Did anyone else find this http://www.soundonsound.com/sos/sep11/articles/loudness.htm as disturbing as I did? I feel there are several distortions and misrepresentations in this article which may reflect badly on Sound On Sound's reputation for accuracy. The main article appears to be accurate enough and well researched, (though I'll flatly disagree with some of the author's conclusions without apology, as I think the methodology unsound) but the sidebars contain some real headscratchers.

Quote:

---

"Limiters reduce loudness ranges, don’t they? Well, yes — and no. In fact, this issue is much more complex than it seems. Imagine you’ve got an audio file that is normalised: you can’t add any more gain without getting distortion. Using a limiter or a compressor on such a file will nevertheless add gain to its content: the RMS levels will be increased. This adds dynamic range to the medium: instead of being, in the case of a 16-bit file, 96dB, it will increase to perhaps 100 or 105 dB. On the diagram to the right, this additional available dynamic range is illustrated by the grey rectangle. From that point of view, limiters don’t decrease the loudness range, they increase it."

---

Firstly the author is referring to level not gain; the two are not interchangeable. Second, there's no distinction between clipping distortion of the kind commonly exhibited by a limiter pushed too hard and distortion caused by digital overs. Perhaps he should have simply said "overs". Next, the phrase, "adds dynamic range to the medium" cannot possibly be true, as digital PCM audio encoding has a fixed dynamic range based on bit depth and sample rate that no limiter can affect. By the author's own nomenclature, dynamic range and loudness range are not the same thing. He appears to be trying to imply that a limiter can increase the loudness range of a normalized audio file, when this is patently not the case. (technically, dynamic range increases, as the waveform gets taller, on average, but only in the sense of the difference between RMS average and digital black, which the author went out of his way in the main article to note was an incorrect usage of the term, which should properly be reserved for the absolute dynamic range of the digital system as I described, above) Furthermore, the decibels in this paragraph are not referenced to anything; this makes them essentially meaningless, though I presume dBSPL is what's meant, in context. Is this author trying to pull a fast one, or am I completely misreading this?

Edited by Jennifer Jones (30/09/11 10:45 AM)

Sounds like he's mixing up perceived values and actual values. I can see how a limited file can give the perception of greater dynamic range - but it's at the expense of clipping and distortion.

Just read it - laughably poor logic. Good grief SOS - what are you letting in these days!!!

Measuring dynamic range with variations in RMS - that's an idiotic argument in musical terms. The whole article suffers from confirmation bias. I'd love to have a chat with the author and understand why he has offered such flawed reasoning.

Edited by Jennifer Jones (30/09/11 11:01 AM)

That was a point I found quite hard to understand while editing the article, and it's entirely possibly I've introduced some confusions in the process of editing! At the time I asked Emmanuel to clarify it and he sent me the following explanation -- I hope he won't mind me reproducing it here:

Suppose you've got a medium whose dynamic range is 50dB.
Suppose you're recording a signal on it, for instance vocals from a mike.

50dB is not much. Let's say the singer is yelling at some point of the recording: you've got to set a low input gain to avoid distortion. But then the singer whispers: this whisper is lost in the medium's background noise.
Conversely, you can increase the input gain to get a clear recording of the whispers, but then, when the singer's yelling, the medium generates distortion.

Now insert a compressor/limiter between the mike and the medium. Set the recording device's input gain high. The singer's whispering: the compressor is transparent because this whisper is below the threshold. The whisper's recording is above the medium's background noise because the recording device's input gain is high enough. The singer's yelling: the compressor springs into action, the peak levels are attenuated, and the yelling is recorded correctly onto the medium, without distortion. A source with a higher dynamic range than the medium's is recorded without distortion.

At this point, one might argue that the dynamic range of the medium was not increased, but the dynamic range of the source was decreased to be "crammed" into the medium's dynamic range. Well, yes and no.

Yes, because the dynamic range of the signal (dynamic range of the RMS, highly correlated to loudness range, which is what eventually matters) was indeed reduced.

No, because the RMS dynamic range / loudness range of what's actually recorded onto the medium is now greater than 50dB. Having reduced the signal's crest factor at high levels, the compressor/limiter did actually add 5 or 10dB of additional RMS dynamic range / loudness range to the original 50dB. The RMS dynamic range of what's recorded onto the medium is now 55 or 60dB, while it would have been 50dB max without any dynamic processing.

Let's put that in numbers (arbitrary values just for the sake of the example).
Dynamic range of the medium: 50dB
Dynamic range of the original vocals: 70dB
Dynamic range of what's recorded with a compressor/limiter: 60dB.

Better?

There are some definition issues here. The dynamic range of a medium is generally given as the maximum level minus the minimum level that can be recorded. If expressed in peak levels in the digital domain, it's a clear definition. If expressed in peak levels in the analog domain, it gets more complicated: why would you consider the background noise's peak levels? It's the RMS level that counts. But then, you have to reason with RMS levels, and consider the maximum RMS level you can record. But for a given peak value (generally the max. instantaneous level the medium can handle while still behaving linearly), the RMS level depends on the signal's crest factor, so there is no absolute "RMS dynamic range" for any medium. It depends on the signal.

Also, when I write "Yes, because the dynamic range of the signal (dynamic range of the RMS, highly correlated to loudness range, which is what eventually matters) was indeed reduced.", it's not systematically true. As I show in the article, some signals show a resilience to compression: their RMS dynamic range / loudness range doesn't decrease at all when compression ratios are reasonable. In numbers:
- dynamic range of the medium: 50dB
- dynamic range of the source: 60dB
- dynamic range of what's recorded: 58dB.
The source dynamic range's reduction is negligible compared to the increase of the medium's.

Let's put the increase of the medium's dynamic range in other words.
Low level signal => no compression => signal is not processed => signal takes advantage of the medium's dynamic range
High level signal => compression => signal is processed, its crest factor is decreased => signal's RMS is increased whereas its peak doesn't increase => additional perceptual / RMS / dynamic / loudness range on "top" of the medium's dynamic range.

I hope it gets clearer now.... this is not easy to understand, and not easy to explain....

Edited by Jennifer Jones (30/09/11 11:02 AM)

Quote Mojobone:

---

Furthermore, the decibels in this paragraph are not referenced to anything; this makes them essentially meaningless, though I presume dBSPL is what's meant, in context.

---

Surely the dynamic range of a system can be expressed simply in dB without a reference?

Edited by Jennifer Jones (30/09/11 11:02 AM)

So is the upshot more to do with noise floor and distortion within a given recording medium?...the argument being that we are recording things with compression and limiters that otherwise would not/could not be recorded as the medium distorts or adds too much noise?..thereby we are artificially increasing dynamic range?...D'oh!

--------------------
My head hurts!

Edited by Jennifer Jones (30/09/11 11:02 AM)

Quote Sam Inglis:

---

...because the RMS dynamic range / loudness range of what's actually recorded onto the medium is now greater than 50dB. Having reduced the signal's crest factor at high levels, the compressor/limiter did actually add 5 or 10dB of additional RMS dynamic range / loudness range to the original 50dB. The RMS dynamic range of what's recorded onto the medium is now 55 or 60dB, while it would have been 50dB max without any dynamic processing.

---

Sorry Sam, but I just can't go along with this...

In the example above dynamic range of the medium has in no way been increased. All material still exists within the 50dB window. All that has happened is what in fact we all quite simply understand, a previously greater dynamic range has been reduced to fit within the operating parameters of a recording system.

