My best argument against DBT

I come to praise audio objectivism, not to bury it.

But I can't help myself, I can now describe the flaw that I've been thinking about for decades.  I claim this only as a precautionary principle, why "good enough" is never actually good enough, why we must do at least a little better, as we can.

DBT generally requires consistently audible differences.  That's great, and it's useful to find out where those are.

But consistently audible differences aren't the whole story, by a long shot.

Every audio experience is different, even with the same music, gear, room, everything.  Hearing the same music twice is never exactly the same experience.  It cannot be.

Furthermore, each experience contains a gazillion "percepts" moments in time or correlations among moments in time.  These are never exactly the same sequence.

Now if you looked closely enough at the generation of precepts, if you could, you would see that the difference between percept A and percept B may be extremely small.  A particular underlying sound level might be just fractionally away from falling from A to B.

Those are the differences that might be caused by inconsistently audible differences, differences too small to be consistently audible, but sufficient to flip from one out of the universe of percepts to another.

Such a small change in sound however only changes the over time distribution of percept A vs B at that moment.  It does not consistently change it to A or B.  The entire auditory experience become a story about the actual perceptual sequence, a very incomplete story.  But a tiny change in one tiny percept could change the whole story, in the way a single voter will, on occasion, change an entire election, and according to economist John Quiggen, it happens often enough to justify voting.

So if you are perfectionist, or just a show off, you want the best you can reasonably get.  If you just want something good, you still get something a bit better than absolutely necessary.

Besides the inconsistent sampling issue, there also just the simple issue of type one vs type two error.  There is no way of ruling out that small differences might be 100% unimportant.  You may be willing to accept a 5% null rejection or a 0.01 null rejection.

However it does get interesting when you get to proportion of effort questions.  Most of us should probably not be thinking so much about digital conversion, which is a nearly perfect process, I would stand with the objectivists on this.  Speakers and room acoustics are still the #1 issue, and deserve most of our time.  When phenomena are 3 orders of magnitude better in typical audio performance than what is consistent audible, you really have to wonder if things like that can be very important.

The variability of experience makes it essentially impossible pursue audio improvement strictly through listening.  Virtually all audio designers admit this.  They start not with listening but with ideas, and measurements are usually a key part of refining those ideas, with listening when they get good enough to satisfy the ideas of the designer.  Subjectophiles sometimes themselves flip from one view of the world to another based on a single highly unreliable (and mostly likely incorrectly performed) subjective test.  That is, of course, insanity.  All listening experience, if they are counted at all, should be counted as unreliable evidence without DBT.  Electronic measurements can be quite reliable nowadays, with the best measurements expressed as FFT graphs up to 40kHz or higher and -140dB noisefloor.   It's hard to argue measurements like those are missing something audible.  Single numbers have only very limited use.

If measurements appear to be 3 orders of magnitude better than we need, can we feel safe?  Well mostly likely yes, if they are the right measurments.  Now we can easily expore the effects of jitter on the noise and distortion spectra.  Usually the differences appear far far below what would seem to be audible.  That wouldn't actually be enough in itself to discount the audibility of jitter, because it might introduce audibly distinct (with all the caveats above) time differences.  So actual DBT is necessary, we have a limited amount of that but a threshold around 10-100nS hasn't been overthrown, to my knowledge.  That still doesn't eliminate the desire to have better levels if not hard to do.