Loudness (ITU BS 1770 / EBU R128) Normalization

[chrisjj]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

I'm sorry, it is called "window length" and it is only available when choosing volume normalization (not replay-gain) with the mode "Adaptive (EBU R128)"

Got it - https://i.imgur.com/zsfrrRX.png - thanks.

Thanks for your report. I'll avoid this windowed version!

[gandlz]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

I'm sorry, it is called "window length" and it is only available when choosing volume normalization (not replay-gain) with the mode "Adaptive (EBU R128)"

As far as I see, the replay-gain (track-gain and EBU R128) in combination with replay-gain apply does the same thing as volume normalization (EBU R128 track-gain). This is measuring the average Loudness of the whole track and adjusting the gain so that this average measurement comes to -18LuFS (or whatever you choose). The tracks dynamic is preserved.

According to my former post I have to say that a window length of 16s is almost the same as using the EBU R128 track-gain feature (non-adaptive), what in fact is the way I go for now. 16s to react on a loudness change is way too long. If I compare it with sitting at the mixing desk, hands on fader, I usually react to a loudness impact within 2-4s. This preserves the dynamic impact but also the overall acceptable playback volume. However, my tests with using window length between 2-4s result in heavy volume changes during dynamic parts and the measuring of the processed file approves this as it often plays for a time much longer than the window length outside the target of -18LuFS. That's why I would say that the adaptive feature is not usable.

The Highpass was just a feature request and has nothing to do with this. I should have mentioned this somewhere else maybe.

[chrisjj]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

Where is the window size setting? I don't see it:

Highpass could also be used to remove DC-Offset.

Are you finding DC offset is affecting the loudness calculation?

[gandlz]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

Well, I know this is an old thread and some things are already improved but I want to add a few more:

I tried a lot with adaptive EBU R128 window size and found out that I definitely need more than the default 6s, otherwise for example a slow solo piano intro with a few breaks will become dramatically destroyed with volume going up and down. So I got good results using 16s.
However I noticed some strange behavior, and would also like to provide a song, where nothing seems to work if someone is interested in reproducing and analyzing this. With settings of 16s it seems that nothing is done for the first 16s and then it drops volume quickly. I had a few of this songs, where it seems that nothing gets done until the window size is over. In the mentioned song it continues with loudness changes all the time, mostly drops but then going up to -16LuFS with peaks of -6LuFS, feeling definitely too loud.

But I also had the very other side: On dancing shows they often have a beep at the beginning of a song as a start signal for the dancers. (Well often these beeps are 0dBFS even when the following intro starts with peaks at -18dBFS, which I don&*8217;t want to discuss for now.) In this case it seems to me, that the very first peak gets measured and then the volume is set according to this peak to -18LuFS. The result is a beep at -18LuFS followed by an un-hearable intro keeping its volume for the duration of the chosen window size and then raising up to -18LuFS.

Furthermore I also had songs, where almost no Loudness change was in the original over a period of time, but the processed file has a sudden drop of 3 to 6LuFS.
So for me the adaptive setting is not really usable as its results are always a surprise. It only works with low dynamic music.
I really hope that this algorithm is going to improve!

I measure with Waves Loudness Meter in Pro Tools.

According to the desired volume fader I understand it as an offset to the -18LuFS default setting. So keeping at 0dB results in -18LuFS. An increase/decrease of 1dB is an increase/decrease of 1LuFS. This seems technically correct to me. However, the term desired volume is quite confusing with a setting of 0dB in this scenario.

Another thing: I see there is a Lowpass available but no Highpass. Is this for any reason? I often get bad recordings to play at a show with mumbling speaks or so. Highpass could also be used to remove DC-Offset. Would definitely help me out some time!

Kind regards,
Gandlz

[schmidj]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

Well I use the replay gain DSP, which is properly labeled in LKFS, set to -18. It appears to work properly, the log even tells you when it couldn't set the replay gain as high as requested because of the potential of clipping. Now, I've never actually measured the result, I'd have to play a track on a player which supports replay gain through a SPDF or AES output, record it digitally and examine the loudness of the resulting file. I'm happy with the results, so I've never bothered.

Nonetheless, I looked at the Volume Normalize DSP, which actually applies gain/attenuation to the file and changes the audio. You have somewhat of a point, the user interface is not properly set up as it is in the replay gain DSP. Also the linked documentation still implies that "0" is -23LKFS, but if you read the release notes, you'll find that was changed to -18,

Also be aware that this DSP has no limiter or other dynamics (unless you use the adaptive choice), so a single gain adjustment is applied to the whole file. While the documentation doesn't discuss what happens if the calculted gain would push the signal into clipping, I'd be pretty sure that the plug-in would apply a lesser amount of gain to the file, such that the loudest sample is either set to 0 dBFS or slightly below that number.

