Prediction and statistical evaluation of speech recognition test scores.

A speech test evaluation and presentation system is described. The test presentation subsystem has the flexibility and speed of live-voice testing while using recorded test materials. The speech test evaluation subsystem compares an individual subject's test performance on a monosyllabic word test with that of an average person with the same hearing loss. The elements needed to make such evaluations are discussed. Also, a trial of the procedure is described. The primary purpose of the trial was to obtain data that would provide a basis for statistical probability statements about individual monosyllabic word test results obtained in clinical settings. Data were collected from three audiology clinics in three different types of settings. Except for a few cases with highly asymmetric speech scores, all nonconductive hearing losses were included. Subject ages ranged from 8 to 92 years. Importance-weighted average pure-tone hearing losses ranged from 0.4 to 97.6 dB HL. Fifty-word recognition scores and audiograms for 2609 ears were included in the main analysis. Twenty-five-word recognition scores and audiograms for another 932 ears from one clinic were used in a subsidiary analysis. Results indicated that distributions of absolute speech recognition scores in hearing-impaired samples are highly skewed. However, after transformation of the scores into rationalized arcsine units (rau), the differences between individual subject scores and scores predicted from the audiogram were reasonably well described by the normal distribution. The standard deviation of this distribution of differences, for the data combined across the three audiology clinics, was approximately 13 rau.