How 'resistant' material may be to reduction in dynamic range under compression is simply a function of how it is measured - in time, frequency and amplitude domains.

For me Emmanuelle is simply expressing what we all appreciate but in an opposite way. It seems completely counterintuitive (as shown by some of the feedback I think) and potentially extremely confusing to beginners. If looking at dynamic range in this way led us to new areas of understanding , then it would to my mind have purpose, but as it is, it just seems pointlessly contrary.

Loopy

Edited by Jennifer Jones (30/09/11 11:02 AM)

then he's completely misunderstood what everyone is talking about. He's squashing 60dB of dynamic range into a smaller space - fine. That's what compressions and limiters are for. But the dynamic range is not preserved. The PERCEIVED dynamic range may be - but the actual dynamic range has gone.

He claims that it is "without distortion". Untrue - by using a compressor you have no choice but to introduce distortion. Any change in the waveform is a distortion. Moreover - any harsh use of the compressor will introduce harmonic distortion - which will be audible. IU'm afraid he's fallen for the fools game of "louder is better".


I'll make it simple. If you have a signal whose lowest to highest levels run a range of 60dB and it is then recorded into a medium playing back 58dB of dynamic range - then you have lost dynamic range. The whole article is fun, nice and pretty much an explanation of how to get more performance dBs into a limited dynamic range.

It's called compression and it reduces dynamic range. SOS - you've been had. This is an entropic device - you cannot recover the information once it is lost - you MIGHT like to try an expander to increase the dynamic range of a dynamically limited signal - but the information of the real dynamics are forever lost. It may be IMPLIED by certain mix tricks or compression etc - but the dynamics are GONE. It's a bunk article and if it ever does get posted as a paper to the AES..... get ready to duck.

I mean - c'mon - duh!!!




Oh - and RMS dynamic range doesn't really mean anything.


- just read the definitions in EBU 3342. This does not refer to dynamic range but "loudness range". Not the same thing.

Edited by Jennifer Jones (30/09/11 11:02 AM)

My head hurts. Too many ranges, levels and gains.

My idiot understanding is as follows.

Most recording media has lots of space for very loud and very quiet signals, as long as the input signal is calibrated appropriately so that it's loud enough to be detected (above the noise floor) and not too loud to saturate the medium (clipping, absolute distortion, whatever).

Compressors can help make quiet bits of the input signal sound louder, by (paradoxically) making the loudest bits a bit quieter. This makes a bit of space at the 'loud' end of the recording medium, so we can shift the signal along a bit (increase the input gain on the recording medium) without saturating it.

If the loudest chunks of the input signal are still too loud despite compression, a limiter will absolutely stop the loudest signals from going above a level which will saturate the recording medium. If we limit more and more of the loud end of the signal, we can again shift the signal along a bit (increase the input gain on the recording medium).

Eventually, the 'really quiet' chunks of the input signal, sound much louder on the recording medium. However this comes at a cost: the 'louder' and 'really loud' chunks of the input signal will have been squeezed (compressor) or smashed (limiter), changing their musical waveforms and making them sound different.

Ergo, the whole thing sounds louder, but with less variation between loud and quiet. As the compressors and limiters have done their work and deliberately distorted the recorded waveforms, it's impossible to recreate the original 'really loud' and 'really quiet' signals as that information is no longer there.


As it happens, most tin-pot radios, TV speakers, or car stereos, and a lot of cheap earbuds, don't want really quiet signals because their day-to-day noise floor is much higher than the device (or medium) used to record the signal on and so consumers can't hear it over the car noise or whatever. Therefore, commercially this might not be a problem.

When I'm buying music to listen to in a quiet space with a half-decent stereo (because perversely I like to listen to the artist rather than have them blaring as background music), my system has a low perceived noise floor, and so the lack of variation between quiet and loud, and the distortion on the loud bits, is noticeable, annoying, and gives me a headache after four tracks.

Is that about the nub of it?

Edited by Jennifer Jones (30/09/11 11:02 AM)

Quote turtles:

---

My head hurts.

---

turtles is not alone. What a thread! My head is wrecked!

--------------------
.. is this thing on?

Edited by Jennifer Jones (30/09/11 11:03 AM)

Sounds like the classic case of a person with a little knowledge - its absolutely technically wrong, and its misleading - but in a classic case of Bad Science it has lots of technical words and measurements in there so people without a clear understanding of the subject will be pulled in. I fairly sure Hugh and Paul will be swearing in a back office somewhere.

--------------------
Daniel Davis
Edinburgh Recording Studio Windmill Sound

Edited by Jennifer Jones (30/09/11 11:03 AM)

I've read that article three times trying to understand it.
I had problems understanding some of the terminology as I've never heard of most of the measurment terms "gated RMS variability" "EBU 3342 lounness range" "crest factor" etc and they weren't really explained very well.

I normally treat information in SOS as gospel,so it will be interesting re-reading the article, with the benifit of the comments above.

Whether the article is accurate or not, I really enjoyed reading such an in depth feature with its interesting (if slightly obtuse) graphs.
As an SOS reader for over 20 years, I'd welcome more SOS features aimed at a slightly higher level of understanding.

Edited by Jennifer Jones (30/09/11 11:03 AM)

I read the article whilst sitting next to a pool with a beer in the midst of a Spanish heat-wave, and thoroughly enjoyed it. I particulary liked the concept that I can brick-wall limit my audio to post Death Magnetic levels (incidentally - I love the sound of that CD) yet still maintain great dynamic range so long as I leave some silent bits in it. It all seemed thought provoking but not awfully useful. I'm not sure that "Bollocks" is quite fair so much as "from an alternative perspective" and unfortunately the use of very specifically defined language to describe alternative views is always going to be fraught.

A

--------------------
Andi, www.thedustbowl.net Mixing, Mastering, Audio Editing at The Dustbowl Audio

Edited by Jennifer Jones (30/09/11 11:03 AM)

When things seem bad, I recommend reading Wittgenstein's Tractatus Logico-Philosophicus. Twice.

Edited by Jennifer Jones (30/09/11 11:03 AM)

Following is my answer to criticism expressed by Mojobone, Narcoman, Urthlupe, and Daniel Davis.

To the others, I thank you for trying to understand

Btw my name is Emmanuel, not Emmanuele or Emmanuelle.

Quote Mojobone:

---

Firstly the author is referring to level not gain; the two are not interchangeable.

---

No - reread the sentence. I'm referring to gain.

Quote Mojobone:

---

Second, there's no distinction between clipping distortion of the kind commonly exhibited by a limiter pushed too hard and distortion caused by digital overs.

---

That's simply wrong. Digital overs cause harmonic distortion (harmonics 1, 3, 5 etc.). Well-adjusted limiters don't, and I'm not even sure that limiters pushed too hard do.

Quote Mojobone:

---

He appears to be trying to imply that a limiter can increase the loudness range of a normalized audio file, when this is patently not the case

---

I'm not trying to imply that - at all. Reread the article and the explanation given by Sam above.

Quote Mojobone:

---

technically, dynamic range increases, as the waveform gets taller, on average, but only in the sense of the difference between RMS average and digital black,

---

That's perfectly true. It means that if you put a limiter, you can achieve a higher RMS level from a given signal, and the DR of the medium will have seemed to increase - at the price of the loss of "crest factor". I didn't invent the method, it's been used for a good 50 years.

