Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[MusicBringer]

Hello Folks, I have run the Analyse Audio function for the whole of my library. It took a few days but its done now.
I have read and re-read many existing posts on the subject of Analyse Audio.
I am still unable to understand or interpret the readings.
Whereas the old DR gave me a number such as DR6 which was poor or DR16 which was high quality.
I knew where I stood with that.

Now however we have both DR and DR (R128) plus two sorts of Peak levels and two sorts of Volume Levels.
I understand that the old DR and DR (R128) are looking at/measuring different things but the fact they are both called Dynamic Range confuses me.
Plus the fact that the highs and lows of DR in no way correlate with the Analyse Audio readings of DR (R128).
For example a low DR6 does not necessarily give me a low DR (R128) LU.
Nor does a high DR16 give me a high DR (R128) LU - in fact often the DR (R128) LU is lower.
This is so confusing...

[InflatableMouse]

Let's see if I get this right .

DR (128) measures the difference between the quietest and the loudest part, while DR measures the the difference between average and peak loudness.

[mwillems]

Let's see if I get this right .

DR (128) measures the difference between the quietest and the loudest part, while DR measures the the difference between average and peak loudness.

You're right Mouse, but there's a caveat which is what allows for some of the seemingly anomalous results that OP is seeing. 

DR compares the average and peak loudness, and R128 compares the loudest and quietest parts of the track, but with specific parameters.  R128 Dynamic range is the distance between the 95th percentile for loudness and the 10th percentile for loudness.  This is the case so that track lead ins, momentary silences, or fade outs don't set the bottom of the "dynamic range" and so that one or two very loud noises don't set the top of the range.  That's why R128 DR can actually be lower than the crest-factor DR: if R128 really measured the difference between the absolute quietest part and the absolute loudest part of the track, it would have to (necessarily) be a larger number than the crest factor DR, every time. 

Check out the R128 spec for more http://tech.ebu.ch/docs/tech/tech3342.pdf

[MusicBringer]

Thank you.

[Jimmy McNulty]

I read some of the PDF and I read the sticky on the subject here, but at this point this subject is over my head.  I would like to learn about it and start using it.

Would someone mind pointing me in the right direction so I can learn not only what DR / DR128 do, but the purpose for them... A beginners guide, if you will?

[JimH]

Why not just use it a little?  The purpose is to prevent large volume fluctuations due to how the tracks were produced.

[Jimmy McNulty]

Sure, I can just use it, but wouldnt it be nice to understand what I am doing?  I dont see the harm in that.


Thanks

[mwillems]

Sure, I can just use it, but wouldnt it be nice to understand what I am doing?  I dont see the harm in that.


Thanks

Here's a quick thumbnail sketch of what it does (hopefully I'm not just telling you what you already know).  Have you ever put on an album and noticed that it was drastically louder than the last album you were listening to?  I have a few albums that every time I put them on I race for the volume control to turn them down.  Similarly, I'm sure you've heard albums that were "too quiet" and you had to turn them up quite a bit to hear them at all (only to go racing for the volume control again when you followed them up with a loud one).  This is the result of the fact that (unlike film) there's no loudness standard in audio engineering.  Because louder audio is often perceived as sounding better, there has been a "loudness war" in mastering music to try and make the loudest sounding music possible, often through compression (reducing the dynamic range to make more of the song "loud").  The two DR measurements provide different measurements of how compressed a given piece of music is (expressed in terms of a decibel range as described above).  

There are two big unfortunate side effects of the loudness wars: some modern music is compressed to within an inch of its life, and albums have drastically varying average volume levels.

There's not much to be done about the first problem (unless you're a recording engineer), but volume leveling mostly solves the second problem.  It does it by analyzing the "loudness" and dynamic range of the music (as described in the R128 paper above), and then adjusting the volume level of the music to a reference level (-23LUFS).  The result is that most music will sound close to the same average volume with volume leveling on.  Volume leveling does not compress the music or affect dynamic range; it just changes playback volume.  By default, when you play an album, the whole album gets adjusted by a fixed amount so that intertrack dynamics aren't affected (i.e. a "quiet" song on an album will still sound quiet compared to the rest of the album).  When you play a mixed playlist, each track is adjusted individually so each track sounds about the same volume (like on the radio).  

