James M. Kates
City Univ. of New York, Graduate Ctr., Rm. 901, 33 W. 42nd St., New York, NY 10036
As a first step in designing an automatic hearing-aid adjustment system, a procedure is being developed for classifying common background noises. The intelligibility and quality of amplified speech depend on the type and level of the background noise as well as on the individual hearing loss. Thus to realize the maximum benefit from a hearing aid, the frequency response and processing characteristics should be set according to the type of background noise, and should change as the noise changes. The noise-classification procedure used 200-ms segments of noise, from which a parameter vector describing the spectral shape and the temporal envelope fluctuations is extracted. Parameter vectors for the different noises, averaged over 2 s of data, have been subjected to a cluster analysis to measure their similarity. Individual 200-ms noise parameter vectors are then assigned to the cluster having the nearest centroid, and performance results will be presented in terms of the percent of noises correctly classified. [Work supported by NIDCD.]