Hi all,
Thanks to sssSutherland for prompting me to post. I don't get down to the Water Cooler as often as I'd like.
Anyway, the blind tests I did with a number of other editors some years ago tested about ten different "lossy" algorithms, including MP3, running at various data rates.
Here are some of the conlusions I reached (and many of the other critical listeners):
At 320 kbps, the best of the compression algorithms are amazingly transparent. It took repeated A/B/X comparisons to isolate artifacts. DJ_Stunna is correct in stating that cymbals and, we found, other percussion instruments are really revealing of flaws. In particular, castinets were very revealing of artifacts. So, no flamenco at 320 kbps!
Another tip-off was increased sibilance on female vocals. We used a Suzanne Vega a cappella solo vocal (really annoying after many repetitions!) that became quite coarse and sibilant at slower data rates. If memory serves, it was also audible at 320 kbps.
One of the algorithms auditioned, developed for the Musicam digital radio system in Europe, was really excellent. I think it came out of that Fraunhoffer Institute in Germany. It was the one to be adopted for digital radio service in Canada. Sadly, the Canadians picked a part of the spectrum for the digital service that the Pentagon used for military purposes, so that aspect of digital radio in Canada never got beyond experimental broadcasts (which I heard; it was excellent).
Rock music wasn't as generally revealing of digital algorithm artifacts as classical, but then again in classical stuff you get a big variety of percussion instruments, the timbre of which may be slightly altered by a poor or too slow algorithm. Put another way, you don't get many castinets in rock and pop music!
Regards,