Quote Mojobone:

---

Furthermore, the decibels in this paragraph are not referenced to anything; this makes them essentially meaningless, though I presume dBSPL is what's meant, in context. Is this author trying to pull a fast one, or am I completely misreading this?

---

As Sam confirms it:
- a logarithmic level measure needs a reference, thus the "FS" in "dB FS" for instance, or the "SPL" in "dB SPL."
- a logarithmic range measure doesn't need a reference. It's essentially a "dimensionless" unit, a ratio. Same thing for the loudness range: EBU3341 loudness is expressed in "LU FS", and EBU3342 loudness range is expressed simply in "LU".

Quote Narcoman:

---

Sounds like he's mixing up perceived values and actual values

---

Read the article again! I give the translation between the two according to ITU1770 and EBU3341, and explain that RMS and EBU3341 loudness are highly correlated in the case of the corpus. Thus, the possibility to consider one or the other indifferently, in the case of this corpus, and in the case of the 3341 norm. You can think the EBU3341 standard is badly defined if you want, that's another story.

Quote Narcoman:

---

Measuring dynamic range with variations in RMS - that's an idiotic argument in musical terms

---

No - if anything, it's a first approximation that may be improved. The approximation is justified by the high correlation between EBU3341 and RMS, and, I repeat, in the case of the corpus I used.

Quote Urthlupe:

---

In the example above dynamic range of the medium has in no way been increased. All material still exists within the 50dB window. All that has happened is what in fact we all quite simply understand, a previously greater dynamic range has been reduced to fit within the operating parameters of a recording system.

---

I agree that this is difficult to understand - refer to Sam's explanation, I can't do much better than that.
The central issue is that very roughly, "instantaneous signal variability" = "RMS variability" + "crest factor". The medium conditions the "instantaneous signal variability", so if you reduce the "crest factor", you increase the "RMS variability".

The lack of common vocabulary does open the door to confusion.

Quote Urthlupe:

---

If looking at dynamic range in this way led us to new areas of understanding , then it would to my mind have purpose, but as it is, it just seems pointlessly

---

It's not quite "new". This method of reducing the "crest factor" in order to increase "RMS variability" for a given "dynamic range of a medium" is not as useful as it was, due to the high dynamic ranges achieved by digital media. In the 50's / 60's, it was another story.
But - consider temporarily the musical accompaniment as "background noise" and the lead vocals as "signal" - that makes you understand why lead vocals are compressed. And next time you compress lead vocals, you know why. Is that so "pointless"?

Quote narcoman:

---

then he's completely misunderstood what everyone is talking about. He's squashing 60dB of dynamic range into a smaller space - fine. That's what compressions and limiters are for. But the dynamic range is not preserved. The PERCEIVED dynamic range may be - but the actual dynamic range has gone.

---

Please re-read the article. And, can you give me a definition of "dynamic range" of a signal while you're at it?
The real question might very well be: why are you so angry?

Quote narcoman:

---

He claims that it is "without distortion". Untrue - by using a compressor you have no choice but to introduce distortion. Any change in the waveform is a distortion. Moreover - any harsh use of the compressor will introduce harmonic distortion - which will be audible. IU'm afraid he's fallen for the fools game of "louder is better".

---

Listen to "The Prettiest Thing" by Norah Jones: heavily limited, perfectly pristine.

"Louder is better"? Me? Did it occur to you that if I wasn't concerned about over limiting, I wouldn't have gone to all the trouble to investigate the issue to such lengths? The basic purpose of the article is to make people think about the actual reasons why recent music is sometimes tiring, to make them think about the loudness war, etc. "Less DR" doesn't cut it, the matter is much more complex. Know your enemy - if you want to fight loudness war, know what it does...

Read this review:
http://productionadvice.co.uk/loudness-war-dynamic-range/
Ian Shepherd is the organizer of the dynamic range day, I suppose he fell for "louder is better", too?

Quote narcoman:

---

I'll make it simple. If you have a signal whose lowest to highest levels run a range of 60dB and it is then recorded into a medium playing back 58dB of dynamic range - then you have lost dynamic range. The whole article is fun, nice and pretty much an explanation of how to get more performance dBs into a limited dynamic range.

It's called compression and it reduces dynamic range. SOS - you've been had. This is an entropic device - you cannot recover the information once it is lost - you MIGHT like to try an expander to increase the dynamic range of a dynamically limited signal - but the information of the real dynamics are forever lost. It may be IMPLIED by certain mix tricks or compression etc - but the dynamics are GONE. It's a bunk article and if it ever does get posted as a paper to the AES..... get ready to duck.

---

Cool down. Read the explanation again. I also state in another paragraph that limiters reduce the RMS variability.
The RMS variability of the original content will be reduced, OK, but the eventual RMS variability will be increased - and that was the point.
The issue is not so simple - I draw your attention again to "instantaneous signal variability" = "RMS variability" + "crest factor". Think it over.

What you're doing is that you read parts from the article, forget about the other parts, draw bad conclusions, and then for some reason think that I'm trying to fool the journal. Isn't that weird? Really, why would you think that I would try to trick SOS? What would I get from it?

Quote Daniel Davis:

---

Sounds like the classic case of a person with a little knowledge - its absolutely technically wrong, and its misleading - but in a classic case of Bad Science it has lots of technical words and measurements in there so people without a clear understanding of the subject will be pulled in. I fairly sure Hugh and Paul will be swearing in a back office somewhere.

---

Sorry to disappoint you, but Hugh's reaction was "It was in every way a superb article."
(See http://www.soundonsound.com/forum/showflat.php?Cat=&Number=937023& page=2&view=collapsed&sb=5&o=&fpart=1#937023.)

If you think that the article is technically wrong, then I guess both the ITU and the EBU are wrong too, since I take a lot of notions from them.

@Narcoman & Daniel Davis, I really wonder why some people are so aggressive? If you disagree, just say you do… don't tell me that I have little knowledge, or that this is Bad Science. You can even feel free to prove that I'm wrong. But opinions like that, where does that lead anyone?


Emmanuel

Edited by Jennifer Jones (30/09/11 11:03 AM)

This is clearly a very complicated topic, and some of Emmanuel's claims and arguments are 'challenging' -- but there's nothing wrong in that. It's only by exploring different theories and propositions that we can arrive at a definitive answer. I suspect there is also some confusion in parts because of different interpretations being placed on apparently familiar terms.

I'll let Emmanuel defend his own article, but would remind everyone to play nice please

There is one small point that I must challenge, though:

Quote Emmanuel D.:

---

That's simply wrong. Digital overs cause harmonic distortion (harmonics 1, 3, 5 etc.).

---

Digital overs don't result in harmonic distortion. They actually produce anharmonic distortion. The process is that digital overs initially generate harmonic distortion within the quantiser because, having run out of numbers, the waveform tops are clipped. However, these harmonics naturally extend beyond half the sampling rate, and they are therefore aliased within the frequency domain to appear at frequencies below their fundamental. The distortion products appear at frequencies which are the mathematical product of the harmonics and the sample rate, not a normal linear harmonic series.

hugh

--------------------
Technical Editor, Sound On Sound

Edited by Jennifer Jones (30/09/11 11:03 AM)

Quote Emmanuel D.:

---

Please re-read the article. And, can you give me a definition of "dynamic range" of a signal while you're at it?
The real question might very well be: why are you so angry?