The only reason I'd use the volume normalize function is if my player didn't support replay gain, and if that was the case, I'd keep a copy of my audio files which did not have volume normalize applied, as this effect is essentially not completely reversible.

You might have better luck running two passes, the first to calculate replay gain, the second to use "replay gain applied" which takes the value of the replay gain in the file and adjusts the level of the file (or a copy of the file if you've written the output string to write the file to a different location) so the resulting file plays at the calculated LKFS level.

[PureChuckles]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

Just an FYI. I perused through the threads and saw pretty much what I see now. "Not much". What you say above is true and for anyone that is wishing to actually use a functional Loudness Normalization, they most likely understand all of this. It is needed for anyone that uses it professionally, as we are all aware of regulations in various regions. So it is not a small deal.

The problem is using dBpoweramp, the Volume Normalize DSP function in dBp is broken. It doesn't adhere to the value... it applies the incorrect type of normalization... and on the incorrect scale.

At this time, it should be labeled as, Loudness Abnormalization.

That is why I pointed out the free utility called bs1770gain. Feel free to point out some usable solutions if you have them. Unfortunately, I have invested time in scripting dBpoweramp for past routines and I slowly need to redo things.

[schmidj]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

It would be worthwhile for people to search old threads on this subject. As I have posted here before, the choice of -23LuFS/LKFS or thereabouts in the various loudness standards was based on the levels used in professional television broadcast applications (for which the standard was written and intended). Professional broadcast audio levels in the plant are very much lower than the levels that have typically been used for digital consumer products like CDs. There are various historic reasons for this, not the least of which has been "loudness wars" in the music business. But historically, most CDs were normalized for a peak level somewhere slightly below 0 dBFS. This was typically after a fair amount of dynamic processing (limiting/compression) to reduce the dynamic range, particularly on "pop" music.

The end result of this is that if you actually take some commercial CDs and ingest the audio to an editor with an accurate loudness meter, you'll typically see loudness of -12 to -18 LKFS. This is some 6 to 12 dB louder than the levels recommended by SMPTE and now the loudness specs for professional television broadcast.

So why wouldn't you want to use -23 for your library. If you put the files with the -23 audio on your smart phone, you probably won't be able to turn the headphones up enough to properly listen to the music, particularly in Europe, which has limits on headphone loudness to protect hearing. The gain, and the limits in smartphones were based on the louder files, recorded at -12 LKFS or so.

The other issue is when playing the files on a device with more than one source like a receiver with a tuner or a car radio. You might well be able to turn the gain up enough to hear the -23 tracks at a reasonable level, but if you then switch the input to, let us say the tuner, you are going to be blasted with really loud audio from the tuner. This is because the circuit designers designed the gain structure for the levels to come close to matching with audio sources like commercial CDs or downloads, which are usually in that -12 to -18 LKFS range.

Finally, there has been some considerable discussion and research in the commercial recording business about using the loudness measurements for CD and download releases, but they all revolve around suggested levels of -14 to -18 LKFS. There have been several suggestions to standardize on -16. -18 as used now by dBpoweramp is a nice conservative choice which will play nicely with most consumer equipment with a little more headroom than you find in most pop recordings.

FYI, I am retired from a 36 year career as an audio/video systems engineer with one of the major USA commercial broadcast networks, and was responsible for overseeing the installation of loudness monitoring and control equipment there when the FCC passed a rulemaking enforcing loudness limits on television broadcasting. I also do quite a bit of live music location recording and use loudness software in processing my recordings.

[PureChuckles]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

Well this thread got pretty messed up with my delayed posts.

If anyone cares, here is an ITU R BS 1770 normalization tool. It is command line. Looks as though the developer has been at this for a while, but I've never ran across it before. Started out as R128gain, but he has re-written it as BS1770gain (no GUI for it yet, it is only CLI), So, he has the right idea in the options. It is GNU General Public License. Good solution, albeit it doesn't handle my format of audio files.

http://bs1770gain.sourceforge.net/*intro

Cheers

[PureChuckles]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

"REPLY TO POST" doesn't work for me, I'm trying this message as a "QUICK REPLY".. I have tried a couple of times with suggestions of clarity. I somewhat concur with ChrisJJ that EBU-R128 "is" -23.. anything different is not that and should be called "BS 1770". And technically it cannot be referred to as EBU-R128 just by Track (Integrated) normalization unless it passes all specifications, most notably the deviation. So unless the deviation was within spec on a track normalize, it wouldn't pass.

My initial reply attempt follows below (although I have edited some as I feel it still wasn't accurate):
-=-==-
Thanks for the quick reply. I believe you might be pointing out a mis-interpretation by the developer(s). Even though EBU states the target is "-23" (LuFS), in order to make compliant for R128 standards, from what I can see dBpoweramp is making three errors.

