Here are my favourite samples I use to test Vorbis at q4 and q5. They represent Vorbis's long known problem with sharp noise and feature some nifty pre-echo on top of that.
The samples were taken from the following tracks:
Autechre — [Gantz Graf EP #01] Gantz Graf [3:58];
Abfahrt Hinwil — [VA — Everything Is Green #02] Everything Is Green [4:25];
Nine Inch Nails — [Quake OST #01] Quake Theme [5:08].
foo_abx v1.2 report
foobar2000 v0.8.3
2006/10/26 01:01:52
File A: file://C:\Files\test samples\ae_gg.wav
File B: file://C:\Files\test samples\ae_gg.ogg
01:01:52 : Test started.
01:02:18 : 01/01 50.0%
01:03:00 : 02/02 25.0%
01:03:22 : 03/03 12.5%
01:05:29 : 04/04 6.3%
01:06:26 : 05/05 3.1%
01:08:01 : 06/06 1.6%
01:09:23 : 07/07 0.8%
01:09:56 : 08/08 0.4%
01:10:10 : Test finished.
----------
Total: 8/8 (0.4%)
That was rather easy. Vorbis just smeared the granular synthesis.
foo_abx v1.2 report
foobar2000 v0.8.3
2006/10/26 14:42:34
File A: file://C:\Files\test samples\eig.wav
File B: file://C:\Files\test samples\eig.ogg
14:42:34 : Test started.
14:43:13 : 01/01 50.0%
14:43:24 : 02/02 25.0%
14:43:45 : 03/03 12.5%
14:43:55 : 04/04 6.3%
14:44:06 : 05/05 3.1%
14:44:18 : 06/06 1.6%
14:44:31 : 07/07 0.8%
14:44:49 : 08/08 0.4%
14:44:51 : Test finished.
----------
Total: 8/8 (0.4%)
Easy, again. On the other hand, LAME -b 320 fails on this sample completely, even worse than aoTuV at q5 despite twice the bit rate.
BTW, this is my favourite pre-echo sample, the attacks are sharp but not very disturbing, and the background is very clean. The track itself is very nice as well.
foo_abx v1.2 report
foobar2000 v0.8.3
2006/10/26 14:46:39
File A: file://C:\Files\test samples\nin_quake.wav
File B: file://C:\Files\test samples\nin_quake.ogg
14:46:39 : Test started.
14:47:14 : 01/01 50.0%
14:47:30 : 02/02 25.0%
14:47:45 : 02/03 50.0%
14:48:20 : 03/04 31.3%
14:50:09 : 04/05 18.8%
14:53:24 : 04/06 34.4%
14:53:27 : Trial reset.
14:54:10 : 01/01 50.0%
14:54:33 : 02/02 25.0%
14:54:57 : 03/03 12.5%
14:55:22 : 04/04 6.3%
14:57:37 : 05/05 3.1%
14:58:23 : 06/06 1.6%
14:59:18 : 07/07 0.8%
14:59:48 : 08/08 0.4%
14:59:49 : Test finished.
----------
Total: 12/14 (0.6%)
That was much harder. The difference is very subtle, almost nonexistant background noise. At my first attempt, I was confident at the beginning, but shortly after started to fail and decided to take a short break. After that, I once again understood the difference to concentrate on, and passed the test without any further mistakes.
Enjoy.
EDIT: Oh, completely forgot to post my testing "equipment".
Soundcard: Creative SB Live! 5.1.
Headphones: Sharp HP-MD33-S.
It was hard for me. Difference is negligible.
BS; Lancer(SSE) [20061005] (based on aoTuV r1 [20051117])
-q 3
foo_abx v1.2 report
foobar2000 v0.8.3
2006/10/26 21:21:41
File A: file://C:\tmp\eig.ogg
File B: file://C:\tmp\eig.wv
21:21:42 : Test started.
21:22:29 : 01/01 50.0%
21:22:44 : 02/02 25.0%
21:22:49 : 02/03 50.0%
21:23:17 : 03/04 31.3%
21:23:26 : 03/05 50.0%
21:23:37 : 04/06 34.4%
21:23:50 : 05/07 22.7%
21:24:01 : 06/08 14.5%
21:24:08 : 07/09 9.0%
21:24:38 : 08/10 5.5%
21:25:14 : 08/11 11.3%
21:25:36 : 09/12 7.3%
21:25:41 : Test finished.
----------
Total: 9/12 (7.3%)
ae_gg.mp3: Can't ABX lame -V 0
Audigy1 + rarity Electronika_TDS-5M + Bad_ears.sys ;-)
P.S. "eig.wv" - i like it!
Thanks a lot for your great eig sample.
I am pretty deaf as towards pre-echo, but with this sample even me can hear the problem.
Good lord, LAME handles this one poorly. Even at 320kbps, it's still audible! It's much worse than castanets.
... LAME handles this one poorly. ...
I'm pretty sure other mp3 encoders will have similar problems, and that's what makes it interesting for the upcoming listening tst.