I've recently done some testing to find out which technical loudness measure best reflects the difference in loudness as perceived by ear. In other words, I was looking for the "perfect algorithm" that could be used by a radio station or iTunes to adjust song volume.

Basically, I imported a few songs into a Cubase session and adjusted their volume the way I would if I was making a compilation CD with them. I also took various loudness measurements (RMS, LUFS) to see how an algorithm using that particular measure would adjust volume.

I found that LUFS integrated loudness did come closer to my ears than using the average RMS value of a song, but not by much.

Second best was the LUFS short term loudness value (which measures loudness in the loudest 3 seconds). But still not quite there.

The best measure was actually the RMS value of the loudest part of a song (at least 20 seconds). So I am asking myself: Why are radio stations (not sure about iTunes) using LUFS values to adjust song volume rather than an algorithm that measures RMS of the loudest 20 seconds in a song?

I'd be interested to hear opinions on this.

I felt that LUFS especially fell short with songs that have very quiet passages. Suppose you have two versions of the same rock song. One version has a long and quiet intro. If you were to adjust playback volume between them you would naturally only take into account the loud part, i.e. you would play them back at the same volume. Not so with LUFS. The version with the quiet intro would yield a lower LUFS value and would therefore be played back louder by a LUFS algorithm. Isn't that a flaw in this loudness measure?

LUFS integrated on the loudest sections always sounds closest to me, it's not perfect, but better than RMS or my pet peeve, "DR".

LUFS is the closest when comparing on a full range system in a flat room.

That said it can still be thrown off by excessive sub bass down real low. Think round 20hz

It's long but read the parts about gating.

https://www.tcelectronic.com/media/1...undt013011.pdf

This is the one thing I didn't measure. But maybe it's less important whether you use RMS or LUFS and more important whether you compare the whole song or just the loudest part.

The main difference between the basic LUFS case and RMS is nothing else than a well specified freq weighting. It consists of a high pass + a little boost at HF (so called k-weighting) before the RMS calculation.

Most ppl using RMS meters also use some form of freq weighting. a, b, k, whatever. I can't think of many case of audio RMS measurements without at least one highpass somewhere. This is already LUFS.

Another relevant difference comes into play in certain longer term EBU modes (Short Term, Integrated and Loudness Range), they use special mechanisms to ignore program pauses otherwise distorting the RMS value. For music production, this "freezing" often makes no sense and the "Momentary" mode is the main choice. This mode is a simple k weighted RMS.

The public broadcasters already measure in LUFS. There the R128 norm is valid.

For digital radio this norm also comes to play.

https://tech.ebu.ch/docs/r/r128-2014.pdf

LUFS is a scale. Loudness Units (relative to) Full Scale. Exactly like dBFS, or Decibels (relative to) Full Scale.

LUFS has no specified frequency weighting, exactly as dBFS has no specified frequency weighting. It's a metering scale.

Fabien, would you write "The main difference between the basic dBFS case and RMS is nothing else than a well specified freq weighting"?

I would hope not, because that makes no sense.


LUFS has no K-weighting and has no RMS calculation. Again, it is merely a scale, just like dBFS. -18LUFS is the same as -18dBFS without K-weighting.

Insert a 1kHz tone aligned to -18dBFS and look at the level on a LUFS meter: -18LUFS. No K-weighting used in that measurement, no weighting filter.


While Sound Pressure Level meters may use A, B and C weighting, RMS metering does not have a weighting filter.

RMS only gives electrical information about a signal, and contrary to popular belief does not actually measure perceived Loudness. RMS as a measure of Loudness is about as useful as using dBFS to measure the loudness of a signal.

RMS is not the same as EBU R128.

No pauses are ignored in the measurement, but well a -70LUFS threshold is employed, and certain transients are ignored such as gunshots and loud crashes that may skew the measurement. The LFE is also ignored but this forms part of the frequency weighting, where if included the LFE would require +10dB of compensation. There is also channel weighting in the measurement which means 5.1 metering is weighted differently to NHK22.2 metering.