---

Yo big guy. Thanks for responding. No anger here at all (aside from the frustration of your article giving credence to the loudness 'tards in the world). Don't read between the lines.....

Dynamic range - can only be - loudest signal to quietest signal (not silence). Anything else is trying to fit a square peg into a round hole.

Quote Emmanuel D.:

---

Listen to "The Prettiest Thing" by Norah Jones: heavily limited, perfectly pristine.

---

Well this is becoming clearer now. That certainly isn't a pristine recording at all. It's distorted. It SOUNDS distorted and it reports as distorted via any simple DAW analysis. Pristine? It's very hard to listen to outside of a car or a laptop. I thought that the day it came out an I think it even more now. It's still a good record though!

Listen to the Berlin Phil. Something none classical ? What about Tron soundtrack mixed by Alan Meyerson - big distorted sounds in there, but all in the mix and not in the master presentation. OR Crooked Vultures on Vinyl. Great sound. Sounds terrible on CD (although still hugely gratifying on a rawk level). Dirtbombs - Ultraglide in Black. Limited to hell - works in the medium but i'd never say it was a "beautiful recording". I'd say it was a loud and rawks recording. Sometimes it's what we want!

I mixed the album of a well known two piece back in '04. The masters were nothing like my mixes in terms of clarity although they DID sound more powerful. In my job I get to hear a lot of mixes before they're mastered, well known ones too (Crazy by Gnarls Barkley still sticks in my mind as an amazing mix but poor and brittle master). Now - I haven't heard the mixes from Norah Jones' album but if my (not inconsiderable) experience with myriad other releases and projects is anything to go by .... well - you know where I'm going.

Quote Emmanuel D.:

---

Read this review:
http://productionadvice.co.uk/loudness-war-dynamic-range/
Ian Shepherd is the organizer of the dynamic range day, I suppose he fell for "louder is better", too?

---

No - he hasn't given your article a negative or positive response. He claimed it to be an excellent article. Not that he agrees with you. Hes even outlined where your article is wrong and where people will get confused.

Quote Emmanuel D.:

---

Quote narcoman:

---

I'll make it simple. If you have a signal whose lowest to highest levels run a range of 60dB and it is then recorded into a medium playing back 58dB of dynamic range - then you have lost dynamic range. The whole article is fun, nice and pretty much an explanation of how to get more performance dBs into a limited dynamic range.

---

Cool down. Read the explanation again. I also state in another paragraph that limiters reduce the RMS variability.
The RMS variability of the original content will be reduced, OK, but the eventual RMS variability will be increased - and that was the point.
The issue is not so simple - I draw your attention again to "instantaneous signal variability" = "RMS variability" + "crest factor". Think it over.

---

Is there something wrong with your thermostat? You've written a contentious article (mostly the title which is NOT supported by the standard you are looking at - there is a big difference between loudness range and dynamic range. Even the EBU paper says to be CAREFUL in that analysis) and it'll be given strong credence by the loudness loons. It IS that simple. You cannot claim greater dynamic range when you've raised the gain and limited the top. No anger here at all.

You've separated out RMS variability (which can only ever been experienced as a rough guide to perceived volume) from Crest factor. When dealing with an absolute of dynamic range (which ISN'T an RMS factor - it's very mechanism denotes time domain) you can't infer a time domain based event showing a correlation with a static difference measurement. It doesn't make sense. The mathematics of it just doesn't hold water! When dealing with the issues in limiters (that of removed transients) the RMS variability doesn't come into play. It just doesn't matter and has no effect on what the concerns over the loudness wars are about - that of loss of detail through removed transients.

Now - one COULD argue that quieter events are brought forward and may be more hearable - but I'd argue that those details were always there and could have been revealed by a simple turn of the gain knob. Okay - you could counter argue and say the larger transient events would be intolerable if unlimited but that's not what we're debating.

Quote Emmanuel D.:

---

@Narcoman & Daniel Davis, I really wonder why some people are so aggressive? If you disagree, just say you do… don't tell me that I have little knowledge, or that this is Bad Science. You can even feel free to prove that I'm wrong. But opinions like that, where does that lead anyone?
Emmanuel

---

It is (unintentional) bad science is is written like a marketing document. One doesn't have to prove you're wrong when you've not proven you're right! The points raised by your opponents have picked holes in your reasoning. I understand the POINT of your article but your reasoning is flawed. You've also mixed up loudness range with dynamic range (intentionally? Probably not) which EBU3341 goes at great pains to warn the user away from. You've also tried to intimate that your 50 dB range medium correctly encodes a 60dB range piece of work. THAT is bad reasoning. It doesn't - it's an entropic process!! Information is lost.

The problem we have is that SOS has seen fit to publish an article which the loudness loons will run with as proof that it's okay to smash the [ ****** ] out of stuff! Have you seen the number of people who have already jumped on your article as a holy grail!!

The intention of the article may be great (as I can see from your responses it is) but it's poorly written (it mixes up established definitions within the biz) and the title itself, unwisely chosen. I spent many years in academia as a published research scientist in maths for signal communication - one has to be VERY careful about title and wording in any paper. I wrote two papers for the Virtuosi project; the second was a rethink about my broadly panned paper on the energy at the turning points of waveforms through physical media after I was brutally shot down by none other than Carl Feynmann. I wasn't wrong but my argument hadn't been thought through well enough (something my idiot of a supervisor should have spotted).

I recommend a re-think about how you've presented this; your responses here do show you care about this subject but the article plays into the loudness loons hands.

Edited by Jennifer Jones (30/09/11 11:04 AM)

Quote Emmanuel D.:

---

To the others, I thank you for trying to understand

---

I am also trying to understand Emmanuel. Apolgies for mis-spelling your name.

Quote Emmanuel D.:

---

Quote Urthlupe:

---

In the example above dynamic range of the medium has in no way been increased. All material still exists within the 50dB window. All that has happened is what in fact we all quite simply understand, a previously greater dynamic range has been reduced to fit within the operating parameters of a recording system.

---

I agree that this is difficult to understand - refer to Sam's explanation, I can't do much better than that.
The central issue is that very roughly, "instantaneous signal variability" = "RMS variability" + "crest factor". The medium conditions the "instantaneous signal variability", so if you reduce the "crest factor", you increase the "RMS variability".

The lack of common vocabulary does open the door to confusion.

---

In that explanation Emmanuel you wrote - 'The RMS dynamic range of what's recorded onto the medium is now 55 or 60dB, while it would have been 50dB max without any dynamic processing. '

This is simply not correct, what is now recorded onto the medium does not exceed the dynamic range of that medium - 50dB in your example. A fundamentally inaccurate and confusing way to express your point, regardless of a lack of common vocabulary.

Quote Emmanuel D:

---

Quote Urthlupe:

---

If looking at dynamic range in this way led us to new areas of understanding , then it would to my mind have purpose, but as it is, it just seems pointlessly

---

It's not quite "new". This method of reducing the "crest factor" in order to increase "RMS variability" for a given "dynamic range of a medium" is not as useful as it was, due to the high dynamic ranges achieved by digital media. In the 50's / 60's, it was another story.
But - consider temporarily the musical accompaniment as "background noise" and the lead vocals as "signal" - that makes you understand why lead vocals are compressed. And next time you compress lead vocals, you know why. Is that so "pointless"?

