(Since Musepack doesn't appear to have much going on I've been re-doing parts of my stash into Ogg, since my flash player supports it and I won't have to store duplicates in both MPC and Ogg/mp3 files for compatibility.)
While listening to an encode of New World Man by Rush I heard something that wasn't right in the opening... I can't really explain it very well, but it's some kind of distortion or smeared, hollow-sounding pre-effect (like brushing across corduroy). I can ABX it distinctly up to -q10! (16/16).
I took the first 6 seconds or so of the intro and used that, although the artifact is present in the first 500ms, which is where I tested.
Samples here:
http://www.hydrogenaudio.org/forums/index....showtopic=44861 (http://www.hydrogenaudio.org/forums/index.php?showtopic=44861)
foo_abx v1.2 report
foobar2000 v0.8.3
2006/05/21 23:11:42
File A: file://C:\Documents and Settings\Primary Login\Desktop\NewWorldMan[orig].wav
File B: file://C:\Documents and Settings\Primary Login\Desktop\NewWorldMan[OggEnc283-q10].ogg
23:11:42 : Test started.
23:12:19 : 01/01 50.0%
23:13:01 : 02/02 25.0%
23:13:16 : 03/03 12.5%
23:13:25 : 04/04 6.3%
23:13:32 : 05/05 3.1%
23:13:54 : 06/06 1.6%
23:14:07 : 07/07 0.8%
23:14:13 : 08/08 0.4%
23:14:25 : 09/09 0.2%
23:14:35 : 10/10 0.1%
23:14:44 : 11/11 0.0%
23:14:54 : 12/12 0.0%
23:15:06 : 13/13 0.0%
23:15:18 : 14/14 0.0%
23:15:29 : 15/15 0.0%
23:15:40 : 16/16 0.0%
23:15:42 : Test finished.
----------
Total: 16/16 (0.0%)
I first noticed it using aoTuVb4 a while back at -q3 and didn't bother to think much of it until I tried aoTuVb4b/LancerSSE2 -q5.5 and noticed it wasn't a lot better. (since -q5 is considered to be transparent I thought I'd better bring this up for discussion) I used OggEnc v2.83 "oggenc2.83-1.1.2-MOD-generic.zip" to generate the -q10 file since I wanted to make sure it wasn't some ICL compiler/SSE2 issue.
Your results are interesting. Could you please post your findings at the official vorbis dev list (http://lists.xiph.org/mailman/listinfo/vorbis-dev)?