[~ 96kbps / für Ausprobier- oder Testuser sowie Charthithörer]
- MP3
(Lame 3.92 MMX / -b 32 -m j -h -p --abr 96 -B 320 --resample 44.1 -q 0 -d)
- Ogg Vorbis
(q-2)
Preis: 0,29 Euro

• If you only perform 4 trials, the result is quite unsecure. Better keep on until p> 1% (e.g. 7/7)
• All trials count! E.g. after getting 2/4 it's not allowed to start again from 0/0. This would spoil the p-value calculation and make the results meaningless (just in case  ).
• Most likely it'll be easier if you listen to a short passage of the sample (2-3 seconds) often, trying to focus on certain details (e.g. transients like percussion, background noise, high-pitched sounds like cymbals)

• for lame try the --alt-preset commandlines (see Digga's link). Most likely the results will be better at comparable bitrates then with the commandlines you used.
• for ogg vorbis at high bitrates you might want to try gt3b1 encoder available at rarewares.

• Many cheap soundcards resample 44.1kHz audio to 48kHz in a crappy way on playback. Try foobar2000's diskwriter for decoding to .wav using its high quality resampler to 48kHz - or directly use foobar2000's ABX tool with resampler enabled.
• Lossy Encoding -> Decoding causes clipping on loud music. To avoid this clipping affecting your ABX test, decrease volume. The easiest/most elegant way to do this, is to use foobar2000 for decoding with replaygain enabled or - again  - foobar2000's ABX tool for performing ABX tests.

according to your suggestions i did another testing (identical source like the above).

• I can't see if replaygain was used for decoding to avoid clipping (or if you have checked manually (e.g. using a wave editor) if the decoded files clip.
  
  
• I suggested foobar2000 for decoding because, besides applying replaygain, it is able to decode mp3 encoded with lame without offset. With corrected offsets (can be done manually as well) ABX tests are more reliable. This only affects the mp3s IIRC.
  
  
• It would be good to provide a description of the differences you heard (not only which position in the sample, but also what it sounded like), so it's easier for others to repeat the test

If you're asking yourself why the hell i did use such big lame syntaxes and didn't just use --alt-preset standard or perhaps --alt-preset extreme i'm gonna tell you why: the whole test-idea came to me as i wrote that hifi-forum thread where i was talking about selling music on the internet (like popfile.de) and how quality may be to which prices. but that isn't worth discussing because musiclabels won't sell mp3s or other non-drm-fileformats and if they'll do they won't do it with great quality, or am i wrong?

btw: wouldn't have my soundcard if it resampled the output to 48kHz (which i'm quite sure it doesn't) this also done on the original wav? wouldn't that mean that the particular error(s) must have been on all replays so it's meaningless?

... what about you?