Some aspects of methodology in speech audiometry.

The aim of this study was to provide more knowledge about the influence of various factors on the reliability of speech audiometric measurements. On the basis of these data, new material on speech has also been developed for measurements of the speech reception threshold in noise. In a theoretical and a clinical investigation it has been shown that the reliability of speech discrimination scores may be predicted by using the binomial distribution. The reliability mainly depends on the score obtained and on the length of the word list. Curves with confidence intervals as well as diagrams which indicate when two different discrimination results deviate significantly from each other are presented. On the whole, the reliability is surprisingly poor. The reliability of speech reception threshold (SRT) measurements was studied by computer simulations and clinical studies. The SRT results of "subjects" who had poor speech discrimination were unreliable and were greatly affected by the level at which the test started. Some new methods were also simulated, one of which (the 2x4-word method) was 8% faster and also somewhat more reliable in the clinical studies than the ordinary 10-word method. It was also less affected by the patient's speech discrimination. Several speech lists with exactly the same content of sound, but with new sentences, were obtained by editing a list of ten spoken Swedish sentences in a computer. A noise was synthesized from the speech material by the computer to obtain the exact same spectrum of the speech and the noise. This material was tested monaurally on normal-hearing subjects and hearing-impaired subjects. A very steep intelligibility curve was obtained in noise (25%/dB), which gave a standard deviation of only 0.44 dB at repeated threshold measurements in noise for the normal-hearing subjects when the learning effect was outbalanced. This figure increased to 1.1 dB for those with severely impaired hearing. The learning effect between the first and the second thresholds did not exceed 1 dB. The threshold values ranged from -8 to -5 dB signal-to-noise ratio for those with normal hearing and from -7 to +7 dB for those with impaired hearing. The new sentence and noise material developed is well suited for clinical measurements of the speech reception threshold in noise.