For the ABX mode I plan to put into my listening utility, I was thinking that it might be more consistent for me to use something which mirrors the way the main program works. That way, people who are just starting to learn how to do blind tests don't have to learn two different ways of doing it.

So instead of X being the unknown quantity, I would rather randomize A and B, and have X be a constant reference. Then I would ask the listener decide whether A or B is the reference. The statistics are the same as ABX. I'll probably call it something like "repeated listening test panel" or something like that to avoid confusing it with Arny Krueger's PC-ABX method. Maybe I'll even put mini-sliders in there to further reinforce the idea that it's really the same method as the main panel, except repeated for reliability.

Also, I plan to use the mode where the progress is updated after each trial, so there will be no "takebacks." Arny's program allows takebacks if it is in the mode where results are reported only at the end of a test.

ff123

Excellent. Seems like i can ditch PCABX soon, the interface is an offense for the eye in Windows XP.

In the ABX comparator I'm working at (from time to time), I have A and B known references, and X and Y as unknown versions of these. You can switch between these 4 ABXY, and then decide wether X is A and Y is B, or X is B and Y is A. I think this is a more sensitive method.

What do you think?

My reference book lists several different types of difference tests (for taste tests):

1. Triangle test: Although it is statistically more efficient than the paired comparison and duo-trio methods, the Triangle test has limited use with products that involve sensory fatigue, carryover, or adaptation, and with subjects who find testing three samples too confusing.

Present to each subject three coded samples. Instruct subjects that two samples are identical and one is different (or odd). Ask the subjects to taste (feel, examine) each product from left to right and select the odd sample.

2. Duo-trio test: The Duo-trio test is statistically inefficient compared with the Triangle test because the chance of obtaining a correct result by guessing is 1 in 2. On the other hand, the test is simple and easily understood. Two forms of the test exist: the constant reference mode, in which the same sample is always the reference, and the balanced reference mode, in which both of the samples being compared are used at random as the reference. Use the constant reference mode with trained subjects whenever a product well known to them can be used as the reference. Use the balanced reference mode if both samples are unknown or if untrained subjects are used.

Present to each subject an identified reference sample, followed by two coded samples, one of which matches the reference sample. Ask subject to indicate which coded sample matches the reference.

PC-ABX is a Duo-trio test with balanced reference, if I interpret the description properly.

3. Two-out-of-Five test: This method is statistically very efficient because the chances of correctly guessing two out of five samples are 1 in 10, as compared with 1 in 3 for the Triangle test. By the same token, the test is so strongly affected by sensory fatigue and by memory effects that its principal use has been in visual, auditory, and tactile applications, and not in flavor testing.

Present to each subject five coded samples. Instruct subjects that two samples belong to one type and three to another. Ask the subjects to taste (feel, view, examine) each product from left to right and select the two samples that are different from the other three.

4. Simple difference Test: Present each subject with two samples, asking whether samples are the same or different. In half the pairs present a matched pair (the same sample, twice). Analyze the results by comparing the number of "different" responses for the matched pairs to the number of "different" responses for the different pairs, using the chi-square test.


Hmm, maybe I will do duo-trio with balanced reference, after all. It seems to be the simplest for beginners.

ff123

This test seems to be the simplest... quite interesting. I wonder how practical it would be to implement, and what drawbacks/advantages it may have, compared to the duo-trio test with balanced reference (PC-ABX)?

The difference is that it takes many more listenings to arrive at a significant result, thus greatly increasing the chances for listener fatigue.

ff123

Here's a screenshot of my current abx dialog box:

http://ff123.net/abchr/abx.gif

ff123

It's...hummm... kinda ugly, but seems cool.

This one has bigger play/stop buttons for easier use, and I tried to make the screen a little more compact. Can't do anything about the ugliness factor, though

http://ff123.net/abchr/abx1.gif

ff123

Please alias 'X is A' and 'X is B' to an easy to reach key(-combination).

--
GCP

Another request (which I'm badly missing in the Linux-ABX I always use)

When selecting an area of the sample, make it snap to a zero crossing and the start and end. (prevents loud POP's)

--
GCP

looks nice 2 me, simple=beautifull, i would change x=a and x=b into buttons (design vise).

Ohh, that's an interesting suggestion. I guess I'd have to read the wav files to be able to do that. Sounds like one for the distant future.

I will alias the "X is A" and "X is B" buttons to "j" and "k". The reason I used radio buttons instead of pushbuttons is to emphasize that one or the other must be chosen; plus it gives visual feedback as to what's been selected, or if anything's been selected at all.

ff123