[TheStumps]

Firstly, Let's be clear - this method won't deliver a "true" dynamic range measurement if by that we mean a measurement which takes into account negation of true silence and other related factors.
There's a few other factors involved in tools that measure dynamic range.

Also, let's clear the terms correctly.
By "Dynamic Range", I don't mean literal "Dynamic Range".
What almost all content referring to the "popular" employment in the "loudness war" zeitgeist is referring to as "Dynamic Range" is technically "Crest Factor".
That is: the measure of the highest peak to the average (body) level.

There's lots of tools out there for measuring this (Wavelab, TT Meter, Ozone 5, Waves Dorrough, Voxengo SPAN, Dynameter, etc... [some are garbage - I know this list is good meters]).

One thing that I don't really like about any of the meters, however, is that it doesn't give you a second-by-second graphical map of what's going on.
Mostly, they all do one of two things (or only one):
1) Give you live readings of the current value(s)
2) Give you an output (of) value(s) when you feed in a song file (.wav, .mp3, etc...)

Well, that's great, but I'm far more analytical and visually based than that.
An average, min, max, or even real time doesn't help me as much as being able to see the whole song splayed open and showing me the level, peaks, and dynamic range for every second of the song.

So, since I work at a telecommunications company I have this awareness of a tool we use for broadcast TV as a "Kentucky Windage" sort of gauge (it's not what's used on the EQ - it's what we use if we just want to spot check a feed from the receiving side real quick).

It's called Orban Loudness Meter.
[Only registered and activated users can see links. Click here to register]

(Now...I haven't downloaded the updated version, but I'm fairly certain the primary components that I use are still included and function the same since they are the integral components of the system)

The basic rundown
The way that I use this tool is that I calibrate my system to make sure that a test file ([Only registered and activated users can see links. Click here to register]
) plays right about at -3 LKFS.
(If it's off by a smidge...meh, we're dealing with a multitude of factors so don't be shocked if you get -2.8 to -3.4 ish results...just get it as close to -3 as possible. If your system doesn't budge when you move your volume around, that means you don't need to calibrate anything and it's going to capture the sound raw - hurray for you!...it also means you should inherently see this file park at around -3 LKFS.)

Once I have that dialed in, I adjust the settings to capture data every second and save it to a file, turn on the write to log file option, turn on the Momentary, Short Term & Integrated, VU, and LRA meters, and finally hit the "power" button to start the capturing.
Then I click play on the song and let it play through.

Once it's done, I rename the log file (it's just a text file) to whatever I want, remove the header data so it's just the tracking data (see below), import it into Excel (Data tab, From Text option - comma delimited), and remove the rows of data which are obviously when the song wasn't playing (rather easy to spot as they usually have an LKFS value down in the - hundreds or very nearly in that range...you don't want to include these in the calculations because they aren't actually part of your song).

Once I have it in Excel I add two calculations on two additional rows to the right of the captured data:
1) The average of L and R reconstructed Peak (Orban captures peaks by L and R separately - and the new one probably gives an option for even more since it supports surround sound now).
This step is simply =average(L:R) where L and R are the cell value on the first row in the columns respectively for the L and R values for the reconstructed peak.
Now I have a single Peak value.
I drag that the rest of the way down the song's data rows.

2) Once I have the Peak, I subtract the Peak from the value on the same row for the LKFS (which will be printed as whichever BS.1770 you used...I use BS.1770-2...these are differences of standards so it doesn't matter a HUGE amount to your uses. There may be more options in the newer version...I'll probably just stick to 2 if it's still there, or figure out which is the one that fits best to the test sound).

This is my calculated CREST FACTOR (often called "dynamic range" in the internet's lexicon of fast-n-loose language).

I then compile the LKFS, PEAK, and CREST FACTOR/DYNAMIC RANGE columns right next to each other and create a simple LINE graph chart from the data.

BAM!
I have my entire song opened up in front of me per second in regards to "loudness" analysis.

OK, some quick instructions:
Cleaning the data file

Content visible to [Only registered and activated users can see links. Click here to register]
.

The line that starts with "Date,Time,etc...."
Delete everything above that line and save the file.


Adjusting the Settings:

Content visible to [Only registered and activated users can see links. Click here to register]
.

The stuff that's turned on (is orange):

Content visible to [Only registered and activated users can see links. Click here to register]
.

Alright.
Now, Here's what it looks like on the output.
Here's Nirvana's Nevermind:

Content visible to [Only registered and activated users can see links. Click here to register]
.

