Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Pleasurize Music Foundation (Read 60337 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Pleasurize Music Foundation

Reply #25
The English manual is up now.
It seems to me that they got the translation of the critical part badly wrong:
Quote
In order to determine the official DR value, a song or entire
  album (16 bit, 44.1 kHz wave format) is scanned. A
  histogram (loudness distribution diagram) is created with a
  resolution of 0.01 dB. The RMS – an established loudness
  measurement standard – is determined by gathering
  approximately 10,000 pieces of loudness information within
  a time span of 3 seconds (dB/RMS). From this result, only
  the loudest 20% is used for determining the average
  loudness of the loud passages.

But according to their German manual, the histogram has 10000 "drawers" (so 0.01 dB * 10000 = 100 dB ~ Dynamic range of 16 bits).
Actually, if my tests are right, there are only ~1000 RMS values calculated, one each 3 seconds...
Also they seem to have some kind of pop/tick protection, although in the manual they are talking about cases where you should take the DR values on a song-by-song basis and average them.

I'd like to have a look at the complete specifications that they promised to release.

Pleasurize Music Foundation

Reply #26
I'm not sure I completely trust either manual.  Their English is bad enough that the meanings of words are sometimes inverted (see one of the download links, which says "until March 2009", but is clearly meant to go live *from* that month), and my German is not good enough to easily read the German manual.  (I probably could if I really tried.)

However, using some educated guesses, I think I can see what they're doing:

- Oversample the entire file and look for the absolute peak sample in the result.
- Take a large number (this is probably where the 10000 comes from) of overlapped 3-second BS.1770 power readings, sort them, and average the top 20% of these.
- Subtract the latter from the former, and round to nearest whole decibel.

There are a lot of important details missing from that summary, the most critical being whether the averaging is done in voltage space, power space or decibel space.  Assuming they know what they're doing - which seems to be the case - I would confidently guess decibel space.

I think the "10000" number comes from taking a reading every half-second on an 80-minute disc.  That is slightly below 10000 readings in total, and is a perfectly plausible way of doing it.  On a shorter disc or a single track, there would be correspondingly fewer readings taken, still half a second apart.  Alternatively, it could be one reading every second per channel.

They say that the reading should normally be taken on a single file covering the entire disc.  However, if the tracks have wildly differing characteristics (legitimately on a compilation disc), they offer an alternative way to calculate an album score from the individual track readings.  This would give a substantial bonus to compilations which have one clippressed track and many sane ones, for example; it would not benefit an album which merely has some stray clicks in it.

If I've guessed it right, I'm actually quite impressed with the simplicity and effectiveness of their system.  I could probably code up an emulation of it to see if I'm sufficiently right.

The main concern I have is that an album recorded at a low level with a few stray clicks on it would have an abnormally high reading.  As such, if they catch on sufficiently that publishing the highest possible DR score becomes fashionable, it is possible to "game" it - the highest possible score, which would have to be generated synthetically, is likely to be in the 90s.  Any good mastering engineer would, I hope, strive to avoid or remove clicks, so this *shouldn't* be a problem, but I'm just not entirely satisfied with a fragile system.

Pleasurize Music Foundation

Reply #27
Continuing:

I've run their offline tool against a subset of my test archive.  This is on a per-track basis, which is not how they recommend, but it's still instructive.

Let's start with the good news: anything recorded without compressors (Bach, Berlioz, Leftfield) is very consistently in the 11-14 dB range.  There are exceptions in the form of Wachet Auf, which is predictably a little lower (due to the timbre of the only instrument), and one track of the Berlioz, which is all the way up at 17 dB (!) due to the relatively few, brief, yet dramatic climaxes in it.  (Berlioz has downright extreme long-term DR, which seems to have caused some difficulty for the recordist on the final, loudest track.)

Now for the bad news: RHCP "Californication" consistently gets a per-track score of about 4 dB, with at least one stuck at 3 dB.

Wait, did I say that was bad news?  Bad scores, yes, but very good news - since the tool has successfully identified a classic offender from a sea of excellence.

And now for the really bad news.  I tried to use it on some files from a game's soundtrack.  It refused to process them on the grounds that they were 48kHz rather than CD format.  So I would have to resample them before I can measure them.  There's no reason for this restriction that I can imagine, except for cutting down on the test regimen.  The authors specifically state that they do not use any psychoacoustic filtering, which would occasion the algorithm being sample-rate specific.

