Normalize Audio Level - EBU R128 - broken in 17.2

In early Version 16 the Normalize Audio Level function in conjunction with the EBU R128 Mode was reliable in both Relative and Independent Mode. When I did some double checks in other EBU R128 capable tools like LM-Correct or Adobe Audition there were no issues detected with the Resolve Files.

But in early Version of 17.x I had issues with the relative normalization with multiple Clips, since they would fail the EBU R128 restrictions afterwards. But I had a workaround. If I made a mixdown in Fairlight first and then apply the same correction only on this one Mixdown, then it still worked. Until 17.2 came. Now I can not produce any correct result. I believe the Clips end up 6 or 9db louder than they should be.

I am using Resolve 17.2 on Windows 10 (with automatic Updates active) and up to date Nvidia Studio.
Like I said, I am only delivering in EBU R128. I have not tested any other possible Norm so far.

Whenever I see something be off by 6dB I suspect pan law errors (or misunderstandings).

Annoying question then, but; have you double-checked routing and controlled for those sorts of things?

Thank you Mattias for your Reply.

Yes. Might be a misunderstanding on my part. The strange thing would be that I have been lucky with EBU128 deliveries for a long time and only now things are different for some reason. I do not know what it could be.

The part with the Pan Law is interesting. I was not aware about that issue but as far as I am able to understand it after some research, it seems to be limited to speaker reproduction alone? Because I am more concerned with files and EBU Reports that are not aligned to what Resolve is creating for me recently. So I might have wrong settings, a bad workflow or there is indeed something broken in Resolve.

To illustrate my issue, I created a Video where I showcase my workflow and how I and up with bad EBU128 Files.

(on last remark! I was using OBS with Autoducking to record this Video. And due to strong Autoducking settings the System-Sound is sometimes fading Up and Down whenever I breathe into the microphone while playing sound. Not good)

Pan Law is not applicable.

EBU R128 specifies a True Peak Limit of -1dBTP and the Preset in the "Normalise Audio Level" window therefore uses a fixed True Peak target which is not user definable.

For broadcast compliance, use the "ITU-R BS.1770-4" preset and specify a -3dBTP target level.
The Target Loudness defaults to -23 LUFS.

Via the Timeline menu > Bounce Mix to Track. Bounce to a new Destination Track.

Does the "Normalise Audio Level" function's "ITU-R BS.1770-4" preset produce the required True Peak and Loudness Target when applied to the bounced Audio Clip?

Use the "Analyse Audio Levels" contextual menu item on the bounced Audio Clip and confirm compliance.

Does the "Normalise Audio Level" function's "ITU-R BS.1770-4" preset produce the required True Peak and Loudness Target when applied to the bounced Audio Clip?

That look like the perfect workaround for me. I imported it into Audition, and it told me that it fits my custom EBU R128 (Target Loudness: -23 LUFS, Tolerance 0,5 LU, Max True Peak Level: -3 dBTP). That is great.

So that's a good quick fix for me, but I would like to point out to the Blackmagic Team that EBU still seems to be broken and even more important (at least on my system) Relative - Normalization is not working and produces random results in at least EBU and ITU-R BS.1770-4.

When I look into the manual at page 786 about "To normalize one or more selected audio clips:" it says

When Set Level is set to Relative, all selected clips are treated as if they’re one clip, so that the highest peak level of all selected clips is used to define the adjustment, and the volume
of all selected clips is adjusted by the same amount.

for me that sounds like a temporary bounce file is created that gets the Normalization treatment. The dB correction is then applied to all selected clips Or in other words it is similar to an automated version of Reynaud Venter proposed workflow.
But here "Relative" fails. If I export the Result of the Relative "ITU-R BS.1770-4" mix, it is off by +2.4dB according to Audition. And even Resolve agrees that the mix is incorrect. If I create a bounce based on this relative mix and apply another "ITU-R BS.1770-4" Resolve is applying a +2.4dB correction on the Bounce Clip. Exactly what Audition would do too.

Pan Law is not applicable.

What I was getting at was that if there was a routing error pan law would absolutely apply. In other words if for some reason the stereo signal was panned center by mistake the level would be different.

I created a Video where I showcase my workflow and how I and up with bad EBU128 Files.

for me that sounds like a temporary bounce file is created that gets the Normalization treatment. The dB correction is then applied to all selected clips Or in other words it is similar to an automated version of Reynaud Venter proposed workflow.

"When Set Level is set to Relative, all selected clips are treated as if they’re one clip,
so that the highest peak level of all selected clips is used to define the adjustment,
and the volume of all selected clips is adjusted by the same amount. This is good if
you have a series of clips, such as a dialog recording, where the levels are consistent
with one another, and you want to normalize all of them together."

I don't read it the way you do (i.e. no intermediate bounced file). If the process that the software uses is simply replacing "highest peak level" with "highest loudness", then I wouldn't be surprised if it simply took the loudest individual clip it could measure and offset it to hit your target, then offset all other clips by the same amount.

That would mean that the summed value you see at the output would be higher by definition.

I could be wrong of course, but it actually seems like a quite likely way they could have coded this. Instead of measuring peak in step one, just use loudness. Then the rest follows as usual. That would give a result that might seem counterintuitive.

Personally I think the way to go about this is to mix to target as you're working. Don't go back and tweak clips this way to satisfy specs. And I mean this regardless of whether or not clip normalization worked the way I think you want it to work.

PS (added): One reason for not doing it the way you are is that if you have any dynamic processing AFTER the clip-level you now change how it behaves. In other words if you have a compressor or limiter on a bus, or maybe noise reduction sensitive to input level, all that stuff then reacts differently after you've mixed your content because you changed the input into those plugins.