actually messed with the dynamics of the audio, but I don't know the particulars of the algorithm. It did not sound as good as using compression and peak limiting to get RMS values of a piece of audio to a certain value.

* perfectly square wave at 0dBFS would result in 0dB RMS, but no one sane would normalize their music in that manner.

[Ben B]

Normalizing a sound file just changes the overall gain of the sound file. Whether I use peaks or RMS is immaterial. Suppose you have a song with an overall RMS of -31 dB FS, and you use RMS normalization to normalize it to -20 dB FS RMS, the gain of the sound file will be increased by 9dB overall. Now suppose I had a song peaking at -9dB FS, and I peak normalize it to zero, the exact same thing happens - the gain of the sound file is increased by 9 dB.

[Ben B]

Normalizing is NOT a form of dynamics processing.

[Thuneau]

Originally Posted by Ben B

Normalizing a sound file just changes the overall gain of the sound file. Whether I use peaks or RMS is immaterial. Suppose you have a song with an overall RMS of -31 dB FS, and you use RMS normalization to normalize it to -20 dB FS RMS, the gain of the sound file will be increased by 9dB overall. Now suppose I had a song peaking at -9dB FS, and I peak normalize it to zero, the exact same thing happens - the gain of the sound file is increased by 9 dB.

no such thing as "-20dB FS RMS" or "RMS of -31 dBFS" . It's either one or the other. They are related but not the same.

[Thuneau]

Originally Posted by Ben B

Normalizing is NOT a form of dynamics processing.

Peak normalizing is not.
But there are (were?) programs that will normalize RMS to a given value and apply some sort of peak limiting to prevent clipping. Sound Forge used to do it and it was called normalizing. Maybe it still does it- I don't use it so I can't be sure.

[SmoothVibe]

Thanks for your thoughts and comments! I tried to find some information about the RMS normalization feature in Pro Tools 9 but without luck. But I have a theory of what it could mean. The average RMS normalization I suspect performs a normalization process based on an average RMS point of view instead of from a peak point of view. An example, the peak of a song is at -X dBFS, a 100% normalization of that would normalize the song up to -0 dBFS, so let's say X is 2, then it gains the signal by ~2 dB so that the peak will equal -0 dBFS. That should be a pure multiplication of all samples, where the mutiplication factor per sample is constant. But when normalizing to adjust the song's average RMS from the current to the target average RMS level it's up to the algorithm to decide how to do this, which of the song's samples to target and by what factor. I see this as a kind of dynamic and auto-adjusting limiter that I guess could actually be made to increase the perceived loudness more efficiently than a normal static/manually configured limiter... So theoretically it seems like a peak limiter on steroids, it should be a mastering engineer's trick of making any song be able to reach a certain loudness at least possible musical impact, by simply having a look at the song as a whole and then make the decisions, instead of like a normal limiter that simply just does something to whatever comes its way... My ears like the sound of RMS normalization as a way of increasing loudness as opposed to normal peak limiting, especially when going for something extremely loud... This also seems to raise an interesting question? Will a more dynamic input to an RMS normalization process actually give it better performance by giving it more to work with?? That I'll try out by ear tomorrow. I'll create an extremely dynamic mix and then I'll make that extremely loud with RMS normalization, I mean up to a certain average RMS dBFS level. Then I'll peak limit up to the same average RMS dBFS level and do a blind test...

[stinkyfingers]

Originally Posted by Thuneau

dBFS is a peak value at any given sample, RMS is a statistical product of many samples-

dBFS is the unit of measure...both Peak and RMS can be written as dBFS...

edit: i see some are still going on about this, so here's the technical definitions:

http://www.rane.com/par-d.html#decibel

[drbob1]

Smooth, normalization doesn't work by multiplying sample value by a set number, it works by adding. So you're increasing all samples by a set amount. It's certainly possible that RMS normalization is more complex, involving some combination of increasing overall volume and cutting peaks, but that sound perilously close to general compression/limiting using a slightly different algorithm?

Easiest way to be sure, I guess, would be to examine wave forms and see how they're changed by the three processes (compression/limiting, peak normalization, RMS normalization)

[psycho_monkey]

Normalisation, rms or peak, is just a gain change. Limiting is a form of compression.

They're in no way the same thing, and rms normalization is not peak limiting!

[peat]

once you truly understand peak, rms, crest factor, dynamic range and loudness, you'll be set.

you'll have a lot of 'ahhhh' moments

[SmoothVibe]

Originally Posted by psycho_monkey

Normalisation, rms or peak, is just a gain change. Limiting is a form of compression.

They're in no way the same thing, and rms normalization is not peak limiting!

