One-Synth-Challenge: General discussion thread

[schiing]

Oh, I'd say zarf is definitely speaking the truth. I'm moving outside the pure scope of OSC here, but consider classical music, with a far more dynamic expression - I guess that would be the most obvious example (https://www.gearslutz.com/board/remote- ... dness.html).

Well you are talking about loudness range (loudness variation of your entire track LU) and exponent1 and I are talking about overall loudness (average loudness measured over entire duration of material LUFS).

No, sorry. Definitely talking about loudness measured over entire duration. A slow movement with lots of pianissimo will measure entirely differently than a loud scherzo. If you tried to align them using LUFS alone, the volume difference between the two files simply wouldn't match.

[exponent1]

Oh, I'd say zarf is definitely speaking the truth. I'm moving outside the pure scope of OSC here, but consider classical music, with a far more dynamic expression - I guess that would be the most obvious example (https://www.gearslutz.com/board/remote- ... dness.html).

Well you are talking about loudness range (loudness variation of your entire track LU) and exponent1 and I are talking about overall loudness (average loudness measured over entire duration of material LUFS).

No, sorry. Definitely talking about loudness measured over entire duration. A slow movement with lots of pianissimo will measure entirely differently than a loud scherzo. If you tried to align them using LUFS alone, the volume difference between the two files simply wouldn't match.

https://artists.spotify.com/faq/masteri ... is-it-used. This is all I'm asking for/talking about.

People can go ahead and make a -5 LUFS track if they can and want, but since SoundCloud has no volume normalization, it's quite painful(in the OSC context purely) most of the times when you suddenly play a loud track after quieter ones. It's unfair to both the listener as well other artists if loudness becomes the basis of judgment.

[schiing]

Absolutely, totally got that - and I agree. This other thing we're ending up discussing is a bit of a sidetrack.

[TrojakEW]

No, sorry. Definitely talking about loudness measured over entire duration. A slow movement with lots of pianissimo will measure entirely differently than a loud scherzo. If you tried to align them using LUFS alone, the volume difference between the two files simply wouldn't match.

In post you send there is no mention about loudness and music style. It is still about destination platform. Also it looks like you miss one important word in my post (average loudness measured over entire duration of material LUFS).

So we have all together:
Momentary Loudness - measures the loudness of the past 400 Milliseconds
Short Term Loudness - measures the loudness of the past 3 Seconds.
Integrated Loudness - average loudness measured over entire duration of material
Loudness Range - loudness variation of your entire track

Please provide any link that are related to different loudness based on music style. I was not able to find anything about it but if something like this exist I will like to learn more about it. I'm still learning about volume normalization.

[schiing]

Yeah, sorry, Trojak. I didn't do a lot of research before posting the link - it was probably not the best of examples. I'm feeling lazy, so I'm afraid I will have to leave it to you to find more appropriate documentation

My point was to indicate how loudness considerations differ between pop and classical music.

I do indeed understand your point about average loudness - and that is what I have been talking about all along.

Now, as exponent1 pointed out, we're moving outside of the current topic. Adding that to the fact that I'M NOT A GREAT TUTOR, and I'm still unsure as to why we keep talking past each other, I probably shouldn't. But I will try to explain my point once more (sorry if I spell things out too much - I'm just trying to be as clear as possible):

Consider the two files I mentioned in my previous post:

Now, say you record these two symphonic pieces - a quiet one with only a few crescendos, and another loud one, with lots of peaks and dynamic content.

We keep the recording level even during the recording session, and well below peak level; and we get two clean files that are naturally aligned to each other in terms of volume. Now, if we move these files to the audio editor untreated and measure the average loudness of each, the quiet one might clock in at -23 LUFS and the loud one at -18.

Treating these files further, we want to make sure we keep the difference between them. Normalizing the quiet one to -18 and keeping the loud one unchanged would create an unnatural result from a listening perspective, especially so if these are two movements of a symphony that immediately follow each other. Okay.

Now, let's say that these *are* in fact two parts of a symphony. We want to think of and measure them as a whole. Let's do that: We find an average of -20 LUFS between the two of them. Working with both files together, we find that we can raise the average volume between them to -18, say, a bit below peak level.