I'm surprised that you still find understanding the basic concepts of EBU R128 so difficult, while seemingly being more than capable of understanding advanced math.

The OP asked for the difference between RMS and LUFS, so I tried to answer from his point of view. Of course he meant RMS dB FS vs LUFS and asked the wrong question. But I don't care, I assumed to understand him.

BTW, note that I said "audio RMS measurements". They really make little sense without a HP filter. Everybody who's used a compressor in the past is familiar with the fact the pure, unfiltered RMS doesn't correlate much with loudness.

EBU r128, the standard that introduced the concept of LU and LUFS, describes 3 different approaches to measure loudness. All k weighted and RMS based. There's no LU without k-weighting and RMS. I wonder how else it is done in your opinion? Note that the EBU documents don't mention RMS, but they undeniably do it in their algorithm with the square root operation "hidden" as a division of the dB signal (in the dB scale, a division by 2 is equivalent to the square root in the linear domain).

The three modes described, Momentary, Short Term and Integrated all base on the above k-weighted RMS measurement. The last two add additional gating (what I called freezing) handling absolute low level passages and relative (sudden) high or low level passages. These are transients. Program pauses. Silent or near silent regions. I don't understand your criticism. When any gate engages, the resulting value effectively freezes (= the meter isn't affected by that "invalid" event, so it doesn't move).

My main statement here was that in case of music production not needing any forms of program pause, mega dynamic jumps or exceptional events handling, "ebu mode Momentary" is the right choice. This is nothing else than a k weighted RMS measurement!


LU momentary = dB RMS k-weighted

They are equal. None is better.

Only the broadcast specific Short Term and Integrated modes use gating. And even then, they are optional and inconsistently defined (different relative thresholds depending on your community/branch/location). You don't need that stuff for music production, you typically use Short Term and Integrated to control a program/stream over long periods. The latter (Integrated loudness) is even invalid for short samples, as stated in the main document.

(Have a running ebu meter here, including the loudness range extension, fulfilling all EBU tests. I assume to have understood )

lufs short term is for music IMHO better than "simple rms"
the messurement-approach replay gain use for messure the loudness works for music(streaming) a bit better, because the louder songs (uptempo, rock, agressive music) still stay a bit louder than ballads.
lufs didn't regard >150Hz enough for music, it's was developed mainly for broadcast

It seems that the most important difference between the two has been missed by everyone.

The basic problem with RMS is that it is that the "M" part of it is not standardized. You are free to select any averaging time constant you want, which of course directly impacts the value are reading. With LUFS based measurements, like BS-1770 which seems to have become the norm over here, the time constants and the integration techniques are completely specified. So you can at least compare measurements. With "RMS" you haven't a clue as to what you are seeing or are comparing.

If you want to make a -real- "loudness" measurement, then go get a copy of Orban's free loudness meter and start using the CBS algorithm. It is more psychoacoustically correct then any of the LUFS based techniques (as he has lectured on in detail).

Again you fail to understand the basic terms.

Loudness Units are like deciBels. 1LU = 1dB.

Would you write "There's no dB without k-weighting and RMS"?

No, you wouldn't. That doesn't make sense.

There is no such thing as LU momentary, anymore than dB Momentary is a concept. It just doesn't make sense.

Momentary Loudness is a calculation, the result of which is expressed in Loudness Units, much like Sound Pressure Level is expressed in deciBels.

The dB is useless without a reference, in the same way LU is meaningless without a reference.

?

The OP is asking for the difference between a meter showing RMS dB and a meter showing whatever LU value.

What is the problem in understanding that I meant

"the LU value resulting from a momentary loudness measurement vs the dB value resulting from a k-weighted RMS measurement"

I see absolutely nothing wrong with it. This is a perfectly clear description.

