The people claiming this show the response of a filterless or poorly (slow) filtered step/needle/squarewave response showing a steeper than 1/2 fs (sample frequency) risetime than can possibly be in a recording. They 'link' that to what they hear.
Many of my DAC's allow a filter less option together with different filters.
Essentially the use of a reconstruction filter is the direct consequence of the difference between the theoretical ideal of "Nyquist" and the real world. An ideal filter would have an infinite number of taps.
Now here a fun fact, a digital filter and a digital reverb unit operate on the same hardware principles (which can of course also be expressed in DSP).
So let's see what a "1Million Tap" filter does? The filter shape will be near ideal, if a symmetrical FIR filter is used.
In addition a delay/reverb chain is generated that is 1 million samples long, 500,000 samples will create a pre echo, the others a post echo. And this is 11.3 seconds of pre and post echo each. And no, we do not need an impulse to get this echo and reverb, it applies to everything. The transient or square wave response is simply an illustration of the time domain behaviour.
Now, if we add seconds worth of reverb, will this be potentially audible? And if it is audible, how do people normally describe the addition of reverb to music? What does it do to location accuracy of phantom images in the stereo soundstage?
What is the most distinct quality reviews of the 1million tap filter DAC mention, compared to others?
And does such an additional reverb qualify as fidelity impairment or "distortion"?
A higher sampling rate would resolve this, if we sample at (say) 176.4kHz and use a cascaded set of analogue filters to achieve (say) -90dB @ 88.2 kHz the whole problems disappears in a puff of logic.
But there are a few, shall we call them effects from dispensing with the digital filter that are interesting. Naturally, extra reverb is cancelled. But what else happens?
1) Ultrasonic images are produced.
Due to a number of factors these images tend to be dominated by the signals high frequency content. Combined with a shallow slope analogue filter (which may also compensate for the HF droop) a spectrum of music tends to show a similar one to what would be present had the music's ultrasonic content be recorded.
Is ultrasonic content in music audible or percievable? AFAIK nobody has invalidated any of the experimental results of Oohashi et Al.
2) The ultrasonic images act in effect as signal level dependent "dither" analogous to noise like dither used by pro audio manufacturers, e.g. Apogee UV22.
Anyway, I suggest everyone listens for themselves and selects what they find preferable.
Thor