Of course, we could do a little bit of compression - but compressed snares sound way better than compressed symphony orchestras, and there's simply no way we're ever getting close to a pop standard loudness average of -12 or -10 with all these quiet passages dragging our average down - causing what might be labeled a genre-specific variation to the average loudness/LUFS level.

Let's take it one step further. Let's say we want to add a disco-beat to the slow piece, Raymond Lefèvre-style! If we do this à la 2019 (I know, Lefèvre '19!) we sidechain the devil out of the strings with the kick, highpass the lower string frequencies and turn up the string volume until they fill the entire gap between ∞ and peak level between the beats on the output file - we'll be at -6 LUFS in no time.

If, however, we want to do it the way of the mighty Steely Dan of 1980, we want to keep the strings at a fairly low volume, at a level that would be natural for a pianissimo string section to be if they shared a room with a drum kit. In this instance, we never want to fill the entire gap between ∞ and peak level - we want to keep the space. So in the end we might end up at -15/-16 LUFS. Again, two examples of what might be described as genre-specific variations.

And, just to be clear - all of these output files can of course always be adjusted to similar average loudness/LUFS levels for compilation purposes later on, according to your chosen specifications - even if the thinking behind them is completely different. But the strings will of course always suffer if you push them too hard.

I do hope this gets my point across. And I'm afraid that if you still don't think I understand what you mean when you say "average loudness" (or even worse, that you think I'm wrong! ), we'll simply have to accept that some things just aren't meant to be - and leave it at that. As I said earlier, (and my children will attest to this) I'm a well-meaning, but ultimately very bad tutor!

[zarf]

Man, sometimes this place is educational Thanks Schiing for such an informative post. I would need to see far more evidence to support your assertion that you are a very bad tutor!

[Compyfox]

Okay, sorry... I'm not an OSC participant - but I see the same questions pop up for the Songwriting Competition that I run on the Mix Challenge audio community.

In fact, the rule set encourages to not render (distribute) higher than -14LUFS ILk - because I'm one of those people that thinks that the Loudness War is utter bollocks, and you do not need to compress everything to to it's limits in order to get an impactful sound. You do that through the right FX usage (which is not a thing for the OSC, to my knowledge).

Treating these files further, we want to make sure we keep the difference between them. Normalizing the quiet one to -18 and keeping the loud one unchanged would create an unnatural result from a listening perspective, especially so if these are two movements of a symphony that immediately follow each other. Okay.

What you're currently talking about is not "Loudness Normalization for individual files", but "Loudness Normalization for a whole album". So that all tracks get adjusted relative to their original individual loudness value/balance. In case of the OSC, you have individual entries however, so it would mean "individual files".

Soundcloud does not(!) offer Loudness Normalization (still). Also, the louder you push your material - because you were told for decades to do so, the worse the quality is getting on the platform due to the used CODEC and ceiling limitations (MP3 starts to drastically fall apart at -7dB RMS already!!!)

If, however, we want to do it the way of the mighty Steely Dan of 1980, we want to keep the strings at a fairly low volume, at a level that would be natural for a pianissimo string section to be if they shared a room with a drum kit. In this instance, we never want to fill the entire gap between ∞ and peak level - we want to keep the space. So in the end we might end up at -15/-16 LUFS. Again, two examples of what might be described as genre-specific variations.

The beauty of LUFS meters (especially SLk - I prefer to use the realtime meter to get me in the ballpark for the "whole stream" ILk readout), is that it's basically independent from the program material "density". Modern day Pop music is way more dense mixed than classic rock or orchestra. So if you'd treat each individual track by "genre" and Dynamic Range or Loudness Range (two completely different things!), rather than from a pure technical aspect, you will never(!) find a common ground.

This is why the DR-Meter failed IMO. The ambition was high and directed at "let's get back to more dynamic content", but the target values fluctuated way too much, because it was program material dependent.

LUFS meters ignore that.

A Pop Prouduction at -16LUFS (Mastered for iTunes or Spotify) works nicely compared to a 20minute orchestra piece at the same LUFS value. The productions can still be dynamic, have fluctuating loudness, but the overall "stream" (listening from Pop track to orchestra to techno, back to pop), there is way less reach for the volume knob.