If you read my post you'll notice I also said it was up to the coders of the algorithm to decide how to bring up the song's average RMS when applying RMS normalization. I mentioned auto-adjusting limiters as a way of trying to picture a kind of process where some parameters are non-fixed and where the result is higher RMS, I know it might not be the same thing, but I'm interested in what it is, to my knowledge it's not a constant gain increase...

[SmoothVibe]

Originally Posted by peat

once you truly understand peak, rms, crest factor, dynamic range and loudness, you'll be set.

you'll have a lot of 'ahhhh' moments

Please tell us all how the RMS normalization algorithm in Pro Tools 9 truly works...

[psycho_monkey]

Originally Posted by SmoothVibe

If you read my post you'll notice I also said it was up to the coders of the algorithm to decide how to bring up the song's average RMS when applying RMS normalization. I mentioned auto-adjusting limiters as a way of trying to picture a kind of process where some parameters are non-fixed and where the result is higher RMS, I know it might not be the same thing, but I'm interested in what it is, to my knowledge it's not a constant gain increase...

Yes, it IS a constant gain increase. Otherwise it's not normalisation.

RMS - Root mean square - Wikipedia, the free encyclopedia - for a definition.

If you select a wave and choose peak normalisation, you'll change the waveform so that the highest peak is the value you chose, and everything else will still be the same volume in relation to that. eg if you sound file has a peak of -5 and you choose to normalise to -1, you'll turn everything up by 4dB. No compression will occur.

If you chose RMS normalisation, the processing will analyse the waveform (see formula on the wiki page - my maths is a bit suspect these days, but essentially it measures at regular intervals), find the RMS value, and then gain up or gain down the wave form so the RMS value is the same as the normalisation value selected.

Again, it's one gain process for the whole wave - the only thing that's different is the amount of gain change. normalising to -3dB RMS is close impossible with a continuous music signal - it'll be clipping like crazy. Even heavily limited masters have RMS levels of -12dB.

Does that make a bit more sense? to try to compare it in any way to peak limiting is wrong. If I do RMS normalisation to a file, I could then re-analyse it, find my new peak, and change the gain of the original file to match this, and the result will be EXACTLY THE SAME.

In reality, unless you're doing broadcast, forget about normalisation and RMS levels. Just mix so it sounds good, and master so it competes sonically and loudness-wise. I can't remember the last time I had to use the normalisation process.

[Ben B]

Originally Posted by Thuneau

no such thing as "-20dB FS RMS" or "RMS of -31 dBFS" . It's either one or the other. They are related but not the same.

Any time you are using decibels, a reference must be established. Since we're discussing levels in a digital system the reference is dB FS whether we wish to assess peak levels or RMS levels. I stand by my use of the terms.

[Thuneau]

Originally Posted by Ben B

Any time you are using decibels, a reference must be established. Since we're discussing levels in a digital system the reference is dB FS whether we wish to assess peak levels or RMS levels. I stand by my use of the terms.

Yes, I understand decibels pretty well. In digital audio dBFS refers universally to peak values. RMS values are never written out as dB RMS FS. They are negative and obviously referenced to maximum achievable, at a given resolution. They are written simply as XdB RMS.
The point is that anything below 0 dBFS is a safe level when talking about peaks, while that can not be said about dB RMS. Measurement expressed in dB RMS can be at -20 and clipping or at -6 and not clipping. Adding the FS after RMS serves no purpose and it only adds to confusion and misunderstanding.

[Ben B]

I guess we'll have to agree to disagree on that. I see dBfs RMS used in technical literature on a regular basis. Here' a link to some Genelec literature that uses the term just as I do (e.g. Under the "What Reference Level" heading).http://www.genelecusa.com/faq/multic...igning-levels/

[5down1up]

Originally Posted by Ben B

I guess we'll have to agree to disagree on that. I see dBfs RMS used in technical literature on a regular basis.

keep it simple :

db = describtion
fs = fullscale ( digital 0.0 max )
rms = average
peak = max

it always sounds like rocket science. its just some words !

[psycho_monkey]

Originally Posted by Ben B

I guess we'll have to agree to disagree on that. I see dBfs RMS used in technical literature on a regular basis. Here' a link to some Genelec literature that uses the term just as I do (e.g. Under the "What Reference Level" heading).Genelec Inc. - How do I align the levels of a 5.1 system (using pink noise as a test signal)?

I don't know the exact best way to write it, but you're correct - dB RMS is meaningless. This could be dBFs (RMS) or dB SPL (RMS), dBV or dBu...I think bracketing it might be correct actually.

But you definitely need to mention what it refers to somewhere. Otherwise it's just a gain reference.

[Bill@WelcomeHome]

Originally Posted by SmoothVibe

In another thread we are currently discussing to normalize or not to normalize. I do RMS normalization instead of explicit master bus plug-in limiting, because my ears like that better. Has anyone here compared RMS normalization against using a hardware limiter/compressor such as the Teletronix LA-2A on the master bus? Am I missing out on something by not investing in a hardware limiter/compressor?

