The answer is that they don't discuss the audio quality in bits of resolution, but are specifying dynamic range and noise. So people interested can figure out themselves how many bits that are "good".

So you might get specifications like 106dB SN, which tells us that you don't get 24 bits of high-quality data. But you get better sound than the 16-bit audio cards. Especially since you might have 192kHz sampling rate giving room for filtering if you just want cleaner 16-bit 44kHz CD-quality audio. And having the extra bits means you can perform mixing before downsampling to CD-quality audio.

And, as already mentioned, they don't need good voltage references. Just very noise-free. And they don't require that the voltage reference is long-time stable. As long as the stability of the voltage reference has a time constasnt in minutes and not in milliseconds.