---

I don't think you understand what I said. You might re-read it. I am also (unfortunately) 'not quite new', I have been using methods of controlling dynamic range in various audio recording, playback and restoration systems since the late 70's and hope to god I will understand my reasons the next time I compress lead vocals (and within the context of your article).

For what its worth Emmanuel I don't think you should take some of the comments here so personally - If nothing else your writing has stimulated thought and discussion which is a very positive thing.

Loopy

Edited by Jennifer Jones (30/09/11 11:04 AM)

...those of us who "compress vocals" do it knowing full well we are reducing it's dynamic range to fit in in the artistic or technical parameters of our mix!! we're not fooling ourselves that we've kept the same dynamics in a more limited space.

Edited by Jennifer Jones (30/09/11 11:04 AM)

Quote narcoman:

---

...those of us who "compress vocals" do it knowing full well we are reducing it's dynamic range to fit in in the artistic or technical parameters of our mix!! we're not fooling ourselves that we've kept the same dynamics in a more limited space.

---

... then please, be as kind as to explain what is the "dynamic range" of a vocal part, or for that matter of any piece of music. So that people can know what is "reducing its dynamic range to fit in".

Edited by Jennifer Jones (30/09/11 11:04 AM)

Quote Emmanuel D.:

---

... then please, be as kind as to explain what is the "dynamic range" of a vocal part, or for that matter of any piece of music. So that people can know what is "reducing its dynamic range to fit in".

---

That's a very obtuse request as it's fundamental to your argument....

The dynamic range of a signal is NOT it's perceived dynamics. At all. Ever. It's is purely measured and quantifiable. Now - sometimes there are different WAYS of measuring things - but in dynamic range it is only the quietest part of a signal to the loudest. Windowing a signal in the time domain can lead to skewed results.

The dynamic range of a part, or piece of music, can only EVER be it's quietest signal to it's loudest. That is it. ANY other measure is not dynamic range. You appear more and more to be referring perceptual coding - which is what SOME limiting does, gives the impression of louder music (where simple gain may have sufficed) within a limited dynamic range application.

If the minimum part of a vocal is down at -40dB(FS) and you're highest at -15dB(FS) (both very unlikely!!) then your dynamic range is bounded by those two values. The dynamic range of your signal is not (not now not ever):

I have a piece with 72 dB dynamic range , I record it through a mic and compressor bringing it down to 25dB dynamic range and that recording now has a 72dB dynamic range. The information is lost. It may have the appearance (or sonics) of maintaining all of it's dynamics since we as humans know what a quiet voice kind of sounds like - we'd hear it as intimate and subjectively "quieter". But it ISN'T quieter. The dynamic range is measurably what is present in the recording no matter what the original was.

You put a limiter at -6dB(FS) on a signal peaking at 0dB(FS) and then raise it's level by 6dB - you have LOST 6dB of dynamic range. You've brought the noise floor up. You may even perceptually reveal details that you didn't hear before - but your dynamic range is reduced. By 6dB as it happens.

Did you not read my prior lengthy response?

Sorry to be potentially a little rude here: is English not your first language and are we having subtle language issues that you may not be aware of? Perhaps that is at the root of the issue? Sorry if that isn't true - just wondering

Edited by Jennifer Jones (30/09/11 11:04 AM)

Quote Urthlupe:

---

This is simply not correct, what is now recorded onto the medium does not exceed the dynamic range of that medium - 50dB in your example. A fundamentally inaccurate and confusing way to express your point, regardless of a lack of common vocabulary.

---

I assure you the problem lies in the lack of a common vocabulary, and to the arbitrary acceptation of notions that don't stand a closer examination.

Let's start with the dynamic range of a medium.

The top of the range is well-defined, at least in the case of digital media: a sample should not exceed 0dB FS.
The bottom of the range is less well-defined. Assuming for a moment that the background noise is white noise, should we take the bottom value as the highest sample belonging to this noise? It wouldn't make sense. We should rather consider the RMS of the background noise, or better yet, its loudness.
But there is an immediate problem: the top of the range is an instantaneous value, and the bottom an averaged one.
Evaluating the dynamic range, you'd be subtracting an RMS level (or something not far from it) to a peak level. Now that doesn't make sense.
To make sense, we should be subtracting either an RMS value to another RMS value, or a peak value to a peak value.
But the peak value of a white noise is meaningless as far as its RMS and loudness are concerned.
So let's consider subtracting an RMS value to another RMS value.

The RMS value of the background noise (white or otherwise) is known.
But the RMS value depends on the signal that's to be stored.

Let's focus on this RMS value. We agree on the fact that if we subtract one RMS from another one, then the dynamic range of the medium will be the maximum value of the signal's RMS minus the background noise RMS value.
So, what's the maximum RMS value the signal can take?…. it depends on the signal itself.
It means that the SNR of a medium expressed in RMS values depends as much on the signal as to the medium
(I know it sounds strange, but since we can't express the SNR otherwise, how to go around this?)

One thing is sure, the signal's peak cannot go above 0dB FS. So the SNR of the medium is, still from that RMS perspective, related to the ratio between the signal's peak and its RMS. In other words, to the crest factor. If the crest factor decreases, the "RMS SNR" increases. And, any processing that reduces the crest factor will increase the medium's SNR when expressed in terms of RMS. Conclusion, using a limiter will increase the medium's SNR.

I know it sounds shocking. The core of the problem is: how to define the dynamic range of a medium in the first place? Peak minus peak makes no sense, since peak is almost not related to loudness. Peak minus RMS makes no sense, since those are two different notions. It only leaves RMS minus RMS or loudness minus loudness… and it leads to that conclusion that you apparently find unacceptable.

But really, provide an alternative solution, I'd be happy to hear it, and it's not a figure of speech.

Quote Urthlupe:

---

For what its worth Emmanuel I don't think you should take some of the comments here so personally - If nothing else your writing has stimulated thought and discussion which is a very positive thing.

---

That's good news

Edited by Jennifer Jones (30/09/11 11:05 AM)

Quote Hugh Robjohns:

---

Digital overs don't result in harmonic distortion. They actually produce anharmonic distortion. The process is that digital overs initially generate harmonic distortion within the quantiser because, having run out of numbers, the waveform tops are clipped. However, these harmonics naturally extend beyond half the sampling rate, and they are therefore aliased within the frequency domain to appear at frequencies below their fundamental. The distortion products appear at frequencies which are the mathematical product of the harmonics and the sample rate, not a normal linear harmonic series.

---

I'm fine with that.
So, the reformulation of my argument would be: limiters don't result in anharmonic distortion.
If I'm wrong, correct me, but the kind of distortion caused by limiters is usually not a variant of harmonic distortion.

Edited by Jennifer Jones (30/09/11 11:05 AM)

Quote Emmanuel D.:

---

I assure you the problem lies in the lack of a common vocabulary, and to the arbitrary acceptation of notions that don't stand a closer examination.

---

Possibly

Quote Emmanuel D.:

---

Let's start with the dynamic range of a medium.

The top of the range is well-defined, at least in the case of digital media: a sample should not exceed 0dB FS.
The bottom of the range is less well-defined. Assuming for a moment that the background noise is white noise, should we take the bottom value as the highest sample belonging to this noise? It wouldn't make sense. We should rather consider the RMS of the background noise, or better yet, its loudness.

