[R. Stanton]

Better than ABX?

ABX has become the standard test for comparison of audio components.
It is a valid test method, yet many people object to it. Their
complaint is often: the ABX test makes it too hard to hear differences
between components.

The ABX test provides three switches to the subject. He can switch to
A, B or X.
A and B are the two components tested. X is selected randomly by
machine. It is either A or B. The subject listens to A, B and X, then
decides if X matches A or matches B.

Normally, we compare components only two at a time. With the ABX test,
we must listen to three sound sources and decide which two are the most
similar. Comments from this group, have shown that many people consider
the ABX test stressful and confusing. One person who used the ABX test
for a group evaluation wrote: "... several of us noted that we had
great difficulty remembering what A had sounded like by the time we
got through with X." This is typical of the complaints about ABX.

I am proposing a new test. Let us call it: the X-Y test.

The computer selects (randomly) one of the following four combinations
of signals, AA, BB, AB, BA, and sends it to switches X and Y. In this
test, the subject uses only two switches. He does an AB comparison of
the two sounds and notes if they are the same, or different.

Sources
B --------------
Switches
A --------------

A --------------
--------- X
B --------------

A --------------
--------- Y
A --------------

B --------------
B --------------

After each trial, the computer records the answer, then randomly
selects another pair of signals to go to switches X and Y. The trials
continue until a reasonable accuracy is achieved.


How the test is scored:

The answers to all trials (AA, BB, AB, BA) are counted. Answers are
scored the same way as ABX system scores answers.



What if someone tries to cheat?

Suppose someone tried to cheat by putting down a false answer, such as:
"heard difference" when she actually didn't hear a difference? It
wouldn't work.

1) Answers of: "sounds different" to all trials would give a score
of 50% correct.
2) Answers of: "sounds the same" to all trials would give a score
of 50% correct.
3) Totally random answers to all trials would give a score of of 50%
correct.

A score of 50% correct indicates the subject can *not* hear a
difference. So cheating wouldn't work.

I think the X-Y test would be easier on the subject, than the ABX test,
and would give a more accurate indication of ones ability to hear a
difference in the components.

Bob Stanton