You still have to differentiate between an album, and individual tracks though.



So my suggestion for the OSC is - if you adapt a Loudness rule - maybe go this way (copied from my rule set):

try to be reasonable with mastering (recommended: -14LUFS ILk or -12dB RMS avg/K-12 max)

I think this is where the "average" confusion currently comes from. Because we talk "average signal strength" - what some users commonly call "Perceived Loudness".

ILk is a whole measurement mechanic (average mathematical value of various measurement points throughout a whole program stream - or "Loudness Measurement - filtered - over time"), while RMS Realtime is where you have to find your personal "average" by having an eye on the meter (RMS min and RMS max is not helpful). In case of the K-System: If your track is made in K-12, doesn't exceed -8dB RMS absolute max, then the "average" of the program material (mezzoforte passages) can be around -14dB RMS to -11dB RMS.

I'm fairly certain this is where all this confusion is coming from.



So I'd say "keep it simple".

Maybe set up a rule like: "Grab (for example) the free Youlean Loudness meter, do whatever you want in your host, but try to not go higher than -14LUFS +-2LU (LUFS meters are best used for Mastering purposes!)".

Those that "mastered loud" will be in shock, because "possible lost impact due to the "loud effect" instead of getting back transients for drums", but this might help the quality of the game overall.

If you have signatures activated, I think I have a KVRmark where I explained "Manual Loudness Normalization", and I've talked about these topics over and over and over. Try it in June 2019 - make it clear in the initial post, see how many people will actually adapt it, then go from there.



Regarding the "vote system".

The "favor votes" can be suppressed if really only the participants vote for each other, indeed. However - since you're using a third party engine, you'd need to set up limitation so that only the votes by the actual participants count. Whether or not that is technically possible (you'd need to synchronize participant names, etc) - I don't know. Only the person that created the script can answer this. Plus - there is the topic "data collection".




Whatever you come up with rule wise, and try in a future game - good luck.

[schiing]

Treating these files further, we want to make sure we keep the difference between them. Normalizing the quiet one to -18 and keeping the loud one unchanged would create an unnatural result from a listening perspective, especially so if these are two movements of a symphony that immediately follow each other. Okay.

What you're currently talking about is not "Loudness Normalization for individual files", but "Loudness Normalization for a whole album". So that all tracks get adjusted relative to their original individual loudness value/balance. In case of the OSC, you have individual entries however, so it would mean "individual files".

Well, I understand how it might seem like I do.

But I was really only using this scenario to display how the overall loudness average will be different depending on whether we consider these fictional "movements" as two separate units or as a single unit - to point out, for instance, that you'll have a rough time trying to get your prog track with a 4 minute buildup to -8 LUFS. Or that you usually want to keep down the mastering volume on material that was recorded and/or played at a low level in relation to what it's supposed to be surrounded by. Example through exaggeration: In the OSC, you probably don't want to master your experimental masterpiece of a subtle duck pond at the same level as the tracks from our excellent colleague, Unknown Challenger - people would feel the ducks crawling through their brain!

I'm sorry if I was unclear about that.

[schiing]

Man, sometimes this place is educational Thanks Schiing for such an informative post. I would need to see far more evidence to support your assertion that you are a very bad tutor!

Hey, stop pushing me, man! Nah, thanks a lot for the vote of confidence!

[aciddose]

Yes, they are. I don't know if you've participated in OSC before, but every OSC result post has a google spreadsheet published for everyone else to see. As an example, this is the final voting sheet from the last OSC https://docs.google.com/spreadsheets/d/ ... I0jKvGoorc

Ah right... yes I've seen those. I mean they don't seem to be clearly indexed from anywhere and I was having a hard time finding older ones.

Then yes, it would be a tiny bit of extra work for those counting the votes but they could simply keep two tallies and then add them together for the total score. It would then be possible to assign prizes to any of the three categories: popular, participant or total score. Of course it would be easy to keep using the total score and provide the separate tallies only for education.

