Exporting song multiple times - levels different each time ?

[Spells]

Hi all,

I am sure I've encountered this before but I'm actually trying to dig deeper on the issue.

I have a song. Quite a massive session. I export the song, and feel it's lost a little edge. As an experiment, I export the song again, it's as it should be. Open up RX and some other tests, and on what should be identical exports, it's showing as much as 1db difference in LUFS value.... meaning the exact same session can produce an export that could be 1db louder on Spotify than the other!

Now... I know that analogue modelling plugins will not be the same every time. But I've even tested this with everything on the master bus bypassed. I can get up to 1db in LUFS difference.

Why is this??? It's worrying because it takes the confidence in a session export away. It makes me want to export over and over until I have one that will perform best. I can understand 0.1 db difference perhaps... but 1db?

[EnochLight]

Hi all,

I am sure I've encountered this before but I'm actually trying to dig deeper on the issue.

I have a song. Quite a massive session. I export the song, and feel it's lost a little edge. As an experiment, I export the song again, it's as it should be. Open up RX and some other tests, and on what should be identical exports, it's showing as much as 1db difference in LUFS value.... meaning the exact same session can produce an export that could be 1db louder on Spotify than the other!

Now... I know that analogue modelling plugins will not be the same every time. But I've even tested this with everything on the master bus bypassed. I can get up to 1db in LUFS difference.

Why is this??? It's worrying because it takes the confidence in a session export away. It makes me want to export over and over until I have one that will perform best. I can understand 0.1 db difference perhaps... but 1db?

There are way too many variables to even respond to this without knowing more. Are you using dithering in Reason? What sample rate and bit-rate are you exporting at? What file format? What devices/plugins are you using and what effects? Any VST? Etc...

[Spells]

No dither. All exports at session spec 24/48.

Lots of plugins and VST's. Some UAD analogue modelling bits.

But nothing that can cause a big difference.

[Vincent Vitellius]

Having a distant past in IT support, I can give you a basic strategy for finding the culprit:

Remember to take notes while troubleshooting.

First do a basic sanity check...
Create a new empty project, add one basic synth playing a few simple notes. Does that exhibit the problem? It shouldn't.

Now you need to find out where the problem is located in your problematic project.

First bypass your mastering processing. (EDIT: I see you already did that)
If that eliminates your problem, then you know the problem is in your mastering chain.
If not, continue to mute half the tracks and test. Then the other half.

If only one half of your project is problematic, continue to divide into two halves until only the problematic track is left.

If this strategy doesn't work, for example because several tracks contributes to the problem, then try one track at a time.

If you identify one track that is buggy when solo'ed, then continue to bypass half the devices on that track. Rinse and repeat.

At some point you should be able to eliminate all the stuff that doesn't contribute to the problem, and whatever is left is your answer.

If I had to take a wild guess, you probably have a device with an option set to "random"

[ProfessaKaos]

Do you have any plugins that over sample upon export (so real time no over sampling or less over sampling real time)?

Here's a video I found which somewhat explains that dynamics plugins with over sampling can catch intersample peaks better (because it will actually see and detect them) than a dynamics plugin without over sampling. So my thinking is maybe you have some dynamics plugins which have an option for over sampling upon export. It might be a far reach, but worth knowing about.

[avasopht]

All it takes is some slight randomness anywhere.

You can bounce to wave at each significant stage of processing to essentially freeze the phase/peaks.

If you want to diagnose the cause, freeze everything, and then unfreeze half of the track to see if you see the LUFS difference, and then the other.

That might help you narrow down where it's happening.

[selig]

One other question; you say the difference is 1 dB - do you mean 1.0 dB EXACTLY (to the tenth of a dB)?
Does the metering you’re using show decibels to the tenth of a dB? If not, it only takes the smallest of changes to jump from 0 dB to 1 dB since for all we know the original level was 0.5 dB and now it’s 0.6 (and so rounded up to 1.0).
Otherwise, the bigger the mix the more likely a bunch of random variation of phase will combine differently and result in changes.
And more importantly, remember this number is an average over the entire song, so who knows where that 1 dB change happened. It could be there is ONE moment in the entire song that is different, or it could be the entire song is shifted 1 dB (less likely).
And if this is what is happening, it will in NO WAY affect what you hear. In fact, given the choice I’d use the lower level version and bump it up a dB, giving you a ‘free’ extra dB!
As a fun experiment, try doing a null test with the two files, it may or may not reveal what is happening. And it WILL reveal where it’s happening. And remember a null test only tells you there is some ‘difference’, which could be timing, level, or phase based. And it’s only the level differences (overall or at a specific frequency) that would be audible, despite the inability to null). Meaning, a null test is limited in what it tells you but can still be helpful in the big picture.