Abstract:

A correlational method [K. A. Doherty and C. W. Turner, J. Acoust. Soc. Am. 100, 3769--3773 (1996)] was used to estimate the relative weights of four frequency bands for the 258-item UCLA nonsense syllable test in ten normal-hearing listeners. Random levels of corresponding bandpass noise were added to the speech tokens on each trial. From the trial-by-trial data, correlations between the noise level and listener's performance yielded an estimate of the relative weight placed upon each band. The weighting functions were very similar across all listeners. This procedure appears to be a simple, fast, and reliable method for measuring speech recognition frequency weighting functions in individual subjects. The correlational weighting function showed differences from the frequency importance function derived using traditional articulation index (AI) filtering experiments. Some speech bands that were shown to contain significant speech information by the AI methods were not highly weighted when all four bands were presented to the listeners in the correlational procedure. It is hypothesized that these differences reflect the way in which listeners use speech information that is redundant across frequency regions. [Work supported by NIDCD.]