To illustrate the discrepancy between measurements and hearing, the following paper by Cees sheds some light.
In 2008 I decided to start an investigation to map the sound characteristics of different DACs. The performance of these converter chips was largely unconvincing though I could not yet identify what the underlying causes were. My dissatisfaction was fed by frequent visits to concerts and the seeming impossibility to bridge the gap between live performances and digital recordings. For already many years I'd been interested in the technical side of recording music and had from time to time made some recordings having in my possession a decent set of microphones, a mixer and a studio-tape recorder. Despite barely acceptable S/N ratio and limited dynamics (≤ 60dB) which my equipment could handle back then, the experience felt 'real' and involved me in the music. That is why I understand vinyl lovers despite putting up with pops, noise and scratches inherent in this medium. I wanted to recreate this sense of realism and feeling with digital equipment and started my investigation in the hope of recreating it. Because of my work and experience in the field of electrostatic loudspeakers, I have many contacts there who gave me opportunity to listen to a so-called NOS non-oversampling homemade DAC. Despite its easily apparent shortcomings, I heard something which had been missing all along in other DACs. The emotion, musical involvement and experience was suddenly back. I wondered whether this was the road I should take? From that moment on, I shifted my investigation to studying the defects of contemporary digital equipment generally based on the widely used method of oversampling. As a designer of electronics, I owned advanced measuring equipment. With my roughly 35 years of experience, I should be able to pursue this. Even so, it kept surprising me that as a direct consequence of oversampling, many artifacts are shown in a system which really should not be within the audible range.

Ryohei Kusunoki. Because I was cautiously optimistic about the sound of NOS DACs, I actively started searching for knowledge on the subject. At various fora it soon became apparent that I wasn't alone searching for the 'real' music experience. I also came across an article by Ryohei Kusunoki who explained the often used oversampling methods in a comprehensive manner. The article shows that, when one's vision is based on experiencing live music, it seems odd to rely on oversampling. This is why I set aside all of my knowledge concerning digital registration and imaging and decided to follow my heart.

Both hobbyists and professionals who are convinced of the NOS principle have had to make do with old DAC chips once developed by Philips which hit the market in the 1980s. For their time they were truly amazing. Companies like 47Labs, Zanden, Audio Note and AMR are convinced of the validity of the NOS principle too. Due to the lack of more modern components, they're forced to use old chips like the TDA1543/1541. And they are right in doing so. The more modern chips to market have grown increasingly complicated and are burdened in most cases with FIR filters which, though they make oversampling techniques possible, also make it impossible not to use them. Because I had, as an electronics designer, often developed products for industrial purposes, I had gained much knowledge of industrial components. Especially in the fields of process and medical engineering, DAC chips are used without FIR filters. Would it be possible to utilize these chips for audio products? Which characteristics should I focus on to improve on the 'old' TDA1543 or TDA1541?

It was clear to me that modern DAC chips offered several advantages over the vintage specimens. Since the 1980s, great headway had been made regarding switching noise and the speed and conduction of Mosfet switches, which are used in so-called R2R ladder circuits. Finally the linearity of earlier mentioned circuits was improved. Outside of these improvements, little had changed. However, it seems that these characteristics are important for, amongst others, the ease and naturalness with which the music is portrayed to improve the desired 'real' feel. Speed and bandwidth are features also of importance in amplifiers. This is why op-amps seem to cause many problems in audio circuits. Terms such as slew rate and open loop gain apparently weigh far heavier here than harmonic distortion does. It makes sense that electronics which can portray music well are generally derived from radio-frequency engineering, where the bandwidth can be as high as 500MHz. When using components that allow such bandwidth, completely different design techniques are required to build a product that is both good and stable. Examples of these are the designs made by Nelson Pass.

Despite the general availability of many measuring instruments in the past few years, it has only become apparent that measurements alone will rarely show if a product will be pleasant to listen to. Are our methods insufficient? Absolutely not. In the design phase, many problems are solved using relevant tests. Yet when for example you test for the spaciousness of the soundstage, current tests fall short. Now our own ears are used as additional instruments. Bandwidth and speed are important factors in electronic audio circuits. This stems from the fact that signals which run through  'old-fashioned' semi-conductor circuits can become distorted when tainted by high-frequent components. The technical term for this is a slewing-induced distortion. The circuit in such a case can't adequately follow the input signal. Our own hearing appears to be very sensitive to this. It often leads to a soundstage which is experienced as flat. [Limited slewing rate: the green line (output signal) shows the inability of the amplifier to follow the input signal (red).]
slewing rate

The current output of a DAC chip is one where a variety of audio plus HF signals are prevalent. The switches which control the R2R ladder circuit in the chip create ultrasonic switching noise. No matter how low, this noise is embedded in the audio signal as an extra component. It means that standard parts like op-amps no longer function correctly. Despite that, many manufacturers continue to use op-amps as current/voltage converters even though they are unsuited for the job. It is important that the electronics which convert the current from the DAC chip to voltage be swift and of sufficient bandwidth. From 2008 to mid-2010, I sought a DAC chip which had the characteristics described. Eventually I found the correct chips and used them in the Quad, Octave and Hex DACs. That this choice was correct was made clear from the number of positive reviews we received worldwide. [The open-loop-bandwidth of the NE5534 audio op-amp, without feedback, stops already at 1kHz.].
It remains remarkable that even though a sine wave with a sample rate of 44.1kHz looks relatively choppy, our hearing does not experience it as such [1kHz sine wave sampled at 44.1kHz]. Some things will be enhanced because, assuming CD quality, in one second merely 44'100 analogue values can be presented which does not add accuracy. When looking at an oversampling DAC, a FIR filter will calculate and fill in the intermediary values. A FIR filter should theoretically give better results. In practice however, it turns out that there are no audible differences.
fir filter
[The lower wave form is the NOS signal, the upper reflects the same signal after passing through a FIR filter.] This makes for a fun experiment. Through an amplifier and loudspeakers, let a NOS DAC generate a sine wave. Record it with a microphone and show the result with the help of an oscilloscope. The choppiness originally present in the signal has completely disappeared. The reason for this is that the signal has already passed several filters before it ever reached the microphone. In the case of a Metrum Acoustics DAC, there will be a mild 70kHz 1st-order filter just before the signal is sent to its output terminals.