I "treated" my dad to one of the worst RHCP tracks.  I asked him: "Can you even tell the difference between the verse and the chorus?"  He is competent in singing himself, though more likely 16th century than 21st century music.  Then I contrasted it with some 1995 Leftfield, noting that with the compressors and limiters removed, there was actually room to put things in the background of the music.  He agreed completely, and this was with his mid-50s ears and a huge pile of my computer fans whirring away.

Pleasurize Music Foundation

Reply #28
They use plain vanilla RMS (no BS.1770) on 3-second non overlapped windows if my implementation is right. The 10000 refers to the "Schubladen" (drawers) of the histogram.

Their result for Underworld - A Hundred Days Off:
DR 8
Dynamic Range: 7.8 (left), 7.7 (right)
Peak: -0.17, -0.17
RMS: -10.1, -9.9

This is what I got:
DR 7.786151716671393
Dynamic Range: 7.824601544286622, 7.747530909009571
Peak: -0.16941573141763458, -0.16941573141763458
RMS: -10.06189178486872, -9.909813233507633


Or for Depeche Mode - Playing the Angel:
DR 4
Dynamic Range: 3.6, 3.8
Peak: over, over
RMS: -6.8, -6.9

vs
DR 3.7152858476665918
Dynamic Range: 3.6097436552998055, 3.819560853398123
Peak: -2.6504994725949876E-4, -2.6504994725949876E-4
RMS: -6.846427797626369, -6.884196252943992


I have tested it on many more albums, and the values match up.

Pleasurize Music Foundation

Reply #29
That's bad news, then.

This should be testable by taking a sample, speeding it up by 0.1%/1%/10% etc, and observing by how much the final numbers change. This should be a more prominent effect for higher amounts of dynamic range in the music to start with.

It also overestimates the loudness of low frequency content, which is a good move for being hard on modern hypercompressed masterings, but is harder to defend on psychoacoustic grounds.

EDIT: Gawd, that was fast - I just found a semi-realistic example of this. Current killer sample for speed variation is the Bach's Duetto No. 1 in E Minor by Christophe Rousset on harpsichord. Normal WAV is DR12; after speeding it up 1% via Audacity resampling, it's DR11. (Slowing it down 1% also yields DR11.) The decimal L/R DR values only changed by 0.1db, but that was all that was needed to knock the rounding over. And if it can work to reduce the DR value it can certainly also work to increase it. This is at least a proof of concept that skullduggery might be employed to cheat the meter, and I would strongly suspect that overlapping the windows would fix it.

Ultimately, this means that DR values should only be considered accurate to within +-1 without more info, and given that the difference between "white noise" and "barely acceptable pop" is 7, that implies a significant lack of accuracy.

re the manual: "10,000 pieces of loudness information within a time span of 3 seconds" - yeah, this is pure fantasy if interpreted literally, it implies a block length of 13.23 samples.

The util also bails out on 24-bit 44.1k WAVs too - not really that big of a deal, but it's kind of a hassle. I'm sure (or I hope) they'll fix it later.

Quote
Wait, did I say that was bad news? Bad scores, yes, but very good news - since the tool has successfully identified a classic offender from a sea of excellence.
Heh. I could have told you the results of that test without even knowing what was being tested.

Another test would be to run it between two different masterings of the same record that we absolutely know are significantly different in terms of compression - eg, the Mudcrutch vinyl vs CD masters, or Icky Thump vinyl vs CD - and see how much of a difference exists. Or apply some light compression to an existing track and see how much is necessary before the DR value changes. Those things help establish sensitivity.





Pleasurize Music Foundation

Reply #30
Version 1.2 is up. The offline tool is still the same, though. And they have not fixed the English manual. They should release the source for the algorithm.

It also overestimates the loudness of low frequency content, which is a good move for being hard on modern hypercompressed masterings, but is harder to defend on psychoacoustic grounds.

Isn't the exact opposite the case? The current algorithm will assign a lower DR value to a recording with more bass. This is bad behaviour as bass is the first thing that suffers from hypercompression. An example: SND's first album (RG of -0.59 dB and very dynamic) is just a DR7!

