Why are "Amplitude Statistics" and the "Match Loudness" analysis are different for the same file?

[unauthorizedrob]

Trying to even out chapters in an audiobook.  I have one "gold standard" chapter the producer considers perfect, against which I am comparing all the others.  I am trying to understand the differences which make one chapter better than the other, and also looking for a way to apply the settings for my gold standard to all the others.  I have tried "Match Total RMS to File," not successful.  Now I just ran "amplitude statistics" on the gold file, and they are DIFFERENT from the analysis results in Match Loudness. (e.g. total RMS is 16.60 in Amplitude Statistics, and 19.62 in Match Loudness).

Why are these different?  

What can be done?

Thanks.

Rob

[SteveG_AudioMasters_]

Of course they are - there are nine different algorithms that you can use in Match Loudness (see Match Loudness Settings), and they'll all give you different results. In Match Loudness, if you select Total RMS and set the Target loudness value to 16.6 and hit 'Run' then it will return a total RMS value of just that!

Which should of course give you a bit of a clue as to what you might try...  Analyse the known-good file using Amplitude Statistics, and transfer these settings to Total RMS, put all your chapters in, run it, and they should all come out to the same RMS value.

Analysing files using things like RMS settings isn't simple - there are a number of factors to take account of, and no absolute defined and agreed reference point, so you are always likely to get variations.

[unauthorizedrob]

Hi Steve

thank you.    I did that.   The problem is that all files need to meet that ACX requirement, of 18-23 RMS and -3 peak.  In Match Loudness, the ”good” chapter is Total RMS of -22.69, Peak limited to -3.  They ALL have to fall in that range.

i should have thrown up my hands months ago.  This is really beating a dead horse.  But we have no alternativ;  in too deep and no funds left.  Sorry to drag personal stuff in here, but it colors everything I do.

is there any chance you could listen to two chapters?  I can send you a link to Dropbox.

Thanks.

[personal email address removed - mod]

[SteveG_AudioMasters_]

unauthorizedrob  wrote

Hi Steve

thank you.    I did that.   The problem is that all files need to meet that ACX requirement, of 18-23 RMS and -3 peak.  In Match Loudness, the ”good” chapter is Total RMS of -22.69, Peak limited to -3.  They ALL have to fall in that range.

If all the files sound similar enough, it really is only a matter of adjusting the overall level of each one to get them into range. If a file gives a result that's 3dB too high (-15dB) then all you have to do is lower the level of the file by 3-4dB and it will be within the correct range. If it's under, do the opposite. If the peaks go too high, then run a limiter on them - you almost certainly won't have to do too much. Alternatively, try the built-in method - it might even work...

Browse for your reference file where it says 'file' and in the drag and drop section, put the rest of the files. Hit Run, and save the files. When you check them, they should all be within range.

[unauthorizedrob]

I did that, too.

i am pretty much at the end of my rope. 

The problems are in the files before the Match Loudness.  I got some devastating notes, same as weeks ago, the files do not sound right to the producer, and I have no idea how to make them right.

[SteveG_AudioMasters_]

What have you actually done to the files? Do you have the original recordings intact?

[unauthorizedrob]

I have done a myriad of things, tried for months. 

Yes, original session files are intact.  I have gone back to them in several cases.

Initially, cleaned noise in hundreds of interstitial spaces, replaced spaces with silence, then tried to use noise reduction process to remove bg noises, like intermittent air conditioning, from under words.  Errors in those settings caused voice to have a slight ring, so restarted, tried re-recording sections to repair errors, couldn’t match sound quality.  Used notch filter to remove hundreds of sibilant high pitched whistles. 

When I got notes from the author, several chapters were adjudged “muffled” or “echoey,” so it was back to square one.  I sat with the producer/voice artist, who invariably said the original files sounded fine, but once they got to the edited stage, the notes were the same.  Truth is, I don’t know what I did.  I used your Clarity filter, and tried different variations with the parametric eq to try to correct for too much high end or too much bass, which is how I interpreted “muffled.”  I used DeHummer at 120 Hz to try and brighten the sound. 

[SteveG_AudioMasters_ · Accepted solution]

A lot of those things just won't work properly if you overdo them. The DeHummer absolutely won't brighten the sound, either! Overdoing NR often results in sort-of 'echo-y' or 'underwater' sounds, and this always works better if you do multiple passes at different FFT settings, and don't take very much out each time. If it was basically a good recording in the first place, then doing things like filtering anywhere within the audible range is inevitably going to alter the sound significantly, and unless you are really fortunate, you'll end up losing just as much wanted signal as unwanted - effectively achieving nothing.

The other thing that can make a huge difference to what you do is what you are monitoring on. If you're doing speech, you need two things; a decent set of monitor speakers, and a decent pair of headphones. Just having the monitors isn't enough - you have to learn what they sound like with known-good recordings before you'll really know what they're doing to yours. The headphones are good for judging NR on - the artifacts tend to be more obvious using those, unless you have some really serious monitors.

When you process stuff, make a note of what you did, and save intermediate steps in the processing as separate files - much easier to go back a stage if you need to. When you have a result that's signed off, then you can ditch the intermediates.

I currently have too much work on, so I can't really help you directly at present (I tend to look at the forum whilst waiting for processing to happen, as a rule - leveraged time!).

[ryclark]

Can you post a reasonable length clip from one of your original files in uncompressed .wav format somewhere like Dropbox or Google Drive for us to take a listen to? We could then possibly advise the minimum that you need to do to make the audio acceptable.

[Datta Groover]

SteveG, thanks for your help with this, but I belive there is a glitch somewhere. I have the exact same problem, and the Ampitude statistics SHOULD show exactly what the analysis of Match Loundness shows.
The question is complicated by the noise redution issues unauthorizedrob is having, but the problem is in that discrepancy between the two systems of analysis within Audtion for the SAME sound file.

unauthorizedrob, I would go by what Match Loudness says, and upload. THat is what I've done, taking my peaks down first with a hard limiter (if you change the peaks after it will change the total RMS), then matching to -17db Total RMS.
So far ACX has not rejected any of my files.

[Datta Groover]

Looks like it WAS a glitch in Audition 2019.  I updated to Audition 2020, and there is now no discrepancy between Amplitude Statistics and Match Loudness Analysis.

[Datta Groover]

It appears Amplitude Statistics was using the perceived RMS value, not the actual total RMS value.