This song has a pretty decent "loudness" for broadcast and a decent "DNR" (Crest Factor).
Seeing things above that +10 line is *thumbs up*.
Trending below that line gets less and less *thumbs up*.

Now...fun part of using this way of looking at things is that DNR (hold on here, I'm going to start a fight...) IS NOT THE KING OF GOOD OR BAD.
You can't tell if a song is set well JUST on the DNR alone.
You need the rest of the information (LKFS and PEAKS), and here's Trent Reznor to passively explain just why that is:

Content visible to [Only registered and activated users can see links. Click here to register]
.

You see, in Less Than, Trent did something artistically interesting with the production.
He PURPOSEFULLY reduced the distance between the body and the peaks as the song progressed.
He did this to increase the tension and force of the song.
Notice, he didn't violate sound levels in doing this. He maintained relatively the SAME peak values, but his average body level ("floor" - not actually the floor technically, but in this topic it's the lowest item so it's our floor) up and up slowly over the course of the song.

This takes considerable talent to keep the PEAKS about the same while raising the BODY up to purposefully LOWER the DNR/Crest Factor without employing massive compression or limiters (and we can tell that there's not a huge amount of limiting or compressing to accomplish this because the peaks aren't flat-lined).

So how do I use it?
Well, It tells me when I've made a mistake and didn't realize it.
Like in a recent title I uploaded here, Fugue.

Content visible to [Only registered and activated users can see links. Click here to register]
.

See, my "DNR" Reading for this comes out "good" (about on par with Nirvana, which is decent).
HOWEVER, if I went by that I would be flat WRONG!

I royally screwed this song's settings up by smashing too much too high and consequently the PEAKS are slamming up against the limiter protection and I'm getting this ugly flat line for the PEAKS.

However, I can tell that the general BODY is good, and the relationship between the BODY and PEAKS is generally correct.
Everything is just too loud.
So all I have to do is turn the track, as a whole, (not on the master, but a pre-bus grouping) down about 4 to 5 dbs, reprocess it, and I should be good (as well as probably get a bit of a boost in the "DNR" value since I won't be squashing up against 0 for the PEAKS).

So...There you have it.
A really granular way to look at the dynamics of your song in regards to "loudness".

Cheers,
Jayson

[Daggit]

wow, I don't totally comprehend all this. but i will certainly think about it. i do understand and remember the trent reznor stuff though. what i can say is that with enough practice of looking at a typical audio waveform at the same zooms, a person can start to get an idea of if the levels are too much or too little. after years of looking at the waveforms, it starts to make sense. i like SmexoScope / FreakoScope type of plugins for this (and for the pitch info too).

[TheStumps]

This whole Dynamic Range bit is something I encourage everyone to get familiar with to a basic level because all of the streaming services have moved to using it as a gauge to determine if they will further compress your upload (if your streaming service isn't already doing it...it will soon).
Soundcloud, iTunes, Spotify, etc...

Personally I think this is a terrible idea because of things like NIN, where if you use JUST the DNR, you're assessment that the song is "too loud" will be horribly WRONG....but...well...I don't have any control and this is the silly world we live in now for distribution.

So it's really valuable to know how to reach that DNR value that will reduce the amount of compression the streaming service will apply to your track.

I've seen the topic raised here a few times; folks complaining about given services' quality "sucking" because the compression ruins the song by comparison to the mp3 they uploaded.

Well...it's not just a matter of compression - it's a matter of their system reading your DNR and (often) you not knowing that, or (also often) you not even knowing what your DNR average even is, so you don't know until after you upload that you are going to be hit with a reduction in quality. BOOOOO!!

If you do something like the above OP, then you can avoid the issue more easily because....and here's the really silly thing...it's not about HEAD ROOM with streaming services.
To my knowledge, they don't even look at head room (distance from peak to 0); they only look at Peak to Body ("DNR").

I get it...the logic goes something a bit like this:
"Gee, these amateur EDM artists these days are cranking everything up really loud and slapping a compressor on it to squash it into the allotted bandwidth range and that's making it very loud because of how tight it's packed, and that they just slapped a limiter at 0 and mash everything up to the limit almost."

"Hey! Why don't we check their DNR and if it's really flat, then we compress it to tone it down?"

"That's Genius Steve!"

Well...OK...to an extent, they have a point...but only in context to that specific type of upload.
So because of a few (thousand/million?) bad apples, EVERYONE else is getting screwed from employing purposefully flatter DNR values in otherwise NOT LOUD songs (e.g. NIN style).

