I’m going to stand up for @AE67 here–not because I agree with his point but because I admire his courage in admitting this. He’s choosing–knowingly I’m sure–to step into a trap, a natural bias toward positive opinions in this field: If you admit you can’t hear something, you’re open to charges of having a bad ear or an inferior system with inadequate resolving power. (Them’s fightin’ words.) But if you claim you can hear something, there’s no risk of being called out for it. In this subjective world, where rigorous testing is disrespected, no one will ever ask you to prove it (or if they do, that person won’t be taken seriously by most).
As to the question of hearing differences in lossy-compressed files, the differences are clearly audible. I don’t have direct evidence to cite (although I believe it exists) but I’ll point out a paper many seem to have missed on a related issue: hearing the difference between high-rez and CD-rez files. It’s from none other than Bob Stuart. Here’s a link and a synopsis, by me:
AES - The Audibility of Typical Digital Audio Filters in a High-Fidelity Playback System
Starting with a 24/196 recording from 2L (a Haydn string quartet), Bob Stuart and colleagues tested six “conditions” chosen “to offer a reasonable match to the downsampling filters used in good-quality A/D converters or in the mastering process; we wanted to minimise the ripple depth and maximise the stop band attenuation in order to reduce audible ringing artifacts, as described by Lagadec
[31]” :
Condition / Filter cutoff (Hz) / Further processing
1 / 21591–22050 / None
2 / 21591–22050 / 16-bit quantization
3 / 21591–22050 / 16-bit quantization, rectangular dither
4 / 23500–24000 / None
5 / 23500–24000 / 16-bit quantization
6 / 23500–24000 / 16-bit quantization, rectangular dither
All “conditions” were found to be audible except condition 4.
They write, “These results indicate that it is possible, with considerable effort to ensure a transparent replay system, to discriminate the difference between a selection of high-quality 192 kHz 24-bit music signals and the same audio with standard production processing applied. All forms of processing tested here were audible, except for one condition where performance was significantly different from chance at the 6.7 % level, including emulated downsampling filters at standard sample rates and 16-bit quantization with or without RPDF dither. Differences were demonstrated here in a double-blind test using non-expert listeners who received minimal training.”
My take: The question of whether (eg) 16-bit files sound different from 24-bit files can’t be answered absolutely. This paper shows that in every case where 24 bits were downsampled to 16, the difference was audible in a rigorous trial. Yet, we can’t rule out the possibility that some new method of quantization could be devise that would be transparent at least to the extent that this jury could not hear the difference. But then perhaps a better trained jury, with better ears, could hear it. And round and round we go.
Jim