I'm trying to show some relations and interesting facts, explaining that the difference between both, given a frequency weighted RMS, can be zero, or very close to it. Both technically and regarding the result.

Asking for what is "better" is weird, as they are absolutely equal over huge regions.

Maybe the wording convention police prefers this: The Momentary Loudness value as defined by EBU r128 is a conventional k weighted RMS value. They are exactly the same thing, just labeled differently.

EBU r128 is no rocket science. It's a rather small extension of the conventional k weighted RMS measurement. It's slightly overloaded with redundant and mostly irrelevant terminology.

That is much clearer and defines what you are actually discussing without just randomly using terms.

Except, that statement is incorrect.

The EBU R128 standard uses Mean Square in its measuring algorithms and not the Square Root of Mean Square (square root of the average or arithmetic mean of all the squares of values).

So no, EBU R128 is not RMS just labelled differently.


Firstly, RMS is not K weighted. EBU R128's K weighting is not conventional at all, as it is a framework of elements that in this combination are unique to this particular standard. All other K weighting systems that may share similar naming convention (like Bob Katz's for example) are in fact different.

Have to agree with the OP's point about tracks with more quiet passages giving lower LUFS readings. I've noticed a lot of dance music producers using this to their advantage to fight the "new loudness war" which is about fooling LUFS loudness normalisation on streaming platforms. Long intros and outros and very long dynamic breaks are giving them up to a 2dB advantage over tracks with more dense arrangements.

I think a lot more of the program's quietest material needs to be trashed before the final LUFS figure is calculated.

Cheers

Conundra

Pretty sure there's gating involved in the LKFS/LUFS meter to weed some or most of the quieter stuff.
I'd be a little surprised if people compose music to purposely trick meters, and think maybe the quieter sections are
built into the style of music to create the dynamic to set up the drop, which seems to cater to loud anyways.

The gating function in the LUFS group (R128, BS-1770) is a band-aid fix to make them barely acceptable. Everyone has probably "forgotten" that those algorithms were failing rather badly when they were initially released without the gating in there. The CBS technique doesn't include gating because it responds to audible content in a way that more closely resembles loudness growth and summation. There is no gating in our hearing, which would lead one to wonder why you would need to have it the measuring equipment.

LUFS techniques sorta-kinda (coulda-woulda-shouda) resemble loudness perception, but then they don't. You use them at your own peril. I pointed this all out 5 years ago when I first started posting on GS. The CBS technique was developed 50 years ago and has stood the test of time. It should have been selected as the loudness measuring standard instead of this rms concoction, but the fact that it wasn't shows you what happens when politics and ego get involved with so-called "scientific" endeavor. Now everyone is paying the price.

That's what gating is for. Quiet sections don't factor it to the assigned value.

The gating removes the top and bottom dynamics yes, but if there are sufficient quiet sections remaining, then the overall LUFS will be lower. I've seen it time and again and it is very easily demonstrated. Simply take a track and copy and paste the intro or quiet passage a few times to extend the track and take LUFS readings of both versions.

Also, don't underestimate the shrewdness of artists that are knowledgable and willing enough to use this to give them an edge online, which is where most music is discovered and promoted after all.

Cheers

Conundra

Seeing what Conundra describes quite a bit as well.

FWIW I think at least some of it is kids/up-and-comers just unknowingly copying the arrangements of artists who have done this deliberately. So, the function is starting to dictate the form in those styles. But it's a noticeable trend, whatever the mechanisms.

To expand on some of the root-mean-square (RMS) vs mean-square (MS) discussion above:

MS is in dimensions of power when you are measuring audio voltage levels. So the dB equivalent of a mean-squared voltage ratio v^2, is 10*log(v^2).

RMS is in dimensions of voltage, because when you take the square root of squared voltage ratio v^2, you get back to volts. The dB equivalent of voltage ratio v is 20*log(v).

