What are “taps” in a digital context? I’ve never heard of them before.
PCM digital audio is governed by the sampling theorem (
https://en.wikipedia.org/wiki/Nyquist–Shannon_sampling_theorem), and that latter formally proves that the reconstruction of the sampled signal will be 'perfect' in the band below Fs/2 when the Sinc(x) function (or sin(x)/x) is used as the reconstruction (low-pass, anti-imaging) filter:
https://en.wikipedia.org/wiki/Sinc_function
The sinc function is a single sinusoid peak with wiggles to both sides of it in time, damping out the farther removed one is from the center peak. Sinc stretches back from the Big Bang to Armageddon.
In digital (low-pass) fitlering, using oversampling, Sinc is approximated by storing a number of the function's values along the time line. Each such value can be named a 'tap'. (Not exactly so, but you get the idea.)
Is there some technical conformation you can point to with regard to diminishing returns and the point that starts to occur?
Wholly agree it’s a marketing gimmick at this point but equally I don’t see the evidence for the above assertion.
It is an obvious and trivial fact from mathematics. I doubt anyone would even have done the effort of writing this out to laymen.
As the function values of Sinc quickly approach zero when moving a way from the peak, original samples that far from the peak will ever contribute less to the reconstructed output signal. The original Philips 4 x oversampling filter in the early 80s had a few tens of taps, perhaps slightly over 100 (don't remember). Later over/upsampling filters went to a few 100, at which time the summed far-out samples' contributions would fall below the quantisation noise inherent to the DAC. As this noise is physically limited to 20-22 bits equivalent at best you have a clear limit here.
Even so the error incurrent by truncating (actually windowing or, gulp, apodising) a Sinc function early translates in a less steep transition of the filter when viewed in the frequency domain. You can see this clearly yourself when playing with the filter settings of a software tool like iZotope.
You can also glean a lot from the 'transition' and 'impulse' views of this site that compares tens of sample rate convertors:
http://src.infinitewave.ca/
Back in 2008, on sabbatical, I played some time with oversampling and downsampling filter designs, thinking there might be a market for ultra-high-accuracy convertors. I went into the low millions of taps (IIRC), but in the end it was not worth it. At any rate, products like iZotope saw the light of day and that was it. I went back to my normal business, which tends to have a bit more (positive) impact on mankind than trying to fit millions of angels on the top of a tap.