Because most music is mastered so that its peak level is close to (or sometimes even above) digital full scale, in order for volume leveling to work correctly (without clipping) most music must receive some amount of attenuation (rather than boost).  This means that when you first turn it on you'll need to turn up the volume a little; but after that you'll be able to leave the volume control more or less alone  

Here are some wiki articles that might help:
http://en.wikipedia.org/wiki/Loudness_war
http://en.wikipedia.org/wiki/Alignment_level

[Jimmy McNulty]

Thank you very much for the response.  I have been reading about the 'Loudness Wars' recently, and I was hoping that DR128 was doing something to deal with the dynamic range compression.  I couldnt really tell from the sticky, the jargon was over my head. 

[MusicBringer]

Thank you, again.

Like Mr McNulty "this subject is over my head".
Sure I can just use it too.
In fact I am using it and yes, it does what it says in the tin in that: "most music will sound close to the same average volume with volume levelling on". That's good. It works. And it works well. I like it. I like it a lot.

I wish to use the information the Analyse Audio has gleaned.
For example, where for any given Album, I have a number of old mp3s of original recordings, or perhaps a flac of a modern CD remix that has been subjected to the loudness war compression and has a low crest factor DR, or even a recent HiRes 24/96 flac with nice high crest factor DR.

Now, please tell me how I can use it.
How can I use the non-crest factor information to help me discern which recordings are worthy of keeping and which duplicates are a waste of space and should be deleted. How can I interpret the Analyse Audio results, esp the LU readings.

Thanks,

[mwillems]

Thank you very much for the response.  I have been reading about the 'Loudness Wars' recently, and I was hoping that DR128 was doing something to deal with the dynamic range compression.  I couldnt really tell from the sticky, the jargon was over my head.  

Once music has been mastered in a compressed way, there's really nothing to be done about the compression on the back end (R128 volume leveling can't help with that).  However, because R128 is a broadcast standard, it may help with the loudness wars eventually.  If broadcasters use R128 volume leveling faithfully, compression will no longer provide a "loudness advantage": compressed music will just sound compressed, not loud.  So in time, mastering engineers may "get the message" and ease off compression.  But meanwhile, we're stuck with it.  

I wish to use the information the Analyse Audio has gleaned.
For example, where for any given Album, I have a number of old mp3s of original recordings, or perhaps a flac of a modern CD remix that has been subjected to the loudness war compression and has a low crest factor DR, or even a recent HiRes 24/96 flac with nice high crest factor DR.

Now, please tell me how I can use it.
How can I use the non-crest factor information to help me discern which recordings are worthy of keeping and which duplicates are a waste of space and should be deleted. How can I interpret the Analyse Audio results, esp the LU readings.

If you're comparing different versions of the same recording, you would use the R128 readings the same way you use the crest factor DR readings. Look at each version's R128 DR and compare them across (apples to apples).  A recording with a higher R128 DR is less compressed than a recording with a lower R128 DR.  You can't necessarily compare a crest factor DR to an R128 DR straight across because they are measuring different things (as described above).  For example, I have three CD versions of Yes's Close to the Edge (an album with three songs):  

Code: [Select]

Album Version Song#1 Crest R128 Song#2 Crest R128 Song#3 Crest R128
---------------------------------------------------------------------------------------
Version 1 11 13 11 12.3 11 5.3

Version 2 11 12.7 11 13.5 9 5.5

Version 3 11 12.9 11 12.4 10 4.6
As you can see Crest Factor DR and R128 DR do not agree, especially on the third track, and that's to be expected because they're measuring different things.  Crest factor compares average volume level to peak volume level, R128 compares the 95th percentile for loudness to the 10th percentile for loudness.

In the second version of the third track, the R128 peak level for that track is +0.6 (which means clipping), and looking at the waveform in Audacity, I can see a handful of narrow spots where the track clips.  But the spots where the volume is at or near peak make up a fairly small portion of the track's running time (they are a handful of fairly narrow dynamic peaks). For crest factor DR, all that matters is the actual peak, so it compares the actual peak to the average volume level of the music and gets a "9" db result.  R128 dynamic range ignores the peak and looks at the 95th percentile for volume (the volume level that 95% of the song is less loud than) on the theory that a handful of very small peaks should not set the dynamic range, and so it discards that information and arrives at a smaller number "5.5" dB.  The 10th percentile for loudness forms the bottom of the R128 DR range (rather than the average level for crest factor), and that will often lead to R128 DR ranges that are wider than crest factor ranges, but, in this case, there aren't many quiet moments in this song, so I think the difference here is mostly at the top.

