Is intelligibility of adjacent passbands hypoadditive or hyperadditive?

Abstract

Based on their own findings and reports from other laboratories, H. Müsch and S. Buus [H. Müsch and S. Buus, J. Acoust. Soc. Am. 109, 2896-2909 (2001)] suggested that when heard together, the intelligibilities of adjacent passbands were hypoadditive, and those of disjoint passbands were hyperadditive. A subsequent study employed extremely high order Finite Impulse Response (FIR) filtering that had been shown to effectively eliminate contributions from transition band slopes [R. M. Warren, J. A. Bashford, Jr., and P. W. Lenz, J. Acoust. Soc. Am. 118, 3261-3266 (2005)]. That study measured the intelligibility for each of the 15 possible pairings of six one-octave effectively rectangular passbands (3 dB/Hz filter skirts) that spanned the speech spectrum with center frequencies ranging from 0.25 to 8 kHz. Each pairing, whether contiguous or disjoint, exhibited hyperadditivity. The present study determined whether decreasing the filter skirts to 0.5 dB/Hz (considered quite steep by conventional standards) would produce the hypoadditivity reported in literature for adjacent bands. Results obtained support the hypothesis that redundancy introduced by overlapping transition band slopes could be responsible for the redundancy correction factor employed by some models for estimating intelligibility of paired adjacent passbands. [Work supported by NIH.].