Pleasurize Music Foundation

Reply #31
Good effort but I doubt it will gain any traction. Remember that sound engineers are there to perform service at the request of the artist. Moreover, certain types of music are just meant to sound, well, loud. Given how the Internet is giving rise to a lot of independent labels and musicians who do their own mixing and mastering, to expect everyone to follow the same standard will be virtually impossible. I applaud the effort, however I doubt it will ever take off.
Be healthy, be kind, grow rich and prosper

 

Pleasurize Music Foundation

Reply #32
Version 1.2 is up. The offline tool is still the same, though. And they have not fixed the English manual. They should release the source for the algorithm.

It also overestimates the loudness of low frequency content, which is a good move for being hard on modern hypercompressed masterings, but is harder to defend on psychoacoustic grounds.

Isn't the exact opposite the case? The current algorithm will assign a lower DR value to a recording with more bass. This is bad behaviour as bass is the first thing that suffers from hypercompression. An example: SND's first album (RG of -0.59 dB and very dynamic) is just a DR7!


Right, that's exactly what I'm saying - but on top of that, that is not accurate. The reason bass suffers after aggressive mastering techniques is because it is simply not as audible as other frequencies. The overall loudness of the track simply does not change all that much from simple removal of bass. As an eq choice it tends to coincide with hypercompression, but the lack of a direct correlation between the two means that this bass overemphasis slightly distorts the meaning of DR comparisons across different mastering styles and music genres.

For instance, this means that compression differences in classical music (with very little bass) might make a less substantial DR change than with rock music; even if a classical piece is hypercompressed, the lack of effect on the bass signal means that the same amount of compression does not go as far with manipulating the DR figure.

Pleasurize Music Foundation

Reply #33
Just noticed yesterday that 1.3 is out. The offline tool has an option to scan entire folders now and can output results to a log file. Still only supports MP3 and WAV though, and they haven't mentioned any changes in implementation.

Pleasurize Music Foundation

Reply #34
Unfortunately, it is no more free available for download.
Are there any alternatives?

Pleasurize Music Foundation

Reply #35
Given the above commentary on TT's algorithm, if all you want is offline analysis, pfpf will probably tell you what you need to know. But it was never really meant for prime time and considerable care is necessary to interpret its results (although the same must be said for any dynamics estimator, really..)

Pleasurize Music Foundation

Reply #36
They've really shot themselves in the foot by restricting access to their meter. They'd rather a donation, than get people to use their meter. Desperate administrators always act the same way (get the money in and then we can spread the message). The thing that was spreading the message is the thing they've just locked away. 

I'm still hopeful that Chromatix and Axon will produce a meter that combines their different approaches and yields 3 results: 2 individual scores and one combined score.

C.
PC = TAK + LossyWAV  ::  Portable = Opus (130)

Pleasurize Music Foundation

Reply #37
Given the above commentary on TT's algorithm, if all you want is offline analysis, pfpf will probably tell you what you need to know. But it was never really meant for prime time and considerable care is necessary to interpret its results (although the same must be said for any dynamics estimator, really..)

Well, I've just tried pfpf, looks good, but it is a bit too... big for my laptop screen. I was rather thinking about something like command line tool?  Sorry for being oldschool : )

Pleasurize Music Foundation

Reply #38
They way of acting was since my first contact (about 2-3 Months away) like "give money, kk?" - i asked them, why their dr meter does not provide support for flac.
The answer was: Many formats, many different problems - you could help by provide a amount of cash.
The other question, why not spread the word in the "base" (us freaky little music lovers), instead of the "head" (music engineers, e.g.) was little "arrogant"; To reach the important people.

After a namecalling in a known german magazin (c`t - www.heise.de), they went probably totaly mad.
It´s not like their board is about to collapse, they just earned some "fame" in places for their work (in fact most for the database at dr.loudness-war.info, who supports FLAC Sources).