Which one is more accurate?  It depends on what your criteria are.  If you consider extremely narrow dynamic peaks to be an important part of the picture, then R128 may sometimes give you a misleadingly narrow impression.  If you instead want to get a sense of how dynamic "most of the song" is, crest factor DR may give you a misleadingly good impression (in this case because of a handful of clipped peaks). Personally I look at both and try to understand why when they disagree.

All that being said, I don't think DR (either one) is the be all end all of musical quality.  As it happens, I prefer the second version of the album shown above to the other two.  It has a better R128 DR for two tracks, but a worse one for the 18-minute centerpiece track that makes up half the album.  DR isn't a perfect way to separate the sheep from the goats, it's just more information.

[MusicBringer]

Thank you, yet again.
In fact thank you very much.
mwillems, you have gone to a lot of trouble to prepare your answer.
It is clear and concise and even I can understand it.
Good choice of album too! 

I agree with you that DR is not the be all and end all.
Your ears tell you which version you prefer - and I do the same.

I suppose I was expecting R128 to confirm that my old mp3s at 256 were of lower quality than some of my flac updates. But No; that is not always the case.

Intuition tells me that a higher LU number means a higher quality recording.
How safe am I to follow that very general supposition?

With this information I'm off to herd some sheep and goats now...

[mwillems]

Thank you, yet again.
In fact thank you very much.
mwillems, you have gone to a lot of trouble to prepare your answer.
It is clear and concise and even I can understand it.
Good choice of album too!  

Glad I was able to help  

Intuition tells me that a higher LU number means a higher quality recording.
How safe am I to follow that very general supposition?

If you're comparing two different albums, it's apples and oranges, I don't personally think DR can tell you much in that context.  I would say that if you're looking at two different versions of the same album, a higher LU indicates less compression, but that doesn't necessarily correlate to quality.  

To use a different example: I have three different CD releases of King Crimson's Lizard.  The original CD release has a slightly higher DR than the two later remasters; but it also has a lot of tape noise and strange intermod distortion artifacts from the original Atlantic studio equipment.  The later remasters gently compressed the recording (lowering the dynamic range of some of the tracks by a dB or two), but Fripp also used modern technology to "clean up" all the pops, hisses, and intermod so the recording as a whole sounds much "cleaner."  Which version is higher quality?  In my opinion, both the later remasters are "higher quality" than the original CD release.  I almost never listen to the original CD release unless I'm trying to make a point about why modern remasters are sometimes worth the money.  But you may feel differently if you listened to them; you have to take it album by album.  

I would say that, for me, if I see two versions of the same album and one has drastically lower DR than the other (four or more dB difference), then I take that as a bad sign about the lower DR version, but, really, the best way is to listen and see.  

[MusicBringer]

"a higher LU indicates less compression, but that doesn't necessarily correlate to quality"
That helps me a lot.

Coz I have been wanting LU to indicate quality I have been trying to do just that with the readings.
But rather, as you quite correctly say, a higher LU indicates less compression.
Although I have been reading the words in the descriptions I have not been hearing the words I read 

@mwillems, you have been most helpful. I am glad you are here.
Although I am not so keen on Fripp's King Crimson. Canterbury sound was too jazzy for me.
Give me Yes anytime - well, anytime it's got a high LU 

[6233638]

I suppose I was expecting R128 to confirm that my old mp3s at 256 were of lower quality than some of my flac updates. But No; that is not always the case.

MP3 or other lossy formats do not compress the dynamics/dynamic range of a track, which DR/R128 is a measure of.
They are compressing the data by selectively discarding information. (usually very high and very low frequency information) Assuming it comes from the same source, a FLAC version will always be better than a lossy version.

Intuition tells me that a higher LU number means a higher quality recording.
How safe am I to follow that very general supposition?

All else being equal, a higher Dynamic Range (DR) generally indicates that it's a better master. But not always.
 
 
Something else to keep in mind is that if you're ripping Vinyls, you cannot use the Dynamic Range (DR) values to compare track quality: http://www.youtube.com/watch?v=n-AE9dL5FG8

[mwillems]

MP3 or other lossy formats do not compress the dynamics/dynamic range of a track, which DR/R128 is a measure of.
They are compressing the data by selectively discarding information. (usually very high and very low frequency information) Assuming it comes from the same source, a FLAC version will always be better than a lossy version.