[aciddose]

re: loudness and variation by genre

Yes, this is why the "rule" is for limiting MAX LOUDNESS. If the rule is -12 dB RMS or -14 dB LUFS it doesn't mean you need to take the recording at -18 and compress it or add gain to clip off the peaks until you get the target level. The "target" is actually a limit rather than something you want to ensure. If you end up with a target (maximum) of -14 dB LUFS and you measure your track at -13.5 it means you need to decrease the peak gain slightly or reduce the compression to get back below -14. If you want to try to get comparable loudness from your track at -21 dB you don't need to add gain or compress to get up to -14 to qualify the track, it's already below the limit... but you can adjust your mix to get the loudness you're looking for up to that point.

Such a rule always needs the qualifier "for each peaking section of the track", "not counting fades"; since if you use near silence half the time you can double the loudness for the non-silent portions. That type of cheating has diminishing returns though and generally if your track is mixed in a normal way you'll measure about -12 dB RMS anyway... since orchestra with long fades will measure much lower and there is no genre I'm aware of that uses silence mixed with maxed out sections. So such a qualifier only exists to be applied selectively when you get a track designed (intentionally or otherwise) in an unusual way to contain bursts of extreme loudness.

Remember that these rules in broadcast were a reaction to commercials/advertisements being compressed to get maximum loudness which typically meant that if you were watching a scene with conversation that suddenly cut into a commercial, the voice yelling "FOR ONLY NINE PAYMENTS OF 9.99 YOU GET THIS SET OF EVERSHARP KNIVES, BUT WAIT, THERE'S MORE!!!" might be 100x louder.

So the rule was only to limit the loudness difference, not to eliminate it completely. If the show/film you were watching faded to silence or crickets at night (-34 dB) you could expect the vacuum commercial that came on to be 10x louder at -14 dB.

[Compyfox]

Yes, this is why the "rule" is for limiting MAX LOUDNESS. If the rule is -12 dB RMS or -14 dB LUFS it doesn't mean you need to take the recording at -18 and compress it or add gain to clip off the peaks until you get the target level. The "target" is actually a limit rather than something you want to ensure. If you end up with a target (maximum) of -14 dB LUFS and you measure your track at -13.5 it means you need to decrease the peak gain slightly or reduce the compression to get back below -14. If you want to try to get comparable loudness from your track at -21 dB you don't need to add gain or compress to get up to -14 to qualify the track, it's already below the limit... but you can adjust your mix to get the loudness you're looking for up to that point.

Why make it so complicated?

Look, the main problem here is that SoundCloud doesn't do "Loudness Normalization on playback". In fact, it does not do any Loudness Normalization at all. If it would do so, then tracks that are pressed it their limits are pulled down to a certain "target level" (while lacking all transients), and all tracks that are below the target level are being pulled up (while only clipping away the transient peaks - if there are any - which is way less damaging). In this case, the argument about "too loud mixes" wouldn't exist. But alas, it does.

I checked in on the rule set of the OSC again, and there is actually no mention that "Mastering" (as in Loudness Boost broad tone-shaping on the summing bus) is allowed. So the main problem here seems to be, that people seem to do two things:

a) just mix however they please, slap a "safety limiter" on, and be done (as in: ignore proper gain staging and a certain signal headroom - yes, people still do that, thanks to >32bit float mix engines)
b) master even though they're not allowed to do so


Then make the rules simple:
Use a reference level, do not master your audio material

As reference, you could either use -18dB RMS (avg) or 0VU. That is basically how all analog type plugins work their hotspots at.




"but sir, this is too quiet, there is no impact form what I want in my production"

This is a Challenge to create something out of "one synth" right? The challenge is to also not use/abuse external plugins (at least to my understanding)? So the focus is to let the synthesizer in question speak for itself. And that works best if you do not(!) perform any mastering or loudness boost.

I've written endless posts about Gain Staging (again - if you have signatures activated, see my signatures for KVR Marks). So keep it simple.



If you want to introduce a new rules, to keep the loudness differences fairly in check, then say:
"We recommend/encourage people to mix at a suitable work level - for example the infamous -18dBFS = 0VU = -18dB RMS (avg)"

If you're using a K-System meter, that's anything between -20dB RMS and -16dB RMS (aka K-20). For LUFS, that can vary between -21LUFS and -18LUFS SLk (avg). Depending on how busy the mix is. Nothing will be squashed, and no extensive safety mechanisms (mastering) are needed.