Mathematically, log(v^2) is equal to 2*log(v). So if you recalculate 10*log(v^2) by substitution, you get 10*2*log(v), which of course is 20*log(v), bringing us full circle to dB in voltage. So you should be able to see from these identities that RMS and MS give us the same dB readings.

If I were to place two dB meters side-by-side, one reading dB voltage and the other dB power, you wouldn't be able to tell the difference just by looking at them. They both would be giving the same level readings in dB. So it is basically a moot discussion to claim that MS is different from RMS when you are looking at loudness calculations in terms of dB scales, because you get the same dB readings either way.

Class dismissed.

The measurement system was developed for broadcast where it makes sense. It is now being adopted for music streaming and unfortunately it falls a short in the way you describe. A question of the right tool for the wrong purpose it seems.

Also, according to the full spec, there is a Loudness Range measurement that has to be within 8 LUFS for broadcast. I wonder if these tracks fulfil that requirement. If they don't, then the tracks are not fulfilling broadcast specs and the total LUFS value is kind of meaningless.

Alistair

I found this interesting, did some looking about and found some interesting posts by Robert Orban (here for example https://hydrogenaud.io/index.php/topic,62570.0.html ). Thank you.

Is that really the only meter around with the CBS algorithm though? I might mail Dave Gamble and see if they can implement it in DMG meter products...

If the LUFS was calculated from the same section of track that label's specify as the "preview" to their aggregators, the playing field would be somewhat more level.

what's cool about rms is the possibility of adjusting the time constant to fit the programme. no single number will ever be meaningful so the ability to choose the timing window is a good thing. people that take rms level as gospel should realise that the timing affects the reading by as much as 5 db and more.

macc - Orban has the only implementation of the CBS loudness algorithm that I'm aware of. It's also included as part of his Optimod processing. CBS Laboratories made a hardware version for broadcast station use a long time ago. If you talk to Dave G. and he is interested, he might be able to work out some sort of deal with Orban to use his implementation.

Timesaver - adjust the time constant to 35 msec and leave it there. That will give the closet approximation to the loudness risetime of your hearing. If you want to smooth it out some more, then follow it with a second time constant of your choosing.

So which RMS meter do you people use that can adjust the time and read average + max that works good?

Reynaud, I believe that you are wrong on this, while FabienTDR is correct. You can only use LUFS (or LKFS) for loudness values computed according to the BS.1770 standard. It can be either Short-term, Momentary, or Integrated loudness. They all include K-weighting, plus different sorts of gating (absolute or relative). Here's the definition of LKFS directly from BS.1770:

It states that LKFS is indeed a scale, but it implies K-weighting and a certain calculation formula.


There is definitely K-weighting going on in any LUFS meter. Try a different frequency and you'll get a different reading.


Pauses are indeed ignored for most of loudness measurements. Two most typical loudness measures are Integrated Loudness and Max Short-Term Loudness. They both ignore pauses: the first one uses a -10 LU relative gate, while the second one just takes the max of all measurements.

You need to be careful when quoting LUFS, because LUFS is not a loudness measure by itself. As Reynaud has pointed out, it's a scale. You are probably referring to Integrated Loudness, which is indeed measured in LUFS. This measure includes a certain amount of gating that mostly discards quieter parts of the song from calculation. However the frequency weighting employed (the K-weighting) is not super-accurate for estimation of perceived loudness. It's better than no weighting at all, but otherwise it's pretty simple. If you are comparing songs of similar spectral content, you may find that RMS of the loudest part works just as good or better. However Integrated Loudness is designed to take the whole programme into account, not just the loudest part. So, its objective may slightly differ from yours.

Thank you for joining the discussion Alexey.

If you simply run an 1kHz and 2kHz alignment sine tone, both referenced to -18dBFS through an EBU R128 compliant meter, that -18LUFS reading is weighted? The pre-filtering and gating is applied?