Dynamic range true value

[Compyfox]

I think you're referring to PSR (not PLR) here, aren't you? Shepherd's calculation is the same as Nugen's, who refers to it as "Short-term PLR" to connect it to the EBU standard.

I did indeed mean PSR - edited my post. Thanks for pointing out.

What I found interesting in my poking around is that the PMF DR measurement constructs a ratio based on peak to the loudest 20% of the program, where the EBU standard (if I understand it correctly) gates the top 10% and bottom 95%. This is to exclude things like "background noise" and "loud gun shots" from the measurement, but reflects a film or broadcast viewpoint.

Er... not that simple.

Realtime DR it is hard to just base the value on "20% of the program". So the realtime calculation can be off to the static offline calculation (hence the offline tool). Unless a gating mechanic was introduced - just like in the ITU-R BS.1770-x specs (or EBU R128)


For the EBU standard (or ITU-R BS.1770-x, or... you get the idea), the system is setup to just "listen in" to a certain range that the MLk algo reads out - which the "integrated calculation" takes into account. Loud bursts above a certain threshold (think positive 8LU to 12LU) are mostly "ignored" in this long term calculation, but quiet sections are taken out of the average (ILk). In this case, the "absolute silence" threshold is at -70LUFS (that material will be discarded from the analysis). But the most important gate mechanic is -10LU rel below the absolute gated loudness level (in MLk).

Example: if the signal was at -26LUFS and suddenly drops off 20dB (due to silence in the program material), the "-10LU relative" gate activates and takes the measurement out of the equation. Until the signal rises back up and exceeds the 10LU relative. The threshold is not "fixed" - it "floats" according to the average signal measurements.

Source: https://tech.ebu.ch/docs/tech/tech3343.pdf

As you said, everyone does it a little differently, and at the end of the day, it's the standards in use at the point of distribution that matter most: iTunes, Spotify, etc. Working backward from one's intended use isn't a bad guideline.

Actually... iTunes, Spotify and co these days do use the ITU-R BS.1770-x "internationally accepted standard" and not DR. They loudness analyze according to the ILk values and pretty much discard everything else.

Reading out DR/PSR/PLR/LRA values is just additional bonus in this case. Unless you're at a distribution environment where accurate logging is essential (else - penalty fees). But even then, DR is just out of the big picture.

Insisting on this type of measurement now, with the backdoor to say "it's easier to understand", doesn't help anything IMO.

You have to keep in mind that trying to meter loudness is trying to come up with one number that is an agregate of every single human's hearings - it can't be done. The intent is come up with a way to estimate what people might hear based on a very statsically shakey model. So arguing about which standard or technique is moot - as long as the approach is within the ballpark of the already stastically precarious model, it's good enough. This is perceptual metering and as such is trying to fit a square peg into a round hole: measuring subjetivity.

This is why Loudness Normalization according to ILk (average of a long term program measurement) ultimately dominated - it's neutral.

DR is just too inaccurate and really, really depends on the genre. If you can see my signature, I wrote about this issue back in the days already while the "DR Meter" came up in the first place




Don't get me wrong - I do understand where PMF (Tischmeyer)/MAAT Digital is coming from. They try to offer another "visual indication", another education tool. But to me, it's just unnecessary (just like the push of PSR) - now more so than during it's original incarnation.

IMO, Bob Katz with his K-System offered a better solution to the "Peak Limiting" problem than DR Metering did (which does need another metering spec backbone!). And then EBU R128 snuck in (sadly with not enough easy to understand/follow information back in the days for music AE's). Until people started to connect the dots with using "SLk for mastering music while staying within a certain range" (similar to what EBU R128 S1's "Loudness Parameters for short - form content" has as SLk range, which is +5LU max relative to the desired ILk target) - I was one of them.

I mean, let's be serious for a moment... how many people, that are not that interested in this field, really know/understand the various types of metering tools, and how they are used to their full efficiency? Now add in the "going their own way" meters like DR and PSR - and you'll see thread popping up over and over on audio communities that ask "what is this, why shall I use it" or "what is the best". Then those that do not(!) have an in-depth knowledge, start to say "use this and that", for mixing even.

And then we're back to square 1.

That... is my main crux with this whole topic

[plexuss]

[snip...]
I mean, let's be serious for a moment... how many people, that are not that interested in this field, really know/understand the various types of metering tools, and how they are used to their full efficiency? Now add in the "going their own way" meters like DR and PSR - and you'll see thread popping up over and over on audio communities that ask "what is this, why shall I use it" or "what is the best". Then those that do not(!) have an in-depth knowledge, start to say "use this and that", for mixing even.

And then we're back to square 1.

