Guybrush

Alright, it seems people are not interested in doing this challenge, so I'm gonna take it down. I don't think I have enough results to bother with any kind of sophisticated statistics, but I'll post them here in case someone's curious.

The test was listening to a four different versions of a 30 seconds mp3 encoded in different bitrates. You were asked to arrange sample 1, 2, 3 and 4 in the order lowest bitrate to highest.

The correct sequence was : 2, 1, 4, 3

The bitrates (kbps) were : 128, 160, 256, 320. Those who took the test were assigned numbers. You can see how they did here :

1. 2, 3, 1, 4 = 1 Rights
2. 1, 3, 2, 4 = 0 Rights
3. 4, 1, 2, 3 = 2 Rights
4. 4, 3, 1, 2 = 0 Rights
5. 2, 3, 1, 4 = 1 Rights
6. 2, 4, 3, 1 = 1 Rights
7. 1, 2, 4, 3 = 2 Rights
8. 2, 1, 3, 4 = 2 Rights

(Note that there are many possible ways of counting rights in this dataset. This just counts how many of the samples are in the right place in the sequence.)


The average guessed sequence (counting which BR is found most often in each place) would be : 128, 320, 160, 256

Although there's not much data to base evidence on, from what I can see, people don't seem very good at separating bitrates higher than 160 and not really reliably down to 128 either.

It's the result I would expect to find even with more data, so I'm not surprised. Maybe it'll make some people reconsider the worth of very high BR/lossless files even if there's nowhere as much data as I'd like.

__________________
Something Completely Different