Error/Change in Amplitude Statistics

[-163.77 dB]

That was a long break - hello again, everyone! I came back with a very technical question.

I was trying to replicate Audition's amplitude statistics in MATLAB, and I spend quite a lot of time trying to figure out why I was getting different results. Sometimes just a bit different, but bigger than one would expect from rounding errors, e.g., 0.8 dB, sometimes huge - see Min RMS below. I compared my calculations to what AA3.0 reported. Around 2:30 am (it's 3:01 am now...) I analyzed the same file with all other versions which I have on my computer - CE2K and AA1.0.  And it turned out that the old versions produce identical (if I don't care about rounding errors) as my procedure - it's AA3.0 which behaves differently.
Here is an example, same file, same (I think!) settings.

AA3.0:
   Mono
Min Sample Value:   -6744
Max Sample Value:   3566
Peak Amplitude:   -13.73 dB
Possibly Clipped:   0
DC Offset:   -.003
Minimum RMS Power:   -163.77 dB
Maximum RMS Power:   -24.19 dB
Average RMS Power:   -39.85 dB
Total RMS Power:   -35.25 dB
Actual Bit Depth:   32 Bits

Using RMS Window of 35 ms

AA1.0, CE2K:
   Mono
Min Sample Value:   -6744
Max Sample Value:   3566
Peak Amplitude:   -13.73 dB
Possibly Clipped:   0
DC Offset:   -.003
Minimum RMS Power:   -64.92 dB
Maximum RMS Power:   -24.19 dB
Average RMS Power:   -39.93 dB
Total RMS Power:   -35.25 dB
Actual Bit Depth:   32 Bits

Using RMS Window of 35 ms

Other files sometimes show differences, sometimes they don't. if there is any pattern, usually Min RMS is smaller or much smaller (down to - inf) in 3.0, and it is often detected at the very end of the file, which does not make much sense. Often, but not always there is a difference in the Ave RMS or Total RMS. It is as if 3.0 thought that there is some additional silence at the end?

I am not sure at what stage between 1.0 and 3.0 something changed, I never had 1.5 and I deleted 2.0 bloatware form my machine.

But can anyone explain the difference or give me a name of a person at Adobe who could give me the answer? Is it a bug in 3.0, or does this change has a reason? What reason?

Is it a bug in 3.0?

Yes - it's known that the stats are bugged. There's a thread about this on the Adobe forum where Younglove has discovered that effectively, there are selection errors. And it has already been bought to the developers' attention. There are some circumstances where errors are almost inevitable, in fact.

Thanks, I should have checked that forum. How could they screw the Statistics Box, which basically has not changed since ce times? Oh, it is barely faster... someone overoptimised the algorithm 

The other thing is that the fade-on-mute thing that you mentioned in that thread does depend on the smooth edit boundaries checkbox. When it's cleared, the cut is perfect, see the picture. Which I think is good - I don't want the software to do smoothing fades and such when I don't want it to.