Skip to main content
Topic: ABX vs. ABXY (Read 10044 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Re: ABX vs. ABXY

Reply #25
Why would one play ABXA ...
And how a listener knows  beforehand how sound A and B, huh?

Ever hear of listener training?


There is  important moment. Kamedo2 and me claim about usefulness of ABXY for subtle difference.
You can't "memorize" enough subtle difference.

Training doesn't apply here.


Re: ABX vs. ABXY

Reply #26
I really cannot understand why finding a difference between A vs B is harder than finding a difference between X vs Y.
It isn't, of course. But finding this difference is only the first step.
Exactly. Evaluating A and B is something you do only once in a single ABX session. After that, evaluating X is something you do 5, 8, or 20 times.

Re: ABX vs. ABXY

Reply #27
Step by step:

=ABX=
Listen to A
Listen to B
if difference is subtle and wasn't caught at first listen, listen to A and listen to B again until determining a difference.
Determining a difference does not mean remembering A and/or remembering B. It means one sample does not sound like the other in a concrete part of the audio. And since you found the difference while listening to A and B, you know which of them has the "different part".
Listen to X on that specific part where you located the difference previously, in order to see if it plays as the "bad" one or as the "good" one.
If X happens to be the same than your last listening ( let's say, you last listened to B, and X is B), you will not hear the difference, and might question if, maybe listening to A will show the difference (and determine that X is B).

=ABXY= (as I understand you describe it)
Listen to A?
Listen to B?
Listen to X
Listen to Y
if difference is subtle and wasn't caught at first listen, listen to X and listen to Y again until determining a difference.
Determining a difference does not mean remembering A and/or B, and/or X and/or Y. It means one sample does not sound like the other in a concrete part of the audio. And since you found the difference while listening to X and Y, you know which of them has the "different part".
Now, what? 
Listen to A?  Listen to B?  Don't listen since now you somehow remember what is A and what is B?

Or maybe you simply say that rather than listening to A/B A/B A/B and then detemining X, you continually listen to ABXY ABXY ABXY until you are sure B is X or B is not X?

Re: ABX vs. ABXY

Reply #28
Why would one play ABXA ...
And how a listener knows  beforehand how sound A and B, huh?

Ever hear of listener training?

There is  important moment. Kamedo2 and me claim about usefulness of ABXY for subtle difference.
You can't "memorize" enough subtle difference.

Please provide a relevant reference in a peer reviewed paper or other independent authoritative evidence beyond your personal say-so confirming that exceptional claim.

Quote
Training doesn't apply here.

Again, this looks like an attempt to fabricate much-needed facts on the spot.

And the word subtle itself is very vague. Got a formal, generally agreed upon definition of it?

Re: ABX vs. ABXY

Reply #29
=ABXY= (as I understand you describe it)
I often did this when I participated in listening tests:
1) listen to samples  A and B (A B A B A B...) to find encoding artifact (pre-echo, ringing, etc)
2) listen to samples X and Y (X Y X Y...) to find what sample has this artifact (for example, what sample has smearing of sharp transients)
3) optionally - confirm it using this sample and samples A and B

Re: ABX vs. ABXY

Reply #30
If you listen to XYXYXY then you not only need to tell that they are different (they are), but which is which.

If you Listen to AXAXAX or BXBXBX then all you need to know is if you can hear a difference or not.

Re: ABX vs. ABXY

Reply #31
If you listen to XYXYXY then you not only need to tell that they are different (they are), but which is which.
Or rather how they are different.  If you can tell how they are different, then the difference is greater than JND.

Quote
If you Listen to AXAXAX or BXBXBX then all you need to know is if you can hear a difference or not.
Yes it'd much easier to answer a same or different question.
Kevin Graf :: aka Speedskater

Re: ABX vs. ABXY

Reply #32
I will explain why ABXY is superior to ABX on high rate listening tests.

In the high rate test, only relative quality change between two adjacent play is usable as a useful hint.
Sadly, observed (perceived) quality is subjected to a Gaussian noise.

(Finding artifacts typically doesn't work in a high rate test, because it starts to seem like both tracks have the artifacts. Still I can still say which one is dirtier or cleaner, with limited reliability.)

In the spreadsheet B3:C3, the quality of track A and B is set. A is original, so it will be 5.0. B is set to 4.0 as a convenience.

In an ideal world where humans are noiseless machines, exactly 1.0 quality drop will be felt by human when playing A -> B. Never more than 1.0 nor less than 1.0.
Those ideal world result is represented on the spreadsheet D6:G7.

