Links:
Main: http://mp3.wp.pl/p/test/
Results: http://mp3.wp.pl/p/test/wyniki/
FAQ: http://mp3.wp.pl/p/test/faq/

For 60-100 kbps Ogg Vorbis wins (second: MP3 FastEnc 96 kbps lowpass 14 kHz, third: AAC Liquifier v.5.0 96 kbps, 16069 Hz)

For 101-170 kbps Ogg Vorbis wins (second: MP3 FastEnc VBR 50, third: MP3 Blade, 160 kbps)

For 171-210 kbps AAC (Psytel) wins (second: RA8 176 kbps, third OGG 1.0 Final VBR 6)

Note:
HE-AAC, QT AAC, Ogg Vorbis GT3B1 were not tested.

Note2: "Średnia ocena" - closer to 1 it means codec is better, closer to -1 means codec is worse. >0 means that codec received more positive results than negative, <0 - more negative than positive,

Note3. This news is quite old.

I'm a bit perplex.
PsyTEL 160 is really worse than Blade according to this test, and PsyTEL 128 is better than the same encoder at... 160 kbps ?!

It's funny to see that mpc --standard is just slightly better than Blade 192 kbps, which is really much better than --r3mix.

You see power of blind test....

Blind test for deaf people : that's comical

Were ABX tests asked for each notation ?

I wonder how this blind test took place...maybe you can tell me since it is written in some sort of weirdo language .
However, how come that wma 64 sounds much better than wma 96??? This is incredible!

You can ask authors of test: jastrzab@hub.wp-sa.pl and imhotep@hub.wp-sa.pl

PPL have two samples - A and B (each time they go to website, they have another set of samples). They were asked, which sample is better: Certainly A, Certainly B, Rather A, Rather B, I don`t see any difference. For A: Certainly A = +2 points, Rather A = +1 point, equal = 0 points, rather B = -1 point, certainly B = -2 points.

So there was no listening of the originals?

Edit: that doesn't seem like a very orthodox test...

FastEnc (for MP3s) has got the best results, and I don't wonder why. I think it mutes down critical sounds from the side channels a little bit, i.e. you don't have good channel dynamics but less artefacts.

For Blind Test participants it should be essential to give info about their equipment. The best way to detect side channel loss is via Headphones IMHO. Here you'll notice that FastEnc is worse than thought.

I agree.

IMO, an orthodox test means this:
http://www.itu.int/rec/recommendation.asp?...t=R-REC-BS.1116

The facts that

1. No reference file was included in the test.

2. No statistical analysis has been performed to check if the results are reliable

make this test worthless.

Too bad.

I agree completely. May as well be taking a poll on MTV as to which song people like better. You also have no control over the equipment. More importantly, fidelity with the original isn't measured in any way.

It's an interesting test for marketers deciding what samples of their music should be encoded with to appeal to the largest audience. A completely worthless test for deciding what's best to use to preserve the original sound.

Could someone tell what the numbers in the columns are ?

Translation of "Ilość Średnia ocena Odchylenie standardowe "

Yes, I found an an online translator
The columns are "amount", "average", and "standard deviation".

Example with the WMA 64 and WMA 96 :
WMA 64 is rated 0.216 plus or minus 1.396
WMA 96 is rated 0.152 plus or minus 1.309

Therefore WMA 64 is "probably" (the probability being defined by the standard deviation calculus, maybe 98 % ) somewhere between -1.180 and 1.612, while WMA 96 is probably between -1.157 and 1.461.
Since the intervals overlap WMA 64 is not rated higher than WMA 96.

The highest possible lower bound is -0.844 for OGG 1.0 Final VBR 0 (0.537-1.381), and the lowest possible higher bound is 0.710 for WAV lowpassed at 7.5 kHz.
It means that every result overlap with all other ones, therefore all samples were rated equal

Would be interesting when they included the original wav file in the test. It could very well get a worse score compared to the lossy compressed files

Pio: What translator you used? The only one know of (Babelfish) doesn't support Polish.

http://www.poltran.com/

Every time I read a new post in this thread, the title looks more and more like a joke about how polish people do blind testing...

Btw: I have nothing against polish people....
It just reminds me of freindly jokes between norwegians and swedes

Anyway, how was this test done? 'Post random results' method?

I don't think that most people can percieve the difference between original
and any ~192kbps codec (maybe except poor MP3) on normal sample and I think they haven't used any problem samples.

Hint: Go post Polish jokes at FB2k forum.
Peter will appreciate them.

I wondered about it too, but the average value should have been 0 and the standard deviation 1.960.
However, given that the number of people was not infinite, maybe random answers can actually account for the results.

Oh, my, that one was classy. :B

I posted link to this thread (three times) in comments section on mp3.wp.pl - they deleted/prevented from showing my comment...

I had to clean this thread a bit. Insults never or very seldom lead to improvements... Too bad some people make HA look an elitist arrogant board, although the purpose of this board is to spread information to everybody in non-hostile way.
Although there's something good also, like Pio2001's posts, as usual.