Statistical Derivation of an Importance Function for Speech at Threshold

Abstract

Fletcher has based his prediction of hearing loss for speech from an audiogram upon weights derived from the function ∫0fG1df. His equation for predicting hearing loss is considered analogous to a linear, multiple-regression equation. If the analogy is appropriate then it should be possible to reconstruct the ∫0f G1df function from a set of multiple-regression weights computed by least squares technique from a set of data giving hearing loss for speech and for 250, 500, 1000, 2000, and 4000 cps. Multiple-regression computations were carried out on the set of data given by Fletcher and on a set of data assembled by the writer. There is a discrepancy between the function as derived from data on filtered speech and as derived from the multiple-regression technique applied to data collected from hard-of-hearing subjects. It is argued that this discrepancy is due to the difference between the speech materials used in the original determination (monosyllables) and that conventionally used to measure hearing loss for speech clinically (spondees). [This research was supported by a grant (B-243) from the National Institutes of Health.]