how do dynamic range meters work?

Discussion in 'Sound Science' started by salm0n, Jan 16, 2012.
Page 1 of 7
2 3 4 5 6 7
Next
 
Last
  1. Salm0n
    I've been reading up on the loudness war reciently and decided to download a dynamic range meter for foobar2000. after checking some of my music i noticed that on some tracks there were huge discrepancies between what i percieved the dynamic range to be and what it was measured to be. this was surprising, so i imported two tracks into audacity to visually compare them.
    DynamicRange.png
    (wont link to the tracks as they probably arent appropriate for a family friendly site)
    the top track (DR4) is something that i had considered to be one of the more dynamic and relaxed dubstep tracks i've ever heard, while the bottom track (DR6) was described on release by the artist as "really stupid, really loud".
    looking at and listening to the tracks gives me a completely different impression to what the DR meter gave, so i was wondering how these meters measure DR, as i really can't reconcile the measurements with everything else.
    thanks in advance.
     
  2. Iniamyen
    I don't know all the details of how it's calculated, but the numbers you see are not as simple as "highest signal compared to lowest signal in the recording." Since there is some silence in almost all recordings, this would mean they all have DR approaching 90dB, which we know is not the case.
     
    I looked at the website for that plugin and found this document: http://www.dynamicrange.de/sites/default/files/tech3341.pdf
     
    It has some information about a spec for calculating the DR, I assume this is what's used in the tool. Of particular interest (to me) was the fact that different timescales are considered for the calculation ("momentary," "short-term," and "integrated.")
     
    My guess is that even though the waveforms look a certain way, the calculation is doing something you as a human wouldn't. Perhaps the first track approximates 2 sections of "approximately brickwalled" waveforms, separated by "almost silence," when considering the measurement thresholds and intervals they used. Whereas the second recording may have had a little more variance over the large timescales, even though it is a higher level overall. I just don't know. But there is something to be said here for not trusting a number blindly. Maybe someone who understands the measurements a little more can tell us why they come out this way.
     
    Peter Hyatt likes this.
  3. Head Injury
    Just guessing here, but I believe dynamic range would be calculated as quietest vs. loudest over small intervals. The first track may appear more dynamic because it has loud parts and quiet parts, but look at the difference between the dark and light blue (which basically signifies loud and quiet over very small intervals). The difference between dark and light blue at any given point is smaller than the difference in the second waveform. Notice the RMS value in DR Meter. That's closer to what you're thinking, which is average volume. The second track has less dynamic compression, but is still louder. The two are exclusive, just often found together.
     
  4. Iniamyen
    Quote:

    A "small" interval isn't well-defined, but I get your point. That was what I was trying to convey in my first post - the calculation of dynamic range depends on what intervals you look at.
     
    So it's not equivalent to "eyeing the track," which is what caused the confusion in the first place. Eyeing the track may not be the best indicator of dynamic range [​IMG]
     
     
    Peter Hyatt likes this.
  5. Salm0n
    Thanks for the replies. my initial asumption on how it would be calculated would be sampling the level across the entire track, then taking a mean and an SD/VAR, which would reported as the dynamic range, or something similar. from what i understand you're saying that it calculates based on average rate of change across small intervals (correct me if i'm misinterpreting). i can see how that would explain the differences between what was measured and what i heard.
    i'm still a little confused about how to interpret the measurements, so if anyone could point me to documents explaining how to interpret these mesurements or how the calculations are made that would be great.
    thanks again.
    EDIT:
    Had a second read through that link Iniamyen posted. the maths is a little over my head but it seams to agree with the idea of small intervals being used rather than a measure of variation across the whole song. i'm still not  entirely sure on how to interpret measurements, but i think i've got a better idea now.
     
    Peter Hyatt likes this.
  6. omasciarotte
    Hey Salm0n,

    Don’t confuse R128 measurements with “dynamic range” measurement. The phrase “dynamic range” has several meanings that I can think of (system or aggregate versus format to give you just two examples), so the concept is ill defined and can cause confusion when talking about those concepts. Same goes for other related phrases and concepts, like “peak amplitude” (is that sample peak (SPPM), True Peak (TP), or QPPM quasi–peak program? ) and “loudness,” also a very vague label.

