We've all tried to see if we could tell the difference between a lossless file and a lossy one
Personally, I haven't tried that hard! If it "sounds bad" I'll notice but I don't want to listen carefully for a defect/artifact. I'd prefer to be "dumb and happy" enjoying the music. I grew up with vinyl and I hated the clicks & pops. It was especially distracting when it was my record and knew when that bad click was coming... I'd be stressed-out waiting for the click instead of enjoying the music. I don't want to experience the same thing with MP3s. (I mostly listen to MP3s in my car with an older iPod connected to the car stereo.)
A couple of times when listening to a ("high quality") MP3 that I ripped myself I've
thought I heard a compression artifact. But when I went-back and listened carefully to the CD, it had the same "defect". I have occasionally heard poor-quality lossy files but it's not the kind of thing I normally listen to (except with cell phone calls).
and I was wondering (since I personally can't), what are we supposed to listen to/for? What does being a trained listener actually mean?
For those who are able to tell the difference, where to do you put your attention on? High/medium/low frequencies, voices, acoustic music or certain instruments that tend to reveal the imperfections, a potential distortion?
Any difference that you can hear! At high-quality MP3 settings the "last remaining" artifact seems to be
pre-echo so that's what trained listeners listen for.
I'm not sure if pre-echo is only an issue with MP3 or if it's also a characteristic of other lossy formats. Pre-echo isn't audible (or maybe not as-audible) in every recording. There are "killer samples" that "break" the format and reveal it's weaknesses.
(since I personally can't)
You can at lower bitrates.
Of course at lower bitrates (more compression = more loss = lower quality) artifacts will be more apparent, and there will be other distortions, not just pre-echo. Like I said, I don't listen to low-quality MP3s but you can try it yourself. I
assume the way to train yourself is to start with low-bitrate files that will sound obviously bad, and work "up" from there.
For lossy compression artifacts, I don't
think it's that important. There is a loss of high frequencies, but that's not the biggest issue. Listeners sometimes hear high-frequency distortion but they rarely hear/report a
loss of high frequencies. The idea is, even if you can hear very-high frequencies as pure-tones in a hearing test, in the context of music the highest frequencies are masked (drowned-out) by other sounds.
The loss of high frequencies
is the easiest thing to
measure so people often use a spectrogram to "prove" the file is lossy, or from a lossy source.