Also report peaks as sample peak relative amplitude

[UltraFuzzy]

By "sample peak relative amplitude" what I mean is: sample peak: value is taken from an actual sample relative amplitude: as a fraction of the maximum possible value instead of dBFS

Providing this number allows comparison with EAC and XLD logs and this number is commonly used as a basic "fingerprint" to help determine if a disc is a distinct mastering.

This is almost in good shape to be merged although you'll need to tell me how you'd like this reported in the logs. Right now it's very programmer-oriented:

  Sample peak:
    Peak:       -0.3 dBFS

  True peak:
    Peak:       -0.3 dBFS

  Sample peak relative amplitude (calculated from ebur128 dBFS):
    Peak:        0.965973

  Sample peak relative amplitude (precise):
    Peak:        0.965973

  True peak relative amplitude (calculated from ebur128 dBFS):
    Peak:        0.970680

I started off just calculating from the ebur128 dBFS, but that result is sometimes slightly off compared to XLD, presumably due to rounding error. I then added "precise", which is a simple iteration over all the samples, which as far as I've seen always matches XLD.

I'm not quite sure how best to tidy up the log. The two values that matter here are one of true peaks and Sample peak relative amplitude (precise). Maybe show the values for only True peak relative amplitude (calculated from ebur128 dBFS) and Sample peak relative amplitude (precise) and clean up their names?

Thoughts?

[cyanreg]

I don't understand this at all. Precise? calculated from ebur128 dBFS? And they match?

[UltraFuzzy]

I changed the ebur128 ffmpeg filtergraph configuration string so that it calculates both sample peak and true peak.

"ebur128=peak=true+sample,anullsink"

It gives these in terms of dB. To convert those values to relative amplitudes I used the formula amplitude = 10^(dB/20). Those are the "calculated from ebur128 dBFS" values. That turned out to sometimes have significant round-off error so I added the "precise" method for sample peak which is just a loop over the samples to find the one with greatest absolute value and then return that as a fraction of abs(INT16_MIN), i.e. the largest possible amplitude for 16-bit audio samples. The two methods for sample peak sometimes match but not always. The sample peak values I'm getting from the "precise" method match other rippers.

True peak in dB is what we currently have in the log and what I'd like to add is the value I labeled "Sample peak relative amplitude (precise)" since this is reported by other rippers and is commonly used as a crude "fingerprint" of a disc to identify distinct masterings. I'm unsure of how to nicely report both of these in the log because of the differing units. I was thinking maybe report both of them as relative amplitude. So from the example above, we'd report just those last two values as something like this:

  Sample peak: 0.965973
  True peak:   0.970680

But I wanted to get your opinion.

[cyanreg]

The names are absolutely terrible.

I started off just calculating from the ebur128 dBFS, but that result is sometimes slightly off compared to XLD, presumably due to rounding error. I then added "precise", which is a simple iteration over all the samples, which as far as I've seen always matches XLD.

Then why keep the ebur128 dBFS?

[UltraFuzzy]

The names are absolutely terrible.

lol. I agree. I found these tricky to give good names to and I didn't mean to suggest that this get pulled with those exact names still there. I wanted to get your opinion on what to write to the log so I opened the pull request with the guts still hanging out.

Then why keep the ebur128 dBFS?

Yeah, the sample peak amplitude calculated from ebur128 dBFS is junk and can be tossed, but the true peak amplitude calculated from ebur128 dBFS may still be relevant so for now I kept its sample peak counterpart around just for the sake of symmetry. Again, things are in a very guts-hanging-out state because I wanted to get your opinion on what to include in the log before condensing it.

[UltraFuzzy]

What about this for the log format?

  Sample peak:
    Peak:       -0.3 dBFS

  True peak:
    Peak:       -0.3 dBFS

  Sample peak:
    Peak:        0.965973 (amplitude)

  True peak:
    Peak:        0.970680 (amplitude)

It's a little clumsy but it wouldn't require touching the way that ffmpeg writes to the log.

For reimplementing the ffmpeg printouts ourself, I'm not sure if there's a way to access the two values labeled "Threshold" since they're not in the in the AVOptions array for ebur128. I'm not familiar with using ffmpeg as a library though so I could easily be missing something.