    As I’ve said elsewhere,
    Also in that thread, I mentioned that, this month, MAAT revised the original DROffline to DROffline MkII, which adds R128 metrics and a whole bunch of additional measurements that you folks would find of interest. The original, entry level DROffline is also still available.
     
    Last edited: Jun 26, 2018
  7. omasciarotte
    BTW, if any of you have knowledge in this area and are interested in reviewing the codependent DRMeter MkII + DROffline MkII, please ping me via PM and I’ll see what can be done. Note that DRMeter MkII is a plug–in and requires a VST2, VST3, AU or AAX plug–in host, like Audirvana Plus, Foobar (with VST kludge) or any modern DAW.
     
  8. gregorio
    It's a bit of a misnomer, the DR meter and the DR database isn't actually measuring dynamic range at all. It's effectively measuring an average of the "Crest Factor" over a selected period of the song. It is designed to give an indication of how much compression has been applied, with lower numbers meaning more compression and therefore how much it's dynamic range has been reduced. Of course though, that doesn't really tell us much about how much dynamic range a song actually has and furthermore, it's only a rather vague "indication" of how much compression has been applied, as it's entirely possible to get a lower average crest factor with no compression and a higher crest factor with lots of compression.

    That's all rather an over-simplification and is caused by the fact that although it appears to be entirely obvious and takes virtually no conscious effort, "under the hood" what's actually happening (in our brain) is quite complex as far as loudness and therefore dynamic range is concerned. There is no actual single property of sound which directly equates to loudness, it is effectively just a human perception, despite the fact that it so obviously appears to be an actual property. Hence why the various different measurements, obviously including the DR meter, often do not correlate with our perception of loudness or dynamic range. An exception to this is the measurement method posed above in the EBU tech3341 document, which incidentally is effectively the same as the R128 (also mentioned above), the ATSC A/85 and the ITU BS.1770. This measurement method was achieved by playing different signals to countless test subjects, getting them to rate the relative loudness of those signals, averaging all those results and inventing a "transfer function" which when applied to sound/music most closely matches the average ideas/perception of loudness. It only works under certain conditions but within those conditions it's actually a fairly successful measurement that correlates well to our perception of loudness and dynamic range. It's very different to how the DR meter works though and your observation of the DR meter is correct.

    G
     
    Last edited: Jun 27, 2018
  9. omasciarotte
    Hey gregorio,

    DRi isn’t Crest Factor, though they are related. All the rest of that paragraph is spot on.

    It is designed to give an indication of how much compression has been applied, with lower numbers meaning more compression and therefore how much it's dynamic range has been reduced. Of course though, that doesn't really tell us much about how much dynamic range a song actually has and furthermore, it's only a rather vague "indication" of how much compression has been applied, as it's entirely possible to get a lower average crest factor with no compression and a higher crest factor with lots of compression.

    Agreed, and thanks for the clarification. All “loudness” metrics fail with certain genres, tracks and skillfully crafted content designed to exploit the measurement method. That said, the internationally mandated Loudness metrics (1770 et al) do a very good job at what they were designed for: broadcast loudness normalization with an emphasis on advertisements. Likewise, DRi does a very good job at what it was designed for: an easy to understand (hence the single integer representation) measurement of the dynamic density of popular music. Two different metrics for two different jobs; The R128 family for broadcast/SVoD/OTT/streaming, and DRi for pop music production/special venue/gaming. [Please excuse the promo, but we feel strongly about this…] DRO2 and DRM2 also include metrics not found in other products, like Minimum PSR and a range of channel-specific measurements, that not only help to ID problems during production and post, but also make you all better informed listeners.

    The upshot of all this awareness of Loudness and the Loudness Wars is that artists, engineers and producers are starting to realize that, if you make content artificially loud, it will be gain reduced on playback through any popular realtime distribution service. So, by trying to make it “stand out,” you’re actually making it quieter, aurally receding into the background, relative to the tracks before and after it. The one current exception is radio broadcasting in many parts of the world, where loudness normalization is still being discussed but hasn’t been mandated or implemented.
     