It would be better if they measured all three components, or at least PEAK and DNR and said, "If Peak = threshold of frequency of hitting nearly 0 AND DNR is < n, then compress, else don't compress".

But...whelp...they didn't so learn DNR to a basic level and save yourself some frustration.

And honestly...all you need to know about DNR is:
1) Aim for at least 10, higher if you can pull it off, but don't sacrifice your music to hit something like 14 or 15 DNR...if you get 10 and there's nothing to improve it...leave it

2) DNR (as used by these services involved in the loudness ware) is simply a measure of the PEAK to the AVERAGE level of your track (in the Orban, LKFS is that "BODY" measure).

3) To get an average out of a logging system like the Orban, just take your DNR column and tell Excel to give you the Average of that column - tada!
You can even get the Min/Max with =min and =max and cite the column you have your DNR per second values in.

Cheers!
Jayson

[TheStumps]

UPDATE: I just downloaded and ran the new version of the Orban meter.
I was able to accomplish essentially the same with the udpate.

Also of interest...it now has an "Analysis" option, which is pretty interesting.
You just add a wav or mp3 (sound file) of your choice and it chews through Orban's opinion of its loudness and delivers a line graph back (which is similar to the one produce in the excel variation...though, no DNR or peak info line charts)...though, it's pretty cool to get back the True Peaks Over Peak Alarm because that'll tell you how many times you're hitting the ceiling (in the Fugue's case...lol...61,131 times...sheesh, I really need to fix that).

Interestingly...it looks like they've included an Average "Integrated Loudness" rating that they spit out.
This will often times loosely be the same as a DNR value (because of the way that Orban is set up for TV and Radio broadcast) if you are cranking up to 0 with a limiter, so that's also kind of spiffy if you just want a quick grab of what your song's average will likely appear as to streaming services.

[TheStumps]

Oh WOW!
I just clicked the Analysis tab's Line Graph drop-down and selected Histogram....WOOOOW.
That is going to be really fun to work with!

Content visible to [Only registered and activated users can see links. Click here to register]
.

[White Noise]

[Only registered and activated users can see links. Click here to register]


As far as I know, this can't record data for analysis in excel, but I think it does enough if you just want to target your songs to be not too loud for streaming. As I understand, they use LUFS, which is similar to an RMS measurement of the loudness over the whole song with some limiter and silence tricks to otherwise boost loudness accounted for. For most services, you aim for around -14 LUFS and call it a day. I'm personally OK with it if I'm +/- 2 LUFS from there. And no red squares, that's peaking.

[TheStumps]

Aye.
LUFS and LKFS are the same thing - just different names.
The measures are equal though.

[TheStumps]

Also...I didn't have time earlier, but:

LKFS/LUFS shouldn't be confused with Crest Factor (sloppily referred to as "dynamic range" currently).

LUFS is a measure of distance from 0 VU (well not by definition, but in most DAWs it's going to work out that way; unless you switch things around).

Crest Factor is the distance, in dBs (by consequence) from the Peak ("True Peak") to the Body (Integrated Loudness level average).

If your ceiling is 0, then something around -14 LUFS is going to roughly ballbark in at a "DNR" (Crest Factor) of around the same value in the positive (assuming that you don't compress everything to be, ohh, only 5 dBs away from that -14 LUFS.

However, if you pull all of your tracks down so that the largest peak that you see kisses, say, -3 LUFS, and your main body average is around that -14 LUFS, then your "DNR" won't be roughly around 14, but instead roughly around 11.

This becomes more dramatic if you set things like the early 90's did.
Let's say your top end is a peak of around -10 with an average of around -20.
Well, then roughly 10 will be your Crest Factor/"DNR".

In this case, if you elevated everything to -14 instead of the -20, then your peak would be at about -4 and your "DNR" would still be 10.

So you can't judge the Crest Factor in all cases by eyeballing the LUFS value.
You can if you're roughly calibrating against a 0 ceiling, but as soon as you depart from a 0 ceiling settings for head room and rely less on limiters, then you can't rely on the LUFS alone.

[TheStumps]

Here's some more analysis that shows how this method can be beneficial to employ.

When you're self-taught (which is likely the vast majority of folks around here), one of the ways to learn is by looking at music that already exists from various artists - comparing and contrasting ways things are accomplished.

