ABX Result Calculation : Suggestion Options

[kotrtim]

Is this way of calculation any good?

this is my suggestion: (to test codec's transparency)

at high bitrate such as 192kbps and above, its pretty hard to rate it!
so why not just use A , B & X test (which is available in foobar2000)

Formula = no. of wrong trials divided by no. right trials

Fix the trial to 10

if the result is 9/10, that means 1/9=0.11111, the codec is NOT transparent

if the result is 10/10, 0/10=0, the codec transparent probability is 0

if the result is 5/10, 5/5=1, the codec is transparent

if the result is 1/10, 9/1=9, the codec is NO DOUBT transparent

If the result is 0/10 (which is impossible), 10/0=......No DOUBT transparent

..............

This post has been edited by kotrtim: Nov 24 2004, 10:53

[guruboolez]

(...)
if the result is 5/10, 5/5=1, the codec is transparent
(...)

Did you ever run any ABX program?

http://ff123.net/abx/abx.html
An encoded file is for sure transparent if pval < 0.05 (even below if ABX test is not sequential).
5/10 => pval = 0.624
In other word, 62% of chance that the tester is guessing.

[kotrtim]

An encoded file is for sure transparent if pval < 0.05 (even below if ABX test is not sequential).
5/10 => pval = 0.624
In other word, 62% of chance that the tester is guessing.

oops...thanks guru, I've used ABX in foobar2000, but I don't really understand the pval=XX%

This post has been edited by kotrtim: Nov 25 2004, 11:05

[guruboolez]

Another precision:
0/10 doesn't mean transparency, but total failure. 9/10 is a pretty good result - not the opposite