---

That's the problem. You should be using an instant measured point in the signal (noise included).

Quote Emmanuel D.:

---

But there is an immediate problem: the top of the range is an instantaneous value, and the bottom an averaged one.
Evaluating the dynamic range, you'd be subtracting an RMS level (or something not far from it) to a peak level. Now that doesn't make sense.

---

of course.

Quote Emmanuel D.:

---

To make sense, we should be subtracting either an RMS value to another RMS value, or a peak value to a peak value.
But the peak value of a white noise is meaningless as far as its RMS and loudness are concerned.

---

That is where you're reasoning breaks down. You should average the PEAKS of the noise over time to use a working value. Second, if your signal is always OVER the noise floor then the noise floor is largely not relevant (in practise that gets more difficult if you are using a recording with sections of silence - a voice recording for example. Where you measure matters and is indeed a point for argument. BUT this becomes moot in terms of "noisy" music). You must make a decision to the minimum point.

Quote Emmanuel D.:

---

It only leaves RMS minus RMS or loudness minus loudness… and it leads to that conclusion that you apparently find unacceptable.

---

Yes. It's is unacceptable as a measure of dynamic range. It can only lead to an evaluation of the RMS of dynamic range. The only way you could make this acceptable would be to form an calculable expression of the two RMS values and look at the difference between the differentials of the two values. Bearing in mind that would mean forming a calculus expression for Bryan Adams I think we can dismiss that as a little bit "out there".

Quote Emmanuel D.:

---

But really, provide an alternative solution, I'd be happy to hear it, and it's not a figure of speech.

---

Elective measurement between peaks and troughs OR averaged measurement of troughs against peak (although that makes it somewhat "averaged". I recognise the difficulty in deciding a minimum but it isn't a valid solution to look at differences between RMS values as a measure of storable and worked dynamic range.

Edited by Jennifer Jones (30/09/11 11:05 AM)

Quote narcoman:

---

You've separated out RMS variability (which can only ever been experienced as a rough guide to perceived volume) from Crest factor. When dealing with an absolute of dynamic range (which ISN'T an RMS factor - it's very mechanism denotes time domain) you can't infer a time domain based event showing a correlation with a static difference measurement. It doesn't make sense. The mathematics of it just doesn't hold water!

---

Please notice the quotes I put:
"instantaneous signal variability" = "RMS variability" + "crest factor"
As I explain in a previous answer, I know well enough that you can't write an equation that contains both windowed and instantaneous signal descriptors.
But then what can I do? If I'm too precise, things get apparently too austere for some.
So I sum up, but then I'm not precise enough.

I'm sure you understand
"instantaneous signal variability" = "RMS variability" + "crest factor"
well enough to see through the necessary simplifications.

Quote narcoman:

---

When dealing with the issues in limiters (that of removed transients) the RMS variability doesn't come into play. It just doesn't matter and has no effect on what the concerns over the loudness wars are about - that of loss of detail through removed transients.

---

You just re-phrased the whole point of the article.
Indeed, the loudness war is about removed transients, not dynamic range. I couldn't agree more.

Edited by Jennifer Jones (30/09/11 11:05 AM)

Quote Emmanuel D.:

---

As I explain in a previous answer, I know well enough that you can't write an equation that contains both windowed and instantaneous signal descriptors.
But then what can I do? If I'm too precise, things get apparently too austere for some.
So I sum up, but then I'm not precise enough.

---

Feel free to get as precise as you like - Maths PhD here!! I'll get it

Quote Emmanuel D.:

---

You just re-phrased the whole point of the article.
Indeed, the loudness war is about removed transients, not dynamic range. I couldn't agree more.

---

This is the fundamental disagreement - removal of transients IS the removal of dynamic range based on the correct dynamic range calculation. If you want a mathematical idiom to refer to - when you deal in two RMS values from the same source you are including and RMS of the included noise TWICE. As you limit the signal you merely bring the noise UP along with the signal. The dynamic range lost is precisely the amount of limiting done (of course, if no limiting take place then you're just moving the window).

To quantity your proposal as proof no dynamic range is lost (and it's in the definition of dynamic range that your argument doesn't work - it DOES work as an assessment of RMS values to information) you'll need to formulate a set of mathematical statements. Since including a root mean square in any set means using differential calculus for any instantaneous measurement you've already shown why you cannot compare the min and max values of the RMS as a valid form of dynamic range. What you really have is a measure of the average variability but not of any maxima or minima. Therefore NOT dynamic range. The two are not well correlated.


I can see where you might get the notion that SNR is increased - but you are only talking about the inherent system noise within the gain system POST limiting. However all that has happened in a "limited by 6dB" file (i.e. one in which 6dB has been removed from the top end) is the difference maxima and minima have been reduced. The crest factor is reduced, the RMS is raised (if we apply the 6dB gain to get the limited signal louder and the noise floor WITHIN the signal recorded is raised the same amount. Clipping transients is the SAME as removing dynamic range.


Edited by Jennifer Jones (30/09/11 11:05 AM)

Quote narcoman:

---

That is where you're reasoning breaks down. You should average the PEAKS of the noise over time to use a working value. Second, if your signal is always OVER the noise floor then the noise floor is largely not relevant (in practise that gets more difficult if you are using a recording with sections of silence - a voice recording for example. Where you measure matters and is indeed a point for argument. BUT this becomes moot in terms of "noisy" music). You must make a decision to the minimum point.

---

I will rephrase your first objection. The SNR would be a windowed measure of the signal's peaks minus a windowed measure of the noise's peaks? If so, how does that contradict what I said? On the contrary, it seems to be homogenous to my the conclusions I draw.

Your second objection I'm afraid I'm not getting. How can the noise floor be not relevant in the process of evaluating the SNR?

Quote narcoman:

---

Yes. It's is unacceptable as a measure of dynamic range. It can only lead to an evaluation of the RMS of dynamic range. The only way you could make this acceptable would be to form an calculable expression of the two RMS values and look at the difference between the differentials of the two values. Bearing in mind that would mean forming a calculus expression for Bryan Adams I think we can dismiss that as a little bit "out there".

---

"the RMS of dynamic range"? we seem to be having a vocabulary problem once more
I begin to suspect we eventually agree, though we don't use the same approach exactly
can you rephrase that so that I can be sure?

"calculus"??? we're dealing with discrete measures, evaluating the differential is only subtracting the signal with itself, with a delay of one sample
not so "out there"
what would the derivative bring? can you be more explicit?

Quote narcoman:

---

Elective measurement between peaks and troughs OR averaged measurement of troughs against peak (although that makes it somewhat "averaged". I recognise the difficulty in deciding a minimum

---

Sorry, you'll have to rephrase, it seems that this is a very long piece of reasoning compacted in 3 lines, but it sounds interesting.

Edited by Jennifer Jones (30/09/11 11:05 AM)

Quote narcoman:

---

This is the fundamental disagreement - removal of transients IS the removal of dynamic range based on the correct dynamic range calculation. If you want a mathematical idiom to refer to - when you deal in two RMS values from the same source you are including and RMS of the included noise TWICE. As you limit the signal you merely bring the noise UP along with the signal. The dynamic range lost is precisely the amount of limiting done (of course, if no limiting take place then you're just moving the window).

