- A vanilla
lame -V2 inserts 0.025s of silence at the beginning of its MP3s. - Encoding adds a lot of peak samples to a track. For instance, there are none in Alone Tonight ripped from CD to FLAC, but half the track is covered in red (at the “view entire project” zoom level) for pretty much any lossily encoded version.
- Assuming I'm reading all of the meters correctly, the noise is held to -12dB in the 192k MP3, which matches the actual dynamic range of the envelope. So it’s no wonder that it's widely considered to be transparent.
I am less surprised that the differences between the rip and the encoding are concentrated in the higher frequencies, where there is the most to gain by discarding data. Particularly since LAME runs a lowpass filter on the input, so anything above the cutoff band will appear as a difference.
They say that familiarity with artifacts is a major predictor of how well one will do on an ABX test. In that case, I think I'd rather stay unfamiliar and enjoy my 128k MP3s. (I tested myself with abx-comparator and got 3 of 10 trials correct…)
And yes, I am bad at this hiatus thing.