ISCA Archive ICSLP 1998
ISCA Archive ICSLP 1998

Speech intelligibility testing for new technologies

Susan L. Hura

There are several tests of speech intelligibility currently available which employ a variety of methods. The most appropriate method for testing intelligibility of speech transmitted via telephony is a forced choice task in which listeners hear speech samples and identify what they hear from among a set of alternatives displayed onscreen. This methodology allows tests to be run quickly and scored automatically. A major flaw in existing forced-choice intelligibility tests is the use of unfamiliar words, nonwords, and proper names along with common words. A stimulus set that is mixed in this way may introduce response biases into the test and therefore produce results that are less predictive of actual intelligibility performance. The Intelligibility of Familiar Items Test (IFIT) ameliorates several methodological flaws found in earlier tests. The IFIT uses a stimulus set composed of high familiarity real English words and tests consonants in initial and final word position and vowels in word medial position.

doi: 10.21437/ICSLP.1998-386

Cite as: Hura, S.L. (1998) Speech intelligibility testing for new technologies. Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998), paper 0042, doi: 10.21437/ICSLP.1998-386

  author={Susan L. Hura},
  title={{Speech intelligibility testing for new technologies}},
  booktitle={Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998)},
  pages={paper 0042},