The active membership provides:
Passwortgeschützter Bereich mit Zugang zu diesen exklusiven Vorteilen für Active Member (ab 01.07.09):
Password secured member-access, with exclusive Gadgets for active members (since 01.07.09):

· Freie Nutzung von Software Updates (TT Dynamic Range Meter Plugin und TT DR Offline Meter Software)
Free access to software updates (TT Dynamic Range Meter Plugin and TT DR Offline Meter Software)

· Lizenzfreie Nutzung des DR-Logos (für CD Booklets und Online Shops)
License free use of the DR icons (for CD Cover / Booklets and Online Shop)

· Freie Nutzung der in Kürze gelaunchten DR-Online-Datenbank mit der Möglichkeit,
  Ihre eigenen Releases einzupflegen, damit Musikkäufer Ihre dynamischen Veröffentlichungen finden

Free use of the DR Database, which will be launched in the next time, provided with the possibility of adding your own releases, so that customers can find your releases

· Freier Zugang in eine wachsende Wissensdatenbank, in der Doktor-Arbeiten und Studien rund um
  das Thema Loudness gesammelt werden

Free access to a knowledge base, which will provide studies and dissertation about the loudness war.

· Active Member Newsletter mit Hintergrundinfos
Active Members newsletter with backgroundinformations

· Listung als Active Member auf der Website (Tabelle > view list)
You´ll get a cookie.

Better than google, i hope.

Pleasurize Music Foundation

Reply #39
They've really shot themselves in the foot by restricting access to their meter. They'd rather a donation, than get people to use their meter. Desperate administrators always act the same way (get the money in and then we can spread the message). The thing that was spreading the message is the thing they've just locked away. 
That's not exactly the criticism I think is valid here. I think they are trying to make what they have self-supporting. What this tells me is, and note I am completely speculating here, they have people working on this, either full-time or part-time, and their initial developer sponsorship (Algorithmix) might have pulled the plug on large-scale financial assistance. If they want to keep the momentum going, they have to find some way to monetarize what they have, even if they are a 503© (which are pathologically well audited in the US so I tend not to have any doubts about the veracity of their nonprofit status).

Quote
I'm still hopeful that Chromatix and Axon will produce a meter that combines their different approaches and yields 3 results: 2 individual scores and one combined score.

Unfortunately, you're going to have to wait a long time on me for that. Part of that is that I've been occupied by other audio pursuits recently, but there's a larger issue afoot here....

Quite simply, there is no testing or validation model for the accuracy of a dynamic range meter. TT has its number, and pfpf has its numbers, and the Sparkle-Meter has its numbers, and they do not agree except in the vaguest of terms. Until such a model is devised, and is well justified on psychoacoustic and listening-preference grounds, no progress will be made beyond what we already have.

Because of this, I am frankly skeptical of the DR databases like on dr.loudnesswar.info, or even some people who have wanted to do the same with pfpf. People are incorrectly looking at these numbers as if they had a shred of authoritativeness to them.

So if/when I revisit this it will be on the QC side of things and not on the actual implementation side of things. Incidentally, I stated almost exactly this intent in my registration entry to Pleasurize when their site first started. No response.

Well, I've just tried pfpf, looks good, but it is a bit too... big for my laptop screen. I was rather thinking about something like command line tool?  Sorry for being oldschool : )
What, you mean nobody else has a 1920x1200 display on their laptop like me?

They way of acting was since my first contact (about 2-3 Months away) like "give money, kk?" - i asked them, why their dr meter does not provide support for flac.
The answer was: Many formats, many different problems - you could help by provide a amount of cash.
The other question, why not spread the word in the "base" (us freaky little music lovers), instead of the "head" (music engineers, e.g.) was little "arrogant"; To reach the important people.
I can't really blame them for this; I think the base has been pretty pumped up already. It's a much harder problem to make headway with the people actually producing the music. You essentially have to tell them how to do their own jobs.

But like I said, to put things mildly, that gets harder if your statistics are potentially faulty.

Quote
After a namecalling in a known german magazin (c`t - www.heise.de), they went probably totaly mad.
It´s not like their board is about to collapse, they just earned some "fame" in places for their work (in fact most for the database at dr.loudness-war.info, who supports FLAC Sources).
OK, cough it up. You can't allude to a fight with c't without describing it in more detail

Pleasurize Music Foundation

Reply #40
Axon, regarding the meters. I understand that there is no "model for the accuracy of a dynamic range meter", but don't yours and Chromatix give an indication? As such aren't, for example, 5 meters' indications (that may cover each others backs) better than one (as long as they're all making good attempts)? So if all 5 meters give a poor rating (little DR) doesn't that say something? In the same way that one ABX report (I realise ABX is a tried and tested method, but you get the point) is an indication, but 5 individual results are much more of an indication.

