A study of two standard speech intelligibility measures

Abstract

In the diagnostic rhyme test (DRT) participants identify each test word by chosing one of two response alternatives that differ only in their initial consonants and only by a single binary distinctive feature. Although the DRT is an accepted industry and military standard for measuring initial consonant intelligibility, other intelligibility measures may be more appropriate when stimuli produce near maximum DRT scores. Moreover, the DRT incorporates assumptions that may be valid for natural speech, but are untested for high‐quality, low‐bit rate coded speech. In particular, the DRT implicitly assumes that segmental intelligibility can be adequately measured by examining only single feature confusions. To examine this assumption, subjects were asked to identify the initial consonants of consonant‐vowel syllables. Multifeature confusions were far more common with coded speech than with natural speech. Moreover, the consonant identification procedure reliably discriminated between speech coding devices that had near maximum, statistically indiscriminable DRT scores. These results suggest that open‐response identification procedures may be more suitable than the DRT for evaluating high‐quality, low‐bit rate coders.