Abstract
An adaptable, Finnish language, “speech in noise” test was developed using a personal computer equipped with a sound card. Each of the 1000 test items stored as a separate digitized wave form file on the hard disk of a personal computer consisted of disyllabic words on one stereo track and synchronized speech noise on the other. Because only a few randomly selected words are presented in this test for SRTN (speech recognition threshold in noise or S/N ratio corresponding to 50% recognition) the selection and equalization of test material is considered to be crucial to the achievement of reproducible results in short time. Equalization of the test items (word + noise) in accordance with degree of difficulty, by adapting the noise signal to the properties of the corresponding word, and selection of 510 of the initial 1000 recordings with the smallest SDs are described. The effect of this procedure on the test-retest reliability of testing SRTN is evaluated. Despite contrary expectations, the procedure appears to have no effect on the reliability of the speech recognition threshold in noise (SD from 1.5 to 1.7 dB).
Key Words: