Okay… I just did a real quick study across 3 albums, 33 tracks, comparing what SoundCheck thinks the track level is, versus what R128 thinks it is. The SoundCheck estimate is taken as the negative of the value shown in the Info/File/Volume field in iTunes.
Very preliminary, rough, first cut by hand… but probably already good enough to judge from. And the results do seem to match my perception.
I wouldn’t be too troubled by the apparent slope not matching unity. First of all, this is a crude study, and the uncertainty in that estimate of slope is probably large enough to contain unit slope.
The graph shows that the variability around the best fit line is about plus or minus 1.4 dB. And that is reasonable to expect, given that SoundCheck uses a very different method for estimating perceived loudness, compared to R128.
What we are really comparing here, is the consistency in estimating perceptual levels. It doesn’t matter if there is a difference, so long as it is predictable. And the fact that these data line up, more or less, along a best fitting line, tells us that – yes! it is consistent.
The “important” number to take away from here is the intercept, which shows that when SoundCheck thinks the music is at its target level (0 dB SoundCheck Volume), our R128 system thinks that track is at 7.9 LU23, or about 8 dB hotter than the R128 target level of -23 LUFS.
And that intercept really does match my average expectation. I think that listening to SoundCheck controlled playback in iTunes sounds about 8 dB too hot for my system, which I had calibrated for reasonable audio levels at -23 LUFS.
And that’s okay! It just means that when I play back through iTunes, I should drop the input fader by 8 dB. And with SoundCheck in control of the playback levels on varying track levels, I will be okay with that fader set to its constant attenuation setting.
I will see variations of around a dB or so, between what I would have had with R128 normalizing, and what SoundCheck will give me. But which one is really correct? eh? (ans: neither one!)
[ And, what! you say? Look at that sample in the lower left. It is dragging the slope too much – too much of a lever arm on the rest of that data…
Okay, fair enough, here it is again with that “outlier” point removed…
… so, nothing really changes in my estimate. SoundCheck has a target level around 8 LU23. Just drop the input fader by 8 dB, and forget it.
And that settles a long running string of wild estimates about where the SoundCheck target level really is. It is at 8 LU23, or -15 dBFS. ]
[ Hey! How did you manage to get a consistent best fit, despite that outlier, and with all that noisy variation in the measurements?
I used a technique offered by another Astrophysicist named Peter Stetson, of the Dominion Astrophysical Observatory in Vancouver, B.C. This fitting method uses iterated weighted least-squares fitting, with deemphasis of relative outliers in each pass, until the situation settles down. It doesn’t throw points away. It just de-weights them according to how far from the fit they fall. Very clever idea of his! ]