ASA 124th Meeting New Orleans 1992 October

1aPP9. On the measurement of source separation.

Gregory H. Wakefield

Paul R. Runkle

Dept. of Elec. Eng. and Comput. Sci., Univ. of Michigan, Ann Arbor, MI 48109-1122

It has been conjectured that certain acoustic cues present in a complex waveform are used by the auditory system to decompose that waveform into a set of auditory sources. The disparity between the perceived source(s) and the physical sources in such phenomena as stereophony suggests that there is not necessarily a one-to-one correspondence between the acoustic composition of signals and the perceptual decomposition of the combined signal into auditory sources. In an earlier paper [Wakefield and Runkle, Issues in Advanced Hearing Aid Research Conference (May 1992)], a method was proposed for measuring the strength of such cues adaptively, however, the hypothesis that the measuring technique itself was responsible for the results could not be rejected. In the present paper, an alternative measuring tool is introduced that is capable of independently suppressing and enhancing different spectral regions, unlike the original tool. Results are presented concerning the salience of co-modulation, harmonicity, onset and offset cues, and amplitude spectrum with respect to auditory sources using the new tool. A recognition paradigm is integrated into the proposed method to help validate the results. [Work supported by grants from the NIH NIDCD and from Ford Motor Company.]