  10. gregorio
    1. The ITU 1770 was NOT specifically aimed at broadcast with an emphasis on advertisements, just at recorded audio loudness in general! It agree that the ATSC largely used it for the purpose of controlling advertisement levels, due to the introduction of the CALM Act in the USA, in addition to controlling levels between different channels (and so did the EBU). However a year or two ago, the AES published a recommended set of loudness specifications (based on the same ITU 1770 measurement) specifically for music production/distribution! Furthermore, there's some discussion of developing specs based on ITU 1770 for theatrical film production and distribution. Of all the loudness measurement paradigms, the one's based on ITU 1770 are by far the most accurate but even these only address what the perceived relative loudness (and/or dynamic range) actually is, NOT what it should be. For example, a Hayden string quartet should not be the same loudness as say a song by Metallica or have the same dynamic range as say a Mahler symphony. There's currently no solution for this aspect of loudness or dynamic range and none even on the horizon, as far as I'm aware such a solution is well beyond even the most sophisticated AI. Furthermore, this is NOT purely a music issue, a romantic comedy or other inter-personal drama should obviously not be as loud or dynamic as say a big action blockbuster.

    2. For me, that is it's problem or rather, problems!:
    A. I don't exactly know what you mean by the term "dynamic density" but more importantly; if it was designed as an easy to understand measurement of "dynamic density", why is it labelled and marketed as an easy to understand measurement of "dynamic range"? I assume it's a far easier "sell" to use the term "dynamic range", which many are familiar with, than to have to explain the term "dynamic density" and then "sell" an idea/concept which no one is familiar with and which most wouldn't bother to familiarise themselves with? In other words, it's simplicity and ease of use broadens it appeal but the downside is that most would naturally assume that the DR database is actually a database of dynamic range measurements, which it isn't, and so it's effectively misleading to most and confusing to others, like the OP, who observe the discrepancy.
    B. If it's designed as a measurement of the "dynamic density of popular music" why does the DR database contain measurements of other genres, such as classical music for example? Even if it were restricted to only popular music, it would still have essentially the same problem as the "Loudness Normalisation" (ITU 1770) measurement: Namely, what is the right amount? A DR measurement of say "6" might be entirely appropriate for one piece (or even a whole popular music sub-genre) but entirely inappropriate for another, IE. Some sub-genres require more compression than others. Furthermore, there's the natural dynamic range of popular music genres to consider, which the DR database doesn't. For example, it would be entirely feasible to create a piece of ambient music with a DR measurement of say "7", using little/no compression, while another genre might need a quite severe amount of compression to achieve a measurement of "8". In other words, one piece with little/no compression, which is totally fine might measure into the Red, "Bad" region, while another piece with unwanted, quite heavy compression artefacts might measure in the Orange, "Not too bad" region!
    C. There seems to be a problem with the DR measurement with certain media types, vinyl for example. It appears that certain distortions are introduced in the cutting and/or playback process with vinyl which causes a higher DR measurement than is actually the case, up to "4" higher I believe. So vinyl releases (or rather, rips of vinyl releases) cannot be compared with releases in a digital format.

    If it were me, at the very least I'd have an easy to access Wiki/explanation page on the DR Database site, explaining these issues and providing more detailed guidance on how to use and interpret the presented DR measurements!

    3. Yes, I can see that! IMO though, the DR database is clearly misrepresenting both dynamic range and the loudness war. btw, what is DRO2 and DRM2?

    4. Firstly, let's not forget that it was the engineers who first raised the issue of the loudness war, more than 25 years ago. Secondly, we're not there yet, there's still the issues of: A. How loud a piece/song should be, relative to another type of piece/song. B. No single standard. Youtube is still probably the world's largest music distribution platform and with an integrated loudness specification equivalent to around -12 to -13LUFS, that's not low enough to really punish excessive compression.

