Systematic relationship between variance and mean intelligibility found in Diagnostic Rhyme Test (DRT) scores

Abstract

The mean intelligibility score obtained with the Diagnostic Rhyme Test (ANSI S3.2‐1989) is the average of an ensemble of scores for phonetic features. From analyzing diagnostic scores from multispeaker tests evaluated with eight listeners, it was found that a systematic relationship exists between the mean scores and the variance of the ensemble of feature scores. The data set studied included 64 multispeaker tests (44 tests with equal numbers of male and female talker subjects, and 20 with three male talkers; the tests with male and female talkers included ten 18‐speaker tests and thirty‐four 6‐speaker tests). Twenty‐four diagnostic scores were obtained for each speaker in each test. A least‐squares fit to this data resulted in the linear regression equation: variance (of feature scores) = 2399 − 24.46×mean DRT score, with r2 = 0.96. This corpus of data conformed with the cited equation with a mean‐square error of 0.97 in estimating standard deviation in relation to a test's mean DRT score. From the tren...