That... is my main crux with this whole topic

I think it's pretty easy to explain loudness and DR measurements by eliminating all the details. The details are not important in terms of using the meters. But it's important to include that any meter that is attempting to measure something which is inherently subjective is only a rough estimate. Whether someone chooses to use a meter based on an industry standard or one based on a dev's good idea, is a personal choice - there is no right and wrong in this regard. there are weaknesses of all the perception models out there. Searching for what is "right" in this regard is futile. If you preface the subject with this is can save a lot of arguements.

[Compyfox]

We both know that - and I'm sure we both are capable of getting the job done (with oldschool tools even).

But the main question is.. do others?


So a meter that says (marketing wise) to be "the end of all means" (if it really isn't, IMHO and all that) and "being a standard" (if it's really not) is not helping anything. But YMMV of course

[cturner]

This is why Loudness Normalization according to ILk (average of a long term program measurement) ultimately dominated - it's neutral.

DR is just too inaccurate and really, really depends on the genre. If you can see my signature, I wrote about this issue back in the days already while the "DR Meter" came up in the first place.

True, but ILk an integrated figure. That must be helpful for Spotify and the rest, who can use a batch process to sum a piece of music to a single number and then turn their master volume up or down accordingly.

But that does precious little for the mixing process, where you want an indication of how the current loudness of the mix relates to that integrated value IN THE MOMENT, not after you're done. Certainly the ITU/EBU recognize this in their promulgation of LRA, which relates a certain peak measurement to a measure of loudness over a specific time window.

When everyone could just meter their peaks to stay within spec, as long as every moment stayed below peak you were good. But you're not in that situation with a measurement like ILk because it's integrated over the entire time of your piece. This to me, is the reason for all the fascination with "loudness history" in all these meters: it's an attempt to show how the loudness dynamics of the moment contribute to that integrated number iTunes will derive from your finished product via batch process.

Perhaps we haven't arrived at the best methods and technology for mix metering in a world of batch loudness control, but I'm not sure the old peak metering methods are as helpful as they once were.

[Compyfox]

To make things short:

This is why you have a "histogram" feature and why you use SLk with a range +-4LU (to 6LU) of the desired target (example: -16LUFS)
LRA is "Loudness Range" from lowest to highest loudness measured. And PLR covers PLR = dBTP - ILk
And want to know the peak value (maximum signal strength), then there is the dBTP meter.

So again - why a (revived) DR Meter or a PSR meter?
You already have(!) this information with a fully spec'd (and hopefully accurate working) ITU-R BS.1770-x meter

[plexuss]

So again - why a (revived) DR Meter or a PSR meter?
You already have(!) this information with a fully spec'd (and hopefully accurate working) ITU-R BS.1770-x meter

To sell products.

[cturner]

So again - why a (revived) DR Meter or a PSR meter? You already have(!) this information with a fully spec'd (and hopefully accurate working) ITU-R BS.1770-x meter

I get the impression you think I'm some defender of DR and PMF, and that PSR is bunk. As you pointed out:

EDIT: Nugen Audio MasterCheck Pro shows BOTH PSR (top rectangle window) and PLR (meter bargraph)

Mastercheck also has a PSR "blob" wiggling in its PLR meter, and Shepherd's Dynameter shows PLR, with PSR defined the same way as the Nugen offering, ie. through EBU standards.

After Klanghelm's VUMT, Mastercheck and Dynameter get the most use here, depending on whether I'm mixing with a cocktail in my hand, or not.

[Doctor Doubledrop]

Hello, thank u all for the cool school.. I'll keep this post like a bible

[cturner]

Hello, thank u all for the cool school.

Yes, I learned a lot as well. Later, I ran across this explanation by Dave Gamble, which came out of his TrackMeter manual, that seemed worth sharing:

7.9.2 On Loudness

In case these specifications don’t mean anything to you yet, we provide here a very quick introduction to these terms. Traditionally, audio metering has been based on PPM or RMS meters (also provided here), which respond to voltage, and do not provide a good perceptual match to human experience of loudness. The ITU set about working out a remedy to this. The first stage towards this was to define more precisely how to take an RMS measurement, and to compensate for the fact that different frequencies are experienced as having different loudness.

The latter part was resolved with what’s known as the K-weighting filter, which includes a high-pass filter at roughly 60Hz, and a high shelf at about 1.5kHz with about 4dB of boost. This simple filter has proven remarkably effective in listening tests at compensating for the frequency-nonlinearity of the ear.

The RMS measurement was now to be performed on K-weighted audio. Signals were to be processed per-channel, and then squared and summed. The RMS mean is computed over some number of samples. When that number of samples equates to 0.4 seconds, we call that the Momentary (M) statistic. It has the feel of a traditional PPM meter, but maps better to perceived loudness. When that number of samples corresponds to 3 seconds, we call that the Short-Term (S) statistic. It has the feel of a traditional RMS meter. The benefits are increased correlation with human experience of loudness, and unambiguous, clear definition as to how they are computed.

