142 Audition
over time. This is in contrast to the critical-band approach to
explaining auditory processing, in which only a narrow
region of the spectrum (in the critical band) is processed in a
very short period of time. As pointed out above, explanations
of profile analysis, CMR, and MDI all assume wide-band
spectral processing and procedures like auditory stream seg-
regation emphasize the importance of information integra-
tion over a long period of time.
Bregman (1990) has approached explanations of sound
source segregation from a perceptual point of view, borrowing
many concepts from the Gestalt school of perception. Several
computational models have been developed to account for
aspects of sound source segregation, especially those from
auditory stream segregation experiments. These models are
usually based on pattern recognition computations that inter-
rogate spectral-temporal patterns generated by modeling the
processes of the auditory periphery (Patterson, Allerhand, &
Giguere, 1995). Computational models of the auditory
periphery simulate the frequency-resolving properties of the
cochlear partition (often using a bank of band-pass filters) and
simulations of hair cell transduction of stereocilia displace-
ment to neural discharges in the auditory nerve (Meddis &
Hewitt, 1992). The pattern recognizers are neural nets or sim-
ilar methods of computation that attempt to segregate the
spectral-temporal neural patterns into subparts, whereby
each subpart may reveal the spectral-temporal structure of a
particular sound source. The cues discussed in this chapter, as
well as a priori information about the stimulus context or prior
learning about the stimulus context, are used to segregate the
overall spectral-temporal pattern into these subparts. These
models clearly imply that sound source segregation is based
on processing the spectral-temporal code provided by the
auditory periphery, and hence sound source segregation is a
central process (Meddis & Hewitt, 1992 ). As of yet, little
direct physiological data are available that can be used to help
guide these modeling efforts.AN OVERVIEW OF THE FUTURE STUDY
OF AUDITIONA great deal of what is known about hearing comes from un-
derstanding the causes of hearing loss and its treatment. The
major links in the hearing process that are most vulnerable to
damage are the intricate structures of the inner ear, especially
the hair cells. The study of the function of inner and outer hair
cells and the exact consequences each plays in hearing will
continue to be a major research focus in audition. The recent
suggestions that the compressive nonlinear properties of
cochlear transduction are derived from outer hair cell function
have led to a better understanding of auditory perception in
both people with normal hearing and those with impaired
hearing. Perhaps, however, the most exciting discovery con-
cerning hair cells is the fact that hair cells in birds, fish, and
probably amphibians regenerate after damage due to overex-
posure to either sound or ototoxic drugs (Tsue, Osterle, &
Rubel, 1994). These regenerated hair cells in birds appear to
function normally in support of normal hearing, but addi-
tional work is needed to fully understand the perceptual abili-
ties of these animals with regenerated hair cells. Hair cells in
mammals do not regenerate. The quest is on to determine why
hair cell regeneration occurs in some nonmammals but not in
mammals. The ability to regrow hair cells could, for many dif-
ferent types of hearing loss, be the ultimate hearing aid.
The study of hair cell regeneration is one of many areas in
which genetic techniques are supplying new and important
facts about auditory function. In addition to revealingFigure 5.16 Both the basic MDI task and results are shown. The basic task
for the listener is depicted along the bottom of the figure. The listener is to
detect a decrement in the depth of probe amplitude modulation (difference
between low and high depth). When just the probes are presented the task is
relatively easy. When an unmodulated masker tone with a frequency differ-
ent from that of the probe is simultaneously added to the probe, threshold for
detecting a decrease in probe modulation depth is not changed much from the
probe-alone condition. However, when the masker is modulated with the
same rate pattern as the probe, the threshold for detecting a decrement in
probe modulation depth increases greatly, indicating that modulation depth is
difficult to detect when both the probe and masker are comodulated. When
the masker is modulated, but with a different rate (shown as a faster rate in the
figure) than the probe, then the threshold for detecting a modulation-depth
decrement is lowered. The waveforms are not drawn to scale. Source:From
Yost (2000), based on data from (Yost, 1992b), with permission.