In the real world, humans are moody animals. The same quality change might be perceived as -1.2, -0.2, or even -2.8 sometimes(though this far is not so frequent).
Let's say human perceived it as -1.6. Look at D18:G18. The graph says it is modestly likely(7.5%) that the perception had come from the A->B. Less likely scenario is that you played the identical tracks twice and the perceived -1.6 drop was actually a noise(1.4%). Even less likely is that it was actually an improvement(0.1%) - probably there was a noise outside the listening room.
Apply the Bayes' rule here. In the A->X, It can be A->A or A->B but not the B->B nor B->A so we can exclude this possibility. The likeliness of it being A->B is D18*0.5/(D18*0.5+E18*0.5) = D18/(D18+E18) = 7.5%/(7.5%+1.4%) = 0.85 = 85% after the observation. Very simple. 0.5 is because blind test softwares are known in advance to set the likeliness of X=A being 50%. It says, according to the Bayesian inference, if a quality drop -1.6 is observed, it is 85% likely that it is A->B.
Likewise, In the X->Y test, we can exclude the A->A and B->B possibility. Just look at the A->B and B->A possibility. Similarly, Bayes' rule says 7.5%/(7.5%+0.1%) = 99% likely that it was A->B, and 1% likely that it was B->A (H18:I18).

Both played tracks twice so far, but X->Y is far more confident in saying which is which.
There are minor exceptions, such as H26:K26 where the ABXY is less confident, but according to this spread simulation, in an average situation, after playing tracks twice, ABXY have 89.2% likeliness to be correct, while ABX have 73.3% likeliness to be correct(L:S).





Another spreadsheet simulation says ABX doesn't catch up by playing tracks more, if ABXY is also allowed to play more.


I actually don't take the spreadsheet software to my listening room. Instead I use the analog 'sense' of something similar to H11:I41 and let it accumulate on my mind as I hear more tracks.

Re: ABX vs. ABXY

Reply #33
From a programming perspective, it's a lot easier to just add the "Y Button" and cover both test scenarios. You don't have to ever click it, and you would be doing an ABX test, or use it and you would be doing ABXY.

And if you think that one test is easier than the other, then good! There should be no difficulty inherent to the test itself.

One program to rule them all !

EDIT: Personally I think I do AXY more than anything.

Re: ABX vs. ABXY

Reply #34
I will explain why ABXY is superior to ABX on high rate listening tests.

In the high rate test, only relative quality change between two adjacent play is usable as a useful hint.
Sadly, observed (perceived) quality is subjected to a Gaussian noise.

(Finding artifacts typically doesn't work in a high rate test, because it starts to seem like both tracks have the artifacts. Still I can still say which one is dirtier or cleaner, with limited reliability.)

In the spreadsheet B3:C3, the quality of track A and B is set. A is original, so it will be 5.0. B is set to 4.0 as a convenience.


Sighted evaluation, right?

You believe that ABXY is superior, and so it is for you.

Re: ABX vs. ABXY

Reply #35
From a programming perspective, it's a lot easier to just add the "Y Button" and cover both test scenarios. You don't have to ever click it, and you would be doing an ABX test, or use it and you would be doing ABXY.

And if you think that one test is easier than the other, then good! There should be no difficulty inherent to the test itself.

One program to rule them all !

EDIT: Personally I think I do AXY more than anything.

I find all the stuff in the FOOBAR2000 ABX add-on that is uniquely related to ABXY as a distraction, and think it hurts my results,

Re: ABX vs. ABXY

Reply #36
I use Foobar's ABX plugin, and what I usually do is first listen to A and B and figure out how they sound different, or in other cases, or I won't even need to listen to A and B, because I know what the difference is. This goes for something like volume level. So if I know that A is 0.2 dB louder than B, I will just listen to X and Y and see which one is louder. Then that must be A.
In the same way, after listening to A and B and thinking e.g. that A is brighter than B, then I just listen to X and Y from then onwards and then pick the brightest one as being A.
I hope this makes sense. I find using X and Y is very easy for me. I have basically used this method in all the blind tests I've done, although I also go back to A and B now and again if need be.
As you can see in the listening test forum, I passed with 15 out of 16 correct for a volume level difference of 0.2 dB, and that was done purely using X and Y.
"What is asserted without evidence can be dismissed without evidence"
- Christopher Hitchens
"It is always more difficult to fight against faith than against knowledge"
- Sam Harris

Re: ABX vs. ABXY

Reply #37
I will explain why ABXY is superior to ABX on high rate listening tests.

In the high rate test, only relative quality change between two adjacent play is usable as a useful hint.
Sadly, observed (perceived) quality is subjected to a Gaussian noise.

(Finding artifacts typically doesn't work in a high rate test, because it starts to seem like both tracks have the artifacts. Still I can still say which one is dirtier or cleaner, with limited reliability.)

In the spreadsheet B3:C3, the quality of track A and B is set. A is original, so it will be 5.0. B is set to 4.0 as a convenience.


Sighted evaluation, right?

You believe that ABXY is superior, and so it is for you.

Critics are good in mix with suggestions, ideas ... anything.
But You  don't help either.  Can You see it? 

 
SimplePortal 1.0.0 RC1 © 2008-2018