1) It is IGNORING any target set by the user and instead uses an embedded/set target for R128 to be the value of "-23". So, changing the value did nothing for me.
2) Unfortunately the value of -23 that s being used, is as dB instead of LuFS (ITU R BS1770). For my particular content, -23 db ended up being around -34 LuFS which is very low.
3) As a result of point 2, the Normalization process that dBp is undertaking is as a "Peak" normalization, and so only the peaks were reaching -23 (dB). By some this point and point 2 could be the same.

I'll take a stab that "probably" some type of phrasing we are looking for in the options are:

a) Peak to Peak (dB)
b) EBU R128 (-23 LuFs) ** Which is rather an "adaptive" method that calculates the result based on the R128 standards deviation within the allotted window and corrects leveling dynamics.
c) ITU bs 1770 (custom LuFS/LKFS Integrated) ** Similar to B, but allows the user to set a custom loudness level such as -18 LuFS/LKFS
d) ITU bs 1770 (custom LuFS/LKFS Adaptive) ** Similar to C, but allows the user to set a custom loudness level but includes: Window, Deviation, Attack, Release, TruePeak Max. A somewhat decent UI reference is that of Adobe Audition even though their terms are technically inaccurate, as well.
e) ATSC A/85 (-24 LKFS) ** if so desired to include as an option, which is also an adaptive method. But honestly, just having option D could solve it all... and then have profiles that spec R128 or anything else.
f) Fixed amplification

For my greedy needs, at this time, "C" is all I need. To normalize (match loudness) of a track without changing dynamics to a custom LuFS level. Again, maybe re-thinking it all and just offer option "D" as it could cover all bases, with profiles that can include R128, ATSC, Custom... whatever. Just a thought.

Thanks for the helpful comments
-Charles

** Tons of edits, sorry **

[chrisjj]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

As an aside to this discussion, note that foobar2000's ReplayGain utility also uses the EBU R128 "approach" but uses a loudness value of "-18". There is no setting to change this in foobar2000. I recall this was reported as being done because the -18 setting got close to the old ReplayGain algorithm set to its default level of 89db.

Which bit of the word standard do they not understand?? :-)

[chrisjj]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

I confirmed with Spoon several months ago that with Volume Normalize DSP 12, the EBU R128 was changed without noting it in any way in the codec, that it was defaulting to -18 instead of -23.

That's really disappointing to hear. I hope this gets addressed before too many users invest in the -18 in ignorance.

There needs to be a true EBU R128 type. Clearly the existing one now should be kept but it should be renamed accurately: BS.1770 at -18. It should not be allowed to remain, offering EBU R128 but secretly delivering the non-standard variant.

[garym]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

If it has, then the labelling needs changing, because the whole point of EBU R128 as a standard is that the loudness value is -23. A setting of -18 is not EBU R128. It is just BS.1770 at -18.

BS.1770 is the measure. EBU R128 is a specific value.

As an aside to this discussion, note that foobar2000's ReplayGain utility also uses the EBU R128 "approach" but uses a loudness value of "-18". There is no setting to change this in foobar2000. I recall this was reported as being done because the -18 setting got close to the old ReplayGain algorithm set to its default level of 89db.

[jfkaess]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

I confirmed with Spoon several months ago that with Volume Normalize DSP 12, the EBU R128 was changed without noting it in any way in the codec, that it was defaulting to -18 instead of -23. I had already ripped hundreds of CD's at -23 as it was in Volume Normalize DSP v11, and want to keep all my CD's ripped to the same level, so the stealth change to -18 was disconcerting, and now i have to manually change the setting to -5 in order to keep volume levels the same between CDs/

I still feel very strongly that the Volume Normaize DSP should clearly indicate the actual level LUFS being used.

[chrisjj]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

Instead of having to know (and this is not shown anywhere in the DSP) that the LUFS value for EBU R128 used to be -23 and now has been changed to -18,

If it has, then the labelling needs changing, because the whole point of EBU R128 as a standard is that the loudness value is -23. A setting of -18 is not EBU R128. It is just BS.1770 at -18.

BS.1770 is the measure. EBU R128 is a specific value.

Why not have it clearly marked what the LUFS value is in the DSP with a slider which allows you to change the actual LUFS level and showing the LUFS level the DSP is actually configured to?

I suggest:

Title: Volume/loudness normalisation

Type: BS.1770 loudness (track)
Desired loudness: {slider} {number} LUFS

Type: EBU R128 loudness (track)
Desired loudness: -23 LUFS

[chrisjj]

Re: Loudness (ITU BS 1770 / EBU R128) Normalization

> Also I think the Volume options' slider units would be better stated as "dbFS".

Oops. Of course that should be "dBFS".