Additionally, one doesn't have to use the same methods to produce a synthesised score (i.e. if each could produce an indexed result).

Forgive me if this is naive, I'm out of my depth technically in this area, but very interested in it nonetheless.

C.
PC = TAK + LossyWAV  ::  Portable = Opus (130)

Pleasurize Music Foundation

Reply #41
Quote from: Axon link=msg=0 date=
I can't really blame them for this; I think the base has been pretty pumped up already. It's a much harder problem to make headway with the people actually producing the music. You essentially have to tell them how to do their own jobs.


market follows the demands of the consumer, and the producers make music so loud because the consumers had an wish for it (on it? crap, should take some english lessons) - louder music sounds "better", for many people.
But if you jump over the shark, and the music get toooooo loud, you can´t get go back to normal.
This is what we got with the latest metallica release.
So, if the base would know about the problem and would understand it (some people were serious shocked, as they first seen one of the youtube vids, who provide a sample of the loudness war and his "victims"), the "big" fishes must take a correction of their heading.
The little sound engineers will never rebell against the big corps. If they say no - anyone else will do it. Everyone wants to feed his family and him self.

Quote from: Axon link=msg=0 date=
OK, cough it up. You can't allude to a fight with c't without describing it in more detail


Little translation fault from me - didn´t know "namecalling" was a complete negativ headed word ;-)
they were mentioned in a article - i hope this will fit it better.

Could search and translate it, if you wish

Pleasurize Music Foundation

Reply #42
Any sort of fuzzy logic revolving around a committee of multiple dynamic range estimators is not going to solve the fundamental flaws which are common to all of them.

I know that's vague, but I don't have time to answer this in better detail today.

Pleasurize Music Foundation

Reply #43
Thanks Axon. No, that makes sense. I think I may have underestimated how fundamental the flaws are.

C.
PC = TAK + LossyWAV  ::  Portable = Opus (130)

Pleasurize Music Foundation

Reply #44
I'm sorry to bump this old topic after 4 years, but does anyone know what happen to the Foundation website? Are they gone? Since this morning I can't access their site anymore, there's a german message about the domain, which does not seem to be a temporary issue. Hope I am wrong.

Also coincidentally the site hosting the Dynamic Range Meter plugin for Foobar is unreachable as well: http://www.jokhan.demon.nl/DynamicRange/

Pleasurize Music Foundation

Reply #45
I'm sorry to bump this old topic after 4 years, but does anyone know what happen to the Foundation website? Are they gone? Since this morning I can't access their site anymore, there's a german message about the domain, which does not seem to be a temporary issue. Hope I am wrong.

Also coincidentally the site hosting the Dynamic Range Meter plugin for Foobar is unreachable as well: http://www.jokhan.demon.nl/DynamicRange/


The forum seems to be dead but the foundation website works now 
One thing i don't understand is why they developed a plugin that's not resizable 
If you haven't a full hd screen is unusable, it gets cutted off at the bottom...


Pleasurize Music Foundation

Reply #46
Is there a working link for downloading plugin? Please?
Error 404; signature server not available.

Pleasurize Music Foundation

Reply #47
I couldn't find the original zip file, so I just made a new one with the .dll and .chm files that I have in my fb2k installation: foo_dynamic_range-1.1.1.zip (Dropbox).

Code: [Select]
10b3be7c5dab4b055867588e731ddf79  foo_dynamic_range-1.1.1.zip
c18a6eea356e80a7397f8431fc123cc5  foo_dynamic_range/foo_dynamic_range.chm
65a5392ac0cb21a9bf820518bbead3ee  foo_dynamic_range/foo_dynamic_range.dll

Pleasurize Music Foundation

Reply #48
Thank you, skamp.
Error 404; signature server not available.

Pleasurize Music Foundation

Reply #49
Wait...this is "only" the foobar2000 component to analyze a file, not the real time vst plugin (the one i posted in my image)!!
I can't find a link for the vst though, if someone want it i'll post it when i'll be at my pc.