Most of these oft-referenced analog pieces impart their own sound to whatever is pushed through them. If you want to add another effect to your mix after you thought that you were done, run it through another box.

One thing to consider: analog boxes cannot read ahead.

[Ben B]

Originally Posted by psycho_monkey

I don't know the exact best way to write it, but you're correct - dB RMS is meaningless. This could be dBFs (RMS) or dB SPL (RMS), dBV or dBu...I think bracketing it might be correct actually.

But you definitely need to mention what it refers to somewhere. Otherwise it's just a gain reference.

Usually, the RMS is in subscript, but I didn't know how to do that on GS!

[foldback]

The problem here occurs when a manufacturer (like Avid) takes an industry standard term (like normalization) and changes it into a different product.

I don't have Protools 9 but I understand what you're saying and it sounds to me like the program is doing some interpolation for you. Personally I hate programs that rearrange the 1's and 0's without me understanding what is going on but if you like the results then use it.

Traditionally normalizing has been just a level change that sought to optimize the dynamic range (I rarely normalize).

A Teletronix LA-2a is not a great bus compressor, I say this from experience not GS regurgitation technique. There are plenty of threads here to explore everyones take on bus compression.

The LA-2a IS a great processor. I do know that it was designed as a Leveling Amplifier but that was in the 1960's and was targeted towards broadcast radio at the time. I would not want one of my mixes from today to be slathered with LA-2a, it's just not the right piece for that app today IMO.

An LA-2a adds girth to a sound that passes through it, even if it is not compressing, this is one of the LA-2a best attributes. There are lots of tube devices out that try to add girth or imply they do "tube" magic, the LA-2a actually delivers that elusive thick tube sound because it is a bunch of tubes and transformers. That sound is great for some things but not so great for others. I like it on my voice, I hate it on my partners voice. I like it on bass guitar, I don't like it on electric guitar.

To begin your quest for great bus compressor do a search for the API 2500 and off you go, enjoy the quest and good music to you!

Originally Posted by SmoothVibe

If you read my post you'll notice I also said it was up to the coders of the algorithm to decide how to bring up the song's average RMS when applying RMS normalization. I mentioned auto-adjusting limiters as a way of trying to picture a kind of process where some parameters are non-fixed and where the result is higher RMS, I know it might not be the same thing, but I'm interested in what it is, to my knowledge it's not a constant gain increase...

[SmoothVibe]

Originally Posted by psycho_monkey

In reality, unless you're doing broadcast, forget about normalisation and RMS levels. Just mix so it sounds good, and master so it competes sonically and loudness-wise. I can't remember the last time I had to use the normalisation process.

Thanks! I'm just curious, have you tested the RMS normalization process in Pro Tools 9, did you not like it sonically or do you find it more efficient to do the same with some gear instead that also adds some colors to the mix?

[psycho_monkey]

Originally Posted by SmoothVibe

Thanks! I'm just curious, have you tested the RMS normalization process in Pro Tools 9, did you not like it sonically or do you find it more efficient to do the same with some gear instead that also adds some colors to the mix?

Normalisation in PT does NOTHING sonically - it just makes things louder or quieter. You can test this for yourself. It's just a gain change.

If I wanted to add colour - you can't use normalisation for that. If you want to add colour, you have to use something that adds colour!

[psycho_monkey]

Originally Posted by foldback

The problem here occurs when a manufacturer (like Avid) takes an industry standard term (like normalization) and changes it into a different product.

I don't have Protools 9 but I understand what you're saying and it sounds to me like the program is doing some interpolation for you. Personally I hate programs that rearrange the 1's and 0's without me understanding what is going on but if you like the results then use it.

Traditionally normalizing has been just a level change that sought to optimize the dynamic range (I rarely normalize).

That's all PT does with it's normalisation function. OP is just confused.

[SmoothVibe]

Originally Posted by psycho_monkey

Normalisation in PT does NOTHING sonically - it just makes things louder or quieter. You can test this for yourself. It's just a gain change.

If I wanted to add colour - you can't use normalisation for that. If you want to add colour, you have to use something that adds colour!

Yes I know, I was more referring to the opposite - that maybe you prefer a coloration which is why you sonically don't like RMS normalization, because you get a coloration by limiting, which might be more sonically preferrable to you?

[SmoothVibe]

Originally Posted by foldback

I don't have Protools 9 but I understand what you're saying and it sounds to me like the program is doing some interpolation for you. Personally I hate programs that rearrange the 1's and 0's without me understanding what is going on but if you like the results then use it.

Traditionally normalizing has been just a level change that sought to optimize the dynamic range (I rarely normalize).

Yes I think it does more than just a gain change, whatever it does I find more efficient than doing the same with a limiter.