To quantity your proposal as proof no dynamic range is lost (and it's in the definition of dynamic range that your argument doesn't work - it DOES work as an assessment of RMS values to information) you'll need to formulate a set of mathematical statements. Since including a root mean square in any set means using differential calculus for any instantaneous measurement you've already shown why you cannot compare the min and max values of the RMS as a valid form of dynamic range. What you really have is a measure of the average variability but not of any maxima or minima. Therefore NOT dynamic range. The two are not well correlated.

---

OK now we (almost) understand each other.

In the article, I was careful of never using the word "dynamic range" as far as audio content is concerned:

There remains the question of whether one should use such a term as ‘dynamic range’ at all: there is no official definition for it, and it may be confused with the dynamic range of a recording medium, which is basically the difference between the highest and lowest level it can handle. During the course of this article, therefore, I won’t talk about ‘dynamic range’ in relation to a piece of music. Instead, I will be using ‘RMS variability’, or more generally ‘dynamic variability’. The term ‘dynamic range’ will be reserved for the measure of signal-to-noise ratio of a recording medium. I will use the term ‘loudness range’ in strict reference to the EBU 3342 document, and the term ‘loudness variability’ in other cases involving loudness instead of RMS.

I based my evaluation of "ranges" on the EBU3342 document, as I state. This particular quantity doesn't decrease with the loudness war. Apparently, you already knew that, but many people assume the contrary (just look at Wikipedia). Therefore, it was worth it to propose an experiment that would clearly show that the actual problem is crest factor, not a "range" in the like of which is defined by 3342.

I agree that the box about how limiters increase the dynamic range of the medium can be easily contradicted when considering other approaches of the dynamic range of a medium. However, you will agree that signals with a lower crest factor may allowed to show a higher RMS variability, given that the medium is the same. So that the use of limiters may result in higher "musical dynamics" (Forte.... Mezzoforte... Piano...)... and that was an interesting and useful point.

May I suggest that as far as mathematical concerns go, you PM me? Your approach is intriguing and I want to know more, but I fear that it may scare off many readers?

Edited by Jennifer Jones (30/09/11 11:05 AM)

Quote Emmanuel D.:

---

I will rephrase your first objection. The SNR would be a windowed measure of the signal's peaks minus a windowed measure of the noise's peaks? If so, how does that contradict what I said? On the contrary, it seems to be homogenous to my the conclusions I draw.

---

That's the very problem. It may we windowed but you're asking for a time domain value to be a fair comparison for deducing dynamic range. By it's very nature a max and min can only be from an instant. As I say - I recognise the problem with deciding a minima when noise is present. But REMOVE noise from this as a mathematical concept - the noise is merely a problem of implementation not of reasoning. The noise can be used as valid signal. It makes no difference. Signal is signal.

Quote Emmanuel D.:

---

Your second objection I'm afraid I'm not getting. How can the noise floor be not relevant in the process of evaluating the SNR?

---

Why are we concerned with the SNR when looking at dynamic range? If the lowest part of your signal is under the noise floor then it IS noise. It's under our windowed function for a minima. Again - I recognise deciding the minima as the issue here.

Quote Emmanuel D.:

---

"the RMS of dynamic range"? we seem to be having a vocabulary problem once more
I begin to suspect we eventually agree, though we don't use the same approach exactly
can you rephrase that so that I can be sure?

---

You can't use a time domain collation as a result for deciding a max and min.

Quote Emmanuel D.:

---

"calculus"??? we're dealing with discrete measures, evaluating the differential is only subtracting the signal with itself, with a delay of one sample

---

EXACTLY!!!! You CAN'T use a time domain response for discrete measurement. You must search and find (in digital terms) the biggest and smallest sample value. No RMS.

Quote Emmanuel D.:

---

not so "out there"
what would the derivative bring? can you be more explicit?

---

It would bring a point value between the two RMS values. It wouldn't bring us a max or min for dynamic range!! I'm using it as the math basis for rejecting the hypothesis that RMS can be used as a meaningful basis for asserting dynamic range. It can't.

Quote Emmanuel D.:

---

Quote narcoman:

---

Elective measurement between peaks and troughs OR averaged measurement of troughs against peak (although that makes it somewhat "averaged". I recognise the difficulty in deciding a minimum

---

Sorry, you'll have to rephrase, it seems that this is a very long piece of reasoning compacted in 3 lines, but it sounds interesting.

---

Simple stuff. Nothing complex. If you want to take account of the noise floor (and I'm saying you shouldn't because once recorded the noise IS the signal) you would need to reason a minimum value. Personally as the noise is now considered signal then the minimum value all in is the one to choose. However - really you should look to decide you signal value in some kind of amplitude "window". What IS the minim of the voice as it tails off? Where DO we decide the sound has ended?

But basically - the dynamic range of a signal is only it's loudest amplitude minus it's quietest amplitude. Including noise!! The dynamic range of something peaking at 0dB(FS) that has lots of digital silence is the full 16bit or 24bit range. Now - if those peaks were limited then the dynamic raneg used to be MORE than 16 or 24bit would hold but it has been reduced by limiting. The peaks (notably the transients although not always) have been curtailed.

Edited by Jennifer Jones (30/09/11 11:06 AM)

Quote narcoman:

---

I can see where you might get the notion that SNR is increased - but you are only talking about the inherent system noise within the gain system POST limiting. However all that has happened in a "limited by 6dB" file (i.e. one in which 6dB has been removed from the top end) is the difference maxima and minima have been reduced. The crest factor is reduced, the RMS is raised (if we apply the 6dB gain to get the limited signal louder and the noise floor WITHIN the signal recorded is raised the same amount. Clipping transients is the SAME as removing dynamic range.

---

I agree with you --- but where did you read that I was limiting a file? I'm limiting a signal in order to put it in a file.

When you say: "but you are only talking about the inherent system noise within the gain system POST limiting"
Why yes, that's what I do... that's exactly what I'm doing. Never stated otherwise, or maybe I wasn't clear enough.

Edited by Jennifer Jones (30/09/11 11:06 AM)

Quote Emmanuel D.:

---

I agree that the box about how limiters increase the dynamic range of the medium can be easily contradicted when considering other approaches of the dynamic range of a medium. However, you will agree that signals with a lower crest factor may allowed to show a higher RMS variability, given that the medium is the same. So that the use of limiters may result in higher "musical dynamics" (Forte.... Mezzoforte... Piano...)... and that was an interesting and useful point.

---

Absolutely - musical dynamics can easily be inferred by use AND abuse of limiters. But they are not a scientific measure! It's perfectly simple to produce a mix that elicits no excitement in the user (and they refer to it is flat or boring) yet could easily be shown that it is, in fact, very dynamic. The difference between the artistic term "dynamic" an the quantifiable one is a brutal line! The issue of limiters reducing dynamic range isn't a musical one but a technical one with all the associated wearying distortions.

Quote Emmanuel D.:

---

May I suggest that as far as mathematical concerns go, you PM me? Your approach is intriguing and I want to know more, but I fear that it may scare off many readers?

---

It was merely a flight of fancy - but I don't think producing a minima from averaging of peak values in noise carries any more weight than seeking it's RMS. Suffice to say - the signal and noise become the signal without reference to SNR once recorded (not withstanding any new noise introduced through re-quantisation of signal in the gain process).

Edited by Jennifer Jones (30/09/11 11:06 AM)

Quote Emmanuel D.:

---

I agree with you --- but where did you read that I was limiting a file? I'm limiting a signal in order to put it in a file.

