Speaker Variability in Intelligibility Testing

Frequently, when recorded test lists are used in intelligibility testing of communication equipment, only a very few speakers are used to provide the source material, in contrast with the relatively large number of listeners employed. Our studies have shown that such a small number of speakers as two or three is not sufficient to provide a fair evaluation in all cases. Particularly, in evaluating “speaker-sensitive” equipment, some extraordinary differences among speakers have been found. Thus, it is easy to see how such equipment might be unfairly biased by the use of a small number of speakers, all of whom happened to be not well suited to it. One cannot, how, use large numbers of speakers speaking all test lists since the amount of source material would be prohibitive. A solution is to rotate a large number of speakers within each test list so that each list provides a sampling of all the the speakers used.