There - simplified. Although still but a recommendation and not a strict "of you don't adhere to this, you're disqualified" rule.





Fixing the "favor votes" from non-participants is a whole different can of worms.

[exponent1]

Yes, this is why the "rule" is for limiting MAX LOUDNESS. If the rule is -12 dB RMS or -14 dB LUFS it doesn't mean you need to take the recording at -18 and compress it or add gain to clip off the peaks until you get the target level. The "target" is actually a limit rather than something you want to ensure. If you end up with a target (maximum) of -14 dB LUFS and you measure your track at -13.5 it means you need to decrease the peak gain slightly or reduce the compression to get back below -14. If you want to try to get comparable loudness from your track at -21 dB you don't need to add gain or compress to get up to -14 to qualify the track, it's already below the limit... but you can adjust your mix to get the loudness you're looking for up to that point.

Why make it so complicated?

Look, the main problem here is that SoundCloud doesn't do "Loudness Normalization on playback". In fact, it does not do any Loudness Normalization at all. If it would do so, then tracks that are pressed it their limits are pulled down to a certain "target level" (while lacking all transients), and all tracks that are below the target level are being pulled up (while only clipping away the transient peaks - if there are any - which is way less damaging). In this case, the argument about "too loud mixes" wouldn't exist. But alas, it does.

I checked in on the rule set of the OSC again, and there is actually no mention that "Mastering" (as in Loudness Boost broad tone-shaping on the summing bus) is allowed. So the main problem here seems to be, that people seem to do two things:

a) just mix however they please, slap a "safety limiter" on, and be done (as in: ignore proper gain staging and a certain signal headroom - yes, people still do that, thanks to >32bit float mix engines)
b) master even though they're not allowed to do so


Then make the rules simple:
Use a reference level, do not master your audio material

As reference, you could either use -18dB RMS (avg) or 0VU. That is basically how all analog type plugins work their hotspots at.




"but sir, this is too quiet, there is no impact form what I want in my production"

This is a Challenge to create something out of "one synth" right? The challenge is to also not use/abuse external plugins (at least to my understanding)? So the focus is to let the synthesizer in question speak for itself. And that works best if you do not(!) perform any mastering or loudness boost.

I've written endless posts about Gain Staging (again - if you have signatures activated, see my signatures for KVR Marks). So keep it simple.



If you want to introduce a new rules, to keep the loudness differences fairly in check, then say:
"We recommend/encourage people to mix at a suitable work level - for example the infamous -18dBFS = 0VU = -18dB RMS (avg)"

If you're using a K-System meter, that's anything between -20dB RMS and -16dB RMS (aka K-20). For LUFS, that can vary between -21LUFS and -18LUFS SLk (avg). Depending on how busy the mix is. Nothing will be squashed, and no extensive safety mechanisms (mastering) are needed.


There - simplified. Although still but a recommendation and not a strict "of you don't adhere to this, you're disqualified" rule.





Fixing the "favor votes" from non-participants is a whole different can of worms.

Mister Fox, there is no rule against mastering in OSC. There is a rule that all the mastering has to be done within the same DAW where the song was produced and no external DAW/software or mastering service is to be used for mastering. https://sites.google.com/site/kvrosc/rules

OSC rules are fairly defined otherwise and as such I'm fine with all of them.

I merely asked for the loudness rule/guideline since we listen to the entries in a SoundCloud playlist and uneven loudness across tracks, especially the ones which start off with a bang(for example our excellent colleague Unknown Challenger and a few others) causes a decent amount of discomfort for the listeners.

That's all, the rest of the loudness technicalities are not important for this competition. The guideline is simple, just keep it reasonably loud and don't cause discomfort for the listeners.

[Compyfox]

Since SoundCloud does not(!) have this feature yet (loudness normalization), then you can't do anything about it.

A guideline/recommendation would work - but if nobody wants to adapt it, then you can't do anything about it either.


Either way - I think at this point, we're walking in circles with this topic

[aciddose]

Why make it so complicated?

My description made it simple. If that's to be considered "complicated" then well, I'm sorry you feel that way.