Lapsley Miller, J. A., Drga, V., and Miller, L. M.
Poster presentation at the 2001 ARO conference
Assoc. Res. Otolaryngol Abs. 24, 250.
The energy detector is usually proposed as the ideal observer for the
detection of narrow-band short-duration Gaussian noise, masked by wider-band
Gaussian noise. For some values of the bandwidth-duration (WT) product, however,
the energy detector (filter, square-law rectifier, and true integrator) is not
as good as a full-linear detector (filter, linear rectifier, and true
integrator). Because human decisions in detection tasks are subject to
inconsistency, performance rarely approaches that of an ideal observer. This
makes it difficult, if not impossible, to evaluate whether the energy detector
is appropriate for modeling human detection. By repeating a noise-in-noise
detection experiment multiple times using the same stimuli, then using group
operating characteristic (GOC) analysis and function-of-replications-combined
estimation (FORCE) analysis, the effects of observer inconsistency are minimized
and asymptotic error-free performance estimated, respectively. At least for one
combination of noise bandwidth and duration, we show that a human observer can
do better than the energy detector and do as well as a full-linear detector,
once error due to inconsistency is removed. This finding implies that the energy
detector is not necessarily an appropriate model for the task of noise-in-noise
detection in humans.
View poster in pdf format
08 Nov 2009 04:37 PM