Spotify uses Ogg Vorbis, except the lowest bitrate on mobile, which is HE-ACCv2, and on Chromecast/similar devices, which get AAC (because they can't natively decode Ogg Vorbis).
It is a significantly better codec than MP3, and doesn't suffer from the pre-echo, washy cymbals and badly-encoded high frequencies. At least not until you severely decrease the bitrate.
The most common problem is when the master is hot and encodes clipped waveforms. It sounds even stranger on lossy. In fact I'm not sure if the codecs' filter banks still have perfect reconstruction in this case or if it's just a borked master. In any case, it always tends to be quite clear when I switch between lossy and lossless on Tidal.
Anyway, it's complicated and brittle, and lossless 44/16 is only something like 2x bitrate and bit-perfect (so you can write proper regression tests for codecs very easily) shrug, I think I'll go that route.
disclaimer: occasionally get paid for mastering music, so I'm wired to prefer simplicity, transparency and don't mind some redundancy in ensuring signal integrity.
What struck me though was that I couldn’t tell which was “better”.
In the track there was a section where the background drumming on one setting sounded like aluminum drum sticks while the other like wooden drumstick. Very subtle effect, and took some intense concentration to pick out.
I didn't know that was the case. Do you know why that somehow accentuates MP3 artifacts? It's not obvious to me why it would, since all the processing iMac speakers might perform is high-quality (no recompression involved). I mean, obviously the iMac isn't going to improve the MP3, it just tries to improve the speakers. But why would that make artifacts more noticeable?
If anything, don't "crappy" speakers hide MP3 artifacts, because they're not good enough to expose them?
Now if the masking sound disappears or changes, the crap is no longer masked.
In other words the psychoacoustic principles work only when we don't alter the signal too much after the codec.
Many diminutive speakers make use of techniques like multiband compression (dynamics, not entropy) to produce "larger" sound. That wreaks havoc on the psychoacoustic model of lossy codecs.
However, obvious encoding artifacts abound on Spotify. Do I have a superhuman hearing?
Probably not. My hypothesis is that not everyone authors lossy files as meticulously as Fraunhofver. Also, the performance of mp3 depends on a highly linear and faithful reproduction after decoding. Mp3 is painfully obvious on crappy, processed-to-hell speaker systems like the iMac.
I think the real question is, why bother with lossy codecs? FLAC streams are lightweight by today's standards, and it's just so much simpler.