Yes, thanks for bringing that up; the word "compression" is equivocal in this context and it hadn't occurred to me that there may be some additional confusion there.  Dynamic compression and lossy file format compression are two completely different issues.
 

Something else to keep in mind is that if you're ripping Vinyls, you cannot use the Dynamic Range (DR) values to compare track quality: http://www.youtube.com/watch?v=n-AE9dL5FG8

Absolutely; that video is an exceptionally great illustration of that issue, thanks for that.

[MusicBringer]

Absolutely; that video is an exceptionally great illustration of that issue, thanks for that.

Hear! Hear!

[rjm]

I nominate this for best thread of the year.

[crisnee]

I second it. Great explanations.

[jacqlan111]

Yes, a very good explaination.

Starting from there, I've did the analyses of my favorite tracks and it came out with a pretty good correalation to the ratings I've done from 1 to 10 (some exceptions).

This is what I was looking for and I presume I can extract a text file to play in excell's graphics for the fun of it. 

In a second phase, there must be a way to have an image of the full Spectrum of one entire track, peaks and lows Vs time.

Maybe not in MC19, I understand it's not the purpose of JRiver to provide this, but can you suggest, if you feel it can't do wrong, an other measurement software that provide this kind of graph.  It may not be really necessary, but I'd love to see the picture of my favorite classic orchestral tracks which are so uncommon for their quiet and loud passages that don't belongs to the same statistic population.

I hope you understand what I mean by this statistic population concept that lie within a sampled track of that kind.

regards
jacques

[mwillems]

Yes, a very good explaination.

Starting from there, I've did the analyses of my favorite tracks and it came out with a pretty good correalation to the ratings I've done from 1 to 10 (some exceptions).

This is what I was looking for and I presume I can extract a text file to play in excell's graphics for the fun of it. 

In a second phase, there must be a way to have an image of the full Spectrum of one entire track, peaks and lows Vs time.

Maybe not in MC19, I understand it's not the purpose of JRiver to provide this, but can you suggest, if you feel it can't do wrong, an other measurement software that provide this kind of graph.  It may not be really necessary, but I'd love to see the picture of my favorite classic orchestral tracks which are so uncommon for their quiet and loud passages that don't belongs to the same statistic population.

I hope you understand what I mean by this statistic population concept that lie within a sampled track of that kind.

regards
jacques

There's a freeware program called Audacity that will mostly do what you want, albeit one file at a time.  It's primarily recording software, but if you open up an existing music file in it, it will show you the waveform.  It will also mark spots where clipping occurs if you tell it to, and can do rudimentary frequency spectrum analysis. 

[jacqlan111]

There's a freeware program called Audacity that will mostly do what you want, albeit one file at a time.  It's primarily recording software, but if you open up an existing music file in it, it will show you the waveform.  It will also mark spots where clipping occurs if you tell it to, and can do rudimentary frequency spectrum analysis. 

Thank you mwillems
I've looked at the Audacity images on Google and the graphs may do what I need.  The vertical scale seems small but it's a free software, though I'll try it.

By the way,  back in MC19, is there a way to analyse tracks after DSP processing to see the impacts of the MC19 filtering I'm doing on the dynamic range?

That would be very important to minimize DR losses while fixing different target curves when EQing HR tracks.

regards

jacques

[mwillems]

Thank you mwillems
I've looked at the Audacity images on Google and the graphs may do what I need.  The vertical scale seems small but it's a free software, though I'll try it.

By the way,  back in MC19, is there a way to analyse tracks after DSP processing to see the impacts of the MC19 filtering I'm doing on the dynamic range?

That would be very important to minimize DR losses while fixing different target curves when EQing HR tracks.

regards

jacques

The only graphical representation native to MC that I'm aware of is the Analyzer module in DSP Studio.  It's a real-time spectrum analyzer, so it will only show frequency response in real time (not total energy in the track, dynamic range, etc.). 

If you're just talking about looking at a few tracks, one option you could use is JRiver's format conversion function.  It gives you an option to "apply DSP" to the track (bake the DSP into the audio track).  You could then run the analyzer on the track after conversion to see if your proposed changes affected the DR numbers.  You'd obviously want to backup the files, etc. before hand as you probably don't want to permanently alter your audio files, but this is one way to test what your DSP processing is doing to the file stats.

You may also be able to find a VST plugin that does what you need in JRiver?  Anybody have any suggestions?