Analyze and normalize loudness with SWS

[Breeder]

Originally Posted by nofish

^^ This is quite clever.

Hey, while you're here. I see you added high precision mode at some point to the code.

What was the problem in the first case? All can see in code is that it changes the amount of samples taken from REAPER? If high precision mode fixes something here, why not just use it as default and remove the option from the menus completely? It doesn't seem that much slower here.

[nofish]

You can read about it here, some of the 1770-4 test files didn't pass in normal mode but do in high precision mode.

The explaination is here.

As that's a cornercase imo (very short files with offsetted audio blocks) I added is as additional high precision mode and not as default, as in majority normal mode works just fine.

(Besides things like generate envelope, go to max. short term / momentary don't work correctly in high precision mode and would need a code rewrite.)

These are the EBU testfiles that are supposed to check out with -23.0 max. momentary:

[Breeder]

Originally Posted by nofish

You can read about it here, some of the 1770-4 test files didn't pass in normal mode but do in high precision mode.

The explaination is here.

As that's a cornercase imo (very short files with offsetted audio blocks) I added is as additional high precision mode and not as default, as in majority normal mode works just fine.

(Besides things like generate envelope, go to max. short term / momentary don't work correctly in high precision mode and would need a code rewrite.)

These are the EBU testfiles that are supposed to check out with -23.0 max. momentary:

Thanks for a prompt answer - I did notice different results with some other files too...they weren't short (I think they were a few seconds long). That's why I posted in the first place

Originally Posted by nofish

(Besides things like generate envelope, go to max. short term / momentary don't work correctly in high precision mode and would need a code rewrite.)

Ok, this is good to know cause one wouldn't expect it. I thought analyse code simply wrote results to a class member variable so if working with high precision mode, that result should be used throughout the entire code.

Can you just explain for future reference what happens when one has high precision mode turned on with these things (go to max. short term, envelope etc...)?

[nofish]

Originally Posted by Breeder

Ok, this is good to know cause one wouldn't expect it. I thought analyse code simply wrote results to a class member variable so if working with high precision mode, that result should be used throughout the entire code.

Thing is in 'normal' mode you took a measurement every 3 sec. / 400 ms for short-term / momentary and used these results for creating the graph and go to short-term / momentary positions, the high precision mode uses overlap measuring (much shorter buffer) so this approach would break.

Can you just explain for future reference what happens when one has high precision mode turned on with these things (go to max. short term, envelope etc...)?

https://github.com/reaper-oss/sws/issues/1120



These features are disabled when using high precision mode in next SWS version (for now).

[Breeder]

Awesome, thanks man

[Handyman]

Recently, whenever I use SWS Loudness “Go to maximum short-term” and “Go to maximum momentary” it highlights areas beyond the end of my audio tracks, where there is no audio at all.

Any idea how I can reset this, and make it work properly again?

[nofish]

Originally Posted by Handyman

Recently, whenever I use SWS Loudness “Go to maximum short-term” and “Go to maximum momentary” it highlights areas beyond the end of my audio tracks, where there is no audio at all.

Any idea how I can reset this, and make it work properly again?

Sorry, just saw this now.
There's a bug in current SWS (v2.10.0 #1) when analyzing with high precision mode on and creating graphs / going to max. short-term / momentary.

So for now, turning high precision mode off for analyzing should fix.

Sidenote:
There's currently also another bug which could lead to a crash when analyzing items shorter than 3 seconds (fixed for next SWS version).

[Handyman]

Originally Posted by nofish

Sorry, just saw this now.
There's a bug in current SWS (v2.10.0 #1) when analyzing with high precision mode on and creating graphs / going to max. short-term / momentary.

So for now, turning high precision mode off for analyzing should fix.

Thank you. Turning off high precision mode worked!

[PelPix]

Breeder,

I'm using this to even out loudness between voice over clips in a project I'm working on, and I'm running into inconsistent loudness. I used to have the same problem in iZotope RX, so it's just inherent in the math, but interestingly enough they just rolled out a fix for this in RX 8: "Program loudness gate." It gates the loudness calculation (I.E. it discards parts of the clips that are too quiet (in momentary loudness) from the loudness calculation.) This is way more useful for program material that consists entirely of voice!

Apparently this is part of the new set of revisions to the B.S1770 standard

Documentation:
https://www.itu.int/dms_pubrec/itu-r...0-I!!PDF-E.pdf

Basically signal below -70 LKFS momentary and over 10 LU below a rolling average of the momentary loudness (short-term maybe?) are discarded from the loudness integration.

Any chance you could implement this too?

Thank you for all your hard work!

Edit:

It looks to already be implemented. Interesting...I wonder why I'm getting inconsistent loudness depending on the amount of silence in clips?

[nofish]

Originally Posted by PelPix

It looks to already be implemented. Interesting...I wonder why I'm getting inconsistent loudness depending on the amount of silence in clips?

Yes, current SWS Loudness measurement should be ITU-R BS.1770-4 compliant (including gating).

So your issue lies somewhere else I'd assume.
Make sure you use latest SWS version (v2.12.1 #3 currently) as there have been fixes to SWS Loudness measurement.
Then if you happen to analyze rather short files you can try to enable 'high precision mode' (see https://forum.cockos.com/showthread....98#post2118098).

If issue still persists please open a new issue on the SWS tracker, ideally with steps (test project) to reproduce.
https://github.com/reaper-oss/sws/issues

[djabthrash]

Originally Posted by Kihoalu

(...)

Are the "momentary" and "short term" measurements RMS?
and, if so, what are the RMS time intervals for each?(...)

+1

(been using SWS loudness during gain-staging to measure peaks, but realized i should also look out for RMS)

[nofish]

It's described in ebu tech3341 doc:

The momentary loudness uses a sliding rectangular time window of length 0.4 s. The measurement is not gated.
The short-term loudness uses a sliding rectangular time window of length 3 s. The measurement is not gated.

So yes, it's basically RMS calculation + the Loudness weighting filtering (described in ITU-R BS.1770-4).

[djabthrash]

Originally Posted by nofish

It's described in ebu tech3341 doc:



So yes, it's basically RMS calculation + the Loudness weighting filtering (described in ITU-R BS.1770-4).

Thank you !

After some more thought and research, i think i'll be better off adjusting my levels by looking at the meter visually, and then using SWS loudness on all my tracks simultaneously to make sure they don't peak too high (and also have a look at the "RMS"-ish value in SWS loudness, which won't hurt).