    G
     
  11. omasciarotte
    Agreed and agreed. Once again, thanks for the clarification! That last bit is not quite true. A few EU countries have implemented the “Low Loudness Flag,” metadata that tells playout systems to “…leave this clip alone, don’t mess with it.”

    Yup. Dynamic Density is the lack of dynamics caused by overly aggressive dynamic compression and limiting. An example of low density being a purist acoustic recording like Hoff Ensemble’s HRA recording of Quiet Winter Night (DRi of 13) whereas high density might be The Stooges’ 1997 remastered Search and Destroy, which has a DRi of 1! See http://dr.loudness-war.info for measurements of your faves.

    Cuz folks love to measure and compare everything! :wink: As I mentioned earlier, no single measure provides a complete picture. As to your point about what is “right,” that’s up to the artist and producer. In the case of DR and Loudness, we provide a buttload of information (disguised as user manuals in our Support section) so our customers can make informed decisions.

    Yup, vinyl can only be compared to vinyl since that distribution medium usually entails aurally subtle but significant frequency–dependent dynamic range reduction along with L+R summing at low frequencies, and sometimes HF de–emphasis to overcome problem sections. And that’s just in mastering! Playback and subsequent re–digitization, through an unknown analog chain and ADC, introduce another very complex convolution. Long story short: a digital capture of a vinyl release has significant differences compared to the original master file (or tape/live feed; most all vinyl is cut from a digital master these days), at least to an algorithm.

    Good idea! I was talking to those folks recently about other matters, and I will bring that suggestion up w/them. Please remember we, MAAT, are a young company. We only opened for business a year ago so, still lots to do.

    Thanks for putting up w/my zeal. I will bring your above view up w/the DR database folks as well. DRM2 = DRMeter MkII while DRO2 = DROffline MkII

    Agreed, and nope, no single standard (Ain’t that beauty of standards?). But, as an industry, we are getting better! Widespread adoption of loudness normalization certainly helps. We at MAAT may be a young company but, as individuals, this ain’t our first rodeo and co–founder Tischmeyer also founded the PMF, developer of DRi and the first organization to formally say, “Stop the (Loudness War) madness!”. He’s also a member of the EBU’s ploud group, creator of the R128 guidelines, which were codified long after DRi was introduced. We are excited about our friend and colleague Eelco Grimm’s work to define and promulgate Minimum PSR, a newly defined metric to add to our production toolkits. BTW, thanks for a rational, informed and insightful discussion!
     
    TheSonicTruth likes this.
  12. old tech
     
  13. gregorio
    That last bit was true! A few EU countries have NOT implemented the "Low Loudness Flag", some TV broadcasters in a few EU countries might have but it's not been implemented by music distribution platforms as far as I'm aware. Even if it were, how would that solve the problem? Should our Mahler Symphony and Metallica song be the same loudness, who/what decides and how? And, if our Mahler Symphony does have it's "Low Loudness Flag" set ("loudness normalisation" is bypassed) to avoid the Metallica issue, then presumably so would our Hayden string quartet and then we're right back where we started, with the problem of if our Hayden string quartet should be as loud or dynamic as a Mahler symphony. Of course, I could use all popular music examples, I'm just using classical genres and heavy metal to more clearly illustrate the issue.

    1. To an extent but not everything, how do you measure and compare a fighter jet with an ice-cream or a Mahler symphony with Gangsta-rap? Let take a more specific measurement and comparison example, say the size of an engine and vehicle performance. A larger engine size is often an indicator of greater performance but there's no direct correlation, we can for example have a motorbike with a 1.0l engine that has several times more performance than a truck with a 6.0l engine. In practise, this measurement and comparison is not used for vehicles in general because too many members of the public are aware there's no correlation. If it is used, it's only used within very specific sub-categories of vehicles and even then only under very constrained conditions (excluding turbo charging for example). Again, enough of the public are aware of this for it not to be an issue but that's the root of our problem here, dynamic range appears to be a simple/obvious thing (like engine size) but unlike with engine size, the public are not aware of the complexities or even the basic fact that the DR measurement does not correlate with dynamic range. To continue the analogy, it would be like having a "Vehicle Performance Database", where all vehicles are listed with a single measurement called "Performance" which was essentially just the engine size. As far as actual vehicle performance is concerned, such a database would be largely wrong/useless!