The ITU (and the EBU and the ATSC and many others!) also had a requirement to be able to measure the loudness of an entire program with a single number. Historically this would literally have been a TV program or radio program. It has since expanded to film, and it is expected to transition to music very soon.

Obviously a peak or an RMS meter respond in real-time to audio information, so some further processing was required. The ITU came up with two solutions. One was to compute the RMS of the entire K-weighted (filtered) signal. This is the Ungated (U) signal, as used by ATSC A/85, and is written down as a figure in LKFS (with the understanding that LKFS means dBFS for a signal that has been measured as ITU1770 ungated).

The second solution the ITU proposed was to record all Momentary (M) readings for the duration of the program, at a minimum of ten per second. Any readings below -70dB are discarded, and the average of what remains is computed. Then subtract 10 from that figure, and discard any readings below that number, and recompute the average. This is known as the Integrated (I) (gated) loudness, as used by EBU r128. It is written down as a figure in LUFS (with the understanding that LUFS means dBFS for a signal that has been measured as ITU1770 integrated).

Whilst [Integrated and ungated loudness is often] displayed in realtime, they are only ever meaningful once an entire program has been played through them.

The EBU wanted a further statistical measurement for a complete program, to measure the perceived dynamic range. This is the LRA statistic. It is measured by recording all Short-Term (S) readings for the duration of a program and discarding any readings below -70dB. They are then compiled into a histogram, where the top 5% and bottom 10% are discarded. The difference between the highest and lowest values remaining is the LRA figure, and is a figure in dB (no units, since it’s a ratio).

Finally, True Peak is an estimation of the maximum instantaneous level of a digital signal after it has been played through a DAC. It is possible for a digital audio signal which does not clip to represent a signal that does clip in the analogue domain. This is measured by simply upsampling the audio and measuring peak levels.

[plexuss]

I noticed kilohearts Disperser is going to go on sale soon. I remember being intrigued by this when I tried the demo but passed on it at regular price. I am taking a look at it again and this time trying to understand what it does, so I RTFM. Here is part of what is in the manual. This relates to DR in a way that I hadn't thought about before. This also partly explains why working with audio and specifically digital audio is acutally quite complex. I don't know about you, but I find mixing digital audio very different than mixing analogue audio - with digital, there are so many "traps for young players" and sometimes very complex interactions that go on that are often difficult to mitigtate - things that audibly compromise audio quality. Where with analogue I think, I am guessing, that a lot of interactions that occur with digital are more "smeared" and so less obvious.

Here's another aspect of dynamic range to consider:

The Disperser is an all-pass filter. This means it has a completely flat frequency response, it does not add any gain whatsoever to the sound you put through it. It does however effect the phase of the frequencies in the signal.

If you leave the amount at a low setting a tiny bit of phase offset will be added to your signal, usually not enough to be audible. It is still a usable tool with a low amount setting however, since it can reduce the "crest factor" of a signal. The crest factor is the ratio between the peaks of a signal, and the overall power of the signal. As an example, a pulse wave with a very narrow pulse width has a high crest factor (it has high peaks, but it still does not sound very loud). Adding a little bit of phase offset to such a signal will drastically lower the peaks, without lowering the volume or changing the sound very much (yay, free headroom!). Human voice can have a surprisingly high crest factor, and many radio stations add a little bit of all-pass filtering on the mics to get more headroom. If you have vocals that sound thin even though you compress them heavily then you may have a problem with the crest factor.

[plexuss]

Sometime this server is very irritating.

[Doctor Doubledrop]

I ran across this explanation by Dave Gamble, which came out of his TrackMeter manual, that seemed worth sharing:

Really helpful to understand the argument, thanks!

I noticed kilohearts Disperser is going to go on sale soon

We will see, I didn't know it.. I'll try for sure the demo in my psy bassline. I'm not sure that it could be a nice idea to insert it in the mix channel but it may could be useful in few sounds

[teacue]

I noticed kilohearts Disperser is going to go on sale soon.

Would you mind to say when and where?

[plexuss]

I noticed kilohearts Disperser is going to go on sale soon.

Would you mind to say when and where?

JRR shop sent an email with it listed as part of their $20 for 20 years sales. it says starting weds but then it says deals will occur through throughout the month. I dont see it on sale today so I guess we have to wait. ?? Not sure.

[teacue]

I noticed kilohearts Disperser is going to go on sale soon.

Would you mind to say when and where?

JRR shop sent an email with it listed as part of their $20 for 20 years sales. it says starting weds but then it says deals will occur through throughout the month. I dont see it on sale today so I guess we have to wait. ?? Not sure.

Ah OK.
Thanks for letting me know