Using a logging tool allows you to do this in pretty granular detail - you get a bit of a view into their production (and through inference some of their methods).

So let's take an example - I took Deadmau5' Ghosts-n-Stuff and logged it through Orban.

So firstly, let's look at the pretty picture because I think you can immediately grab a good amount of information from the graph alone.

Content visible to [Only registered and activated users can see links. Click here to register]
.

Now, immediately there's some interesting information here.

For one thing we can easily spot that, by comparison to the previous Nirvana or NIN samples, Deadmau5 has set this song pretty loud.

His body (integrated loudness) is where NIN and Nirvana have their peaks, and his Peak is hovering an average of around ~ -3.5 dBs.

You can also easily spot that he's got a compressor sitting there at that -2 dBs range with a slight forgiveness because his max peak is -1.57, and the Peak line is almost flat.
It's not as flat as what hitting a limiter looks like, as there's more fluctuations involved - so clearly we're looking at a compressor.

Interestingly, this looks to be true of the body as well. Meaning - he probably has a few compressors pushing things into this rather tight tunnel.

His DNR is consequently pretty low - averaging around -8. If you take off that ending after 316 seconds, then he's primarily averaging around 7 to 7.5 dBs of dynamic range.

In other words...a wall of sound with very little variation.


While we're at it, this is a GREAT example to point out one of my opinions.

DON'T USE THE SINGLE HIGHEST PEAK TO COMPARE AGAINST THE AVERAGE BODY (LUFS/LKFS) FOR DETERMINING DYNAMIC RANGE!

You'll find that instruction virtually everywhere on the internet.
I think even the badass Ian Shepherd outlines it this way (I double checked...yes, he does).

I HIGHLY respect Ian Shepherd, but on this point...I respectfully disagree.

Doing so gives you a wider DNR value than you may deserve because of ONE peak.

This Deadmau5 song is a good example.
If you calculate the DNR per the standard instructions you'll find:
(to quote Ian)
"PLR, on the other hand, means simply “peak to loudness ratio”, and is based on the peak to integrated loudness of the music."

Or some variation thereof.

Well, let's do just that for Ghosts-N-Stuff.
Max Peak: -1.57
Average Integrated Loudness: -11.835
Result: 10.265

If you remove the ending at 316 on
Max Peak: -1.57
Average Integrated Loudness: -11.0256
Result: 9.455587

But that, in my opinion, is a lie.
That max peak occurs at second 229.
The Body at that moment is -9.8, not -11.

That's a resulting 8.23 at that second.

It is really bad employment of good data sample test practices to take a single sample from a population and compare that against the average of the sample population and cite the resulting calculation derived from those two values as a representative of some property OF the entire population average.

If you, instead, do the sensible thing and take the AVERAGE peak and compare that against the average integrated loudness, THEN you get a much more honest and accurate value.
Average Peak: -3.60826
Average Integrated Loudness: -11.835
Result: 8.226722

If you remove the ending at 316 on
Average Peak: -3.1477
Average Integrated Loudness: -11.0256
Result: 7.877889


10.2 vs 8.2 is 2 dBs of a difference.
To me, that is a HELL of a difference!

According to the guidelines by the loudness war's DNR party ([Only registered and activated users can see links. Click here to register]
), 10 is fair.
It's not excellent, but it's passable and if you want to leave it there - OK.

Meanwhile, 8 is on the poor side of their "transition" category. In fact, it's just before the "BAD" line.

If you take off the ending at 316, then we actually drop down to the "BAD" category (if we round down, and I ALWAYS round down categorically for "better safe than sorry" reasons).



Now, I've said before that DNR isn't the king number, so we can't look at this track and say "ooooo, it's bad because of a bad DNR."
What makes it suspect for a poor fitting settings (NOT poor quality...these are two different things here...quality is an aesthetic judgement based on a range of factors - and I'd say Deadmau5' production is definitely not poor in quality as you can hear everything and it is very clean)...anyway...back on track...

What makes it suspect for a poor fitting of settings is that it's rammed up to ~-2 dBs while at the same time channeled to a tight 7 to 8 dB DNR.

That's not something I would suggest attempting to do.

(In Deadmau5' case...I suspect this was done on purpose because he probably wanted a VERY punchy sound that was quite like a wall of sound hitting your face without an apology, but even-so...it's not something I personally agree with.)






So...analyzing songs using a log method is really good for perspective and learning.
Try it out, you'll be amazed what you find (especially regarding some of your favorite tunes)!