    2. Agreed. Although I'd add, "or often even a useable picture". Which begs the question then, why does the DR Database present all it's data on the basis that one single measurement does provide a complete picture (and not even a particularly applicable measurement at that)?!

    3. I'm not sure I understand that answer. It's leaving it up to the artists and producers to decide what's "right" which is mostly responsible for the loudness war in the first place and why some sort of enforced loudness control is required to end it.

    4. Going back to point #1, folks generally prefer to only compare things which make sense to compare. You and I know that it only "makes sense" to compare vinyl to other vinyl (and not to CD or other digital distribution formats) but how many other users of the DR Database do? By not making it clear, in fact not even mentioning it at all, the DR Database is in effect stating that it does make sense to compare the DR measurements of vinyl and digital releases, which not only completely contradicts your statement but causes many/most users to blindly make this comparison and thereby be significantly misled!

    1. Thanks for the explanation, but that just leads back to a previous point, how do you measure "lack of dynamics caused by overly aggressive compression and limiting"? First of all, how do you even measure the amount of compression? Using the metric of crest factor for example, a low crest factor can be an indicator of heavy compression but you can have a low crest factor with no compression and a higher crest factor with a lot of compression. It's the same as our engine size and vehicle performance analogy, you can have a smaller engine but superior performance. Secondly, even if you could measure the amount of compression, how do you determine an "overly aggressive" amount? A severely "over aggressive" amount on one piece/genre could be way too little on another. Using our examples again, with our Hayden string quartet we'd reach an "overly aggressive" amount of compression well before any audible non-linear distortion but with Metallica, audible non-linear compression distortion is the minimum starting point! The compressor/s were almost certain to be slammed/over-driven well beyond that starting point and that was back in the 1980's, even before the jump in the loudness war during the mid/late '90's and early '00's!
    1a. Thanks, but what do they actually do? How are they measuring the amount of compression?

    2. Agreed. However, until we have a more universal loudness normalisation and at a level that clearly/obviously penalises excessive compression, then I don't believe we'll see a significant change in the situation. Ultimately, it will probably take a long time, not only on the technical side to achieve a more universal, low loudness normalisation level but also because the changes required are not just in the mastering stage but in the actual compositional/structural and mixing stages. Effectively, an evolution of some/many of the current popular music sub-genres is required.

    G
     
  14. bigshot
    I think it's a mistake to try to reduce creative choices to mechanical standards. Calibration is great and it's very useful, but once you start mixing and making creative choices, it's the skill and taste of the sound mixer that is important, not the charts and diagrams. We should be educating listeners about how mixing and mastering works, and pushing for better application of the tools. Some music is intended to be dynamic. Some music isn't. Narrow dynamics isn't necessarily a bad thing in the right context. We really don't need tools for measuring apples and oranges on the same scale. As I said before, I have harpsichord CDs that would probably register as the most compressed music ever recorded. But that is the correct sound for the instrument. And I have some prog rock albums that are mixed like a wall of mush with a million different things all going on at once with no dynamics and no air around anything. But they were designed to sound like that. You can't run stuff like this through a numerical scale and compare them to numbers generated by a Mahler symphony or Ravel's Bolero.

    Calibration and measurements are for keeping things in spec. Ears are for judging creative issues. Application of dynamics is creative.
     
    Last edited: Jun 29, 2018
  15. TheSonicTruth

    Attn. Gregorio:

    " And I have some prog rock albums that
    are mixed like a wall of mush with a million
    different things all going on at once with no
    dynamics and no air around anything.
    "

    ^This^ is what I, and others here, mean by 'density'. And yes, you can have density without loudness. I have CDs of such audio that was not peak normalized in the least, but still, played at past a certain volume level, sounded like the tarmac at LaGuardia the day before Thanksgiving.
     
    Last edited: Jul 3, 2018
    omasciarotte likes this.
Page 1 of 7
2 3 4 5 6 7
Next
 
Last

Share This Page