When you say: "but you are only talking about the inherent system noise within the gain system POST limiting"
Why yes, that's what I do... that's exactly what I'm doing. Never stated otherwise, or maybe I wasn't clear enough.

---

yeah - that's what I think after reading your posts on here. It's the article that isn't clear - not your thought process.

When limiting a signal at source (a real voice, or a synth etc) you're still raising the noise floor and reducing the dynamics of the original signal. Probably in a pleasing way - but it's a reduction none the less. The "file" bit was an assumption on my behalf. The problem with modern limited masters is the pre-recorded digital source being squished into CD for some willy waving contest on CD changers

Edited by Jennifer Jones (30/09/11 11:06 AM)

Quote narcoman:

---

Absolutely - musical dynamics can easily be inferred by use AND abuse of limiters. But they are not a scientific measure! It's perfectly simple to produce a mix that elicits no excitement in the user (and they refer to it is flat or boring) yet could easily be shown that it is, in fact, very dynamic. The difference between the artistic term "dynamic" an the quantifiable one is a brutal line! The issue of limiters reducing dynamic range isn't a musical one but a technical one with all the associated wearying distortions.

---

Indeed we completely agree with each other. That's the main point of the article, nicely phrased.
Still, many people have been saying otherwise. And that's was why I wrote the article....

Edited by Jennifer Jones (30/09/11 11:06 AM)

ha!!

Reckon that'd be a useful sig strip for both of us : "beware the distinction between dynamic and dynamic".

Edited by Jennifer Jones (30/09/11 11:06 AM)

This reminds me of the two wizards fighting in The Raven.

Anyway a good thread that seems to have resolved itself.

--------------------
http://soundcloud.com/richie-royale
http://www.mixcrate.com/richieroyale

Edited by Jennifer Jones (30/09/11 11:06 AM)

Quote Emmanuel D.:

---

Let's start with the dynamic range of a medium.

---

I think this is one area where there might be confusion in the terminology. Audio can have a dynamic range, but as far as I'm aware a medium can't. A medium has a signal-to-noise ratio, and we try to fit the audio signal's dynamic range inside the available rage described by the signal-to-noise ratio.

That signal-to-noise ratio is a measure of the averaged rms noise floor relative to either the peak signal level or the nominal operating signal level, depending on the device in question, expressed in decibels (with no reference suffix).

Quote:

---

It means that the SNR of a medium expressed in RMS values depends as much on the signal as to the medium

---

Er... no, I'm not convinced (see above). SNR and dynamic range are not interchangeable terms. SNR is a clearly defined measurement.

Quote:

---

One thing is sure, the signal's peak cannot go above 0dB FS.

---

Actually the 'reconstructed' signal can, quite legitimately -- read up on intersample peaks. Although I do understand the point you're trying to make here.

Quote:

---

Conclusion, using a limiter will increase the medium's SNR.

---

I have to say I'm not at all convinced by the logic of your argument, not least because I'm unclear on your base assumptions.

hugh

--------------------
Technical Editor, Sound On Sound

Edited by Jennifer Jones (30/09/11 11:07 AM)

..but having waded through the discussion between Emmanuel and Narcoman the mists are lifting!

hugh

--------------------
Technical Editor, Sound On Sound

Edited by Jennifer Jones (30/09/11 11:07 AM)

Hugh:

I will try to proceed step by step, and also to avoid terminology issues.

We can agree on the fact that for any recording medium, there is a limit above which the signal gets unreasonably degraded. Or degraded above a certain "normative limit" deemed as unacceptable. The degradation might be harmonic distortion, digital clipping, or anything else. This limit can be well-defined, or the phenomenon can be progressive, it depends. In this case, the "normative limit" will be used as a reference.

Also, for any recording medium, there is background noise. It can appear under the form of white noise, or vinyl clicks, or anything else. The background noise is often locally unpredictable, but globally predictable.

Let's say I've got an audio source. This audio source emits a signal, which also features a background noise. I will suppose that the source's signal's background noise is always softer than the recording's medium background noise, no matter what. It's negligible. This is seldom the case, but that aspect can be taken into account afterwards if necessary. All equipment noise except which of the recording medium is also ignored.

First case: let's imagine that the signal features a very low crest factor. There are no peaks, and the signal is very stable. I record it onto the medium. I increase the input level until the medium's upper limit as described is reached. I write down the input level setting. Let's write it as U1 dB. Now I decrease the input gain, until the signal is barely audible. At this point, it gets progressively masked by the medium's background noise. This input setting will be written as D1 dB. For this medium, for this signal, the available range, or whatever else you want to name it, is U1-D1 dB.

Second case: let's imagine that the signal features a very high crest factor, with very salient, short peaks in an otherwise stable context. The maximum input level, before the medium's upper limit is reached, is U2 dB. The minimum input level, where the sound begins to be masked by the medium's background noise, is D2 dB. The available range is U2-D2 dB.

Doing that, I will find that U1-D1 > U2-D2. Why is that?

Signal 2's peaks will reach the medium's upper limit while the majority of signal 2's content, or waveform, as you wish, will be much below. If signal 1's crest factor is C1 dB, and signal 2's is C2 dB, when the medium's upper limit is reached, then the RMS of signal 1 will be roughly C1 dB below this limit. As for signal 2's RMS, it will be roughly C2 dB below the upper limit. I said "roughly". This means that, approximately, U1+C1 = U2+C2. I suppose someone will rightly point out that I don't have the right to write this from a mathematical perspective, because a gain and a crest factor are not the same notion, but if you do the experiment yourself, you will indeed find that U1+C1 = U2+C2.

At the other end of the range, we have approximately D1 = D2. While signal 2's peaks will not be masked, the rest of it will be, and the majority of signal 2 will begin to be masked by the background noise near the same input gain as in the case of signal 1.

So, on the one hand (signal 1), I was able to use U1-D1 dB, with U1 = U2+C2-C1, which means I was able to use U2+C2-C1-D1 dB, and, since, approximately, D1 = D2, to use (U2-D2) + (C2-C1) dB.
On the other hand (signal 2), I was able to use U2-D2dB.
Since C2 > C1, C2 - C1 > 0. It means U1-D1 > U2-D2.

Conclusion, with signal 1, I was able to use more "range" or whatever people call it, than with signal 2.

This is the first part of it - Hugh, are you OK with this part at least? If so, I proceed to the second part.

Edited by Jennifer Jones (30/09/11 11:07 AM)

okay - yes. But the uninteresting fact here is that the dynamic range of U2-D2 is greater than U1-D1.
U1-D1 > U2-D2 isn't some great revelation, it's obvious .

It would be an issue if the dynamic range of the recording medium was 10dB but you're piece of audio to be recorded had a wholly normal 30dB dynamic range. If it was you'd compress, limit or otherwise distort the 30dB range to get it into the 10dB useable audio range. However even the most basic recording medium performs far better than that!

But okay, let's agree you've set up the basics of your premise.

Edited by Jennifer Jones (30/09/11 11:07 AM)

ah come on, I'm moving cautiously, because I want to be sure to be understood
it is a delicate topic, and even you told me that I should be extra careful about formulation
that's what I do

Edited by Jennifer Jones (30/09/11 11:07 AM)

okay. yes. I'll shut up until you've finished!

Edited by Jennifer Jones (30/09/11 11:07 AM)