Detection 133Figure 5.13 (A) A noise band with a spectral notch or gap is used to mask
a signal whose frequency is in the center of the spectral gap. (B) The masked
thresholds for detecting a 1,000-Hz signal are shown as a function of in-
creasing the spectral notch of the band-reject noise. Source: From Yost
(2000), adapted from data of Patterson and Moore (1989), with permission.
effects obtained with tonal maskers and signals. As the width
of the band-reject noise increases, signal threshold is lowered
because there is less power in the critical band of frequencies
near that of the signal. The width of the critical band is pro-
portional to signal frequency as is consistent with frequency
tuning measured in the auditory periphery (Glasberg &
Moore, 1990). That is, as the frequency content of a signal in-
creases, the range of frequencies that are critical for masking
the signal also increases proportionally.
The concept of the critical band as a measure of auditory-
processing channels that are frequency tuned is closely tied to
the biomechanical and neural measures of processing in the
auditory periphery. This combination of physiological and
psychophysical evidence for frequency-tuned channels forms
a significant part of all current theories and models of audi-
tory processing (Moore & Patterson, 1986).
Many data from masking experiments, especially those in-
volving Gaussian noises and tonal signals, can be explained
using the energy detection model (Green & Swets, 1973) from
the general theory of signal detection (TSD). For instance, in
an experiment in which a Gaussian noise masks a tonal signal,
the energy detection model assumes that the energy of the
noise is compared to that of the signal plus noise. The noise
masker energy is a random variable that can be described with
a distribution with a known mean and standard deviation. The
addition of the signal to the noise often increases the mean of
the distribution, but not the standard deviation. As signal level
increases, the normalized (normalized by the common stan-
dard deviation) difference in the means of the distributions in-
creases. On any presentation, listeners use a sample of energy
to decide whether the signal plus noise or just noise was pre-
sented. If the likelihood of the sampled energy is greater than
a criterion value (set by the listener’s response proclivity or
bias), the listener responds that the signal plus noise was pre-
sented, because high signal levels are more likely to produce
high energy levels. A measure of performance (d) can be ob-
tained from the theoretical distributions of signal-plus-noise
and noise-alone conditions, and then can be compared to a
similardmeasure obtained from the listener’s data. Various
forms of the energy model and other models based on TSD
have been successful in accounting for a variety of masking
results (Green & Swets, 1973).Temporal MaskingThe masking data described so far are based on conditions in
which the signal and masker occur at the same time. Masking
also takes place when the signal and maskers do not tempo-
rally overlap. Forward masking occurs when the signal comes
on after the masker is turned off and backward masking oc-
curs when the signal precedes the masker. For the same tem-
poral separation between signal and masker, there is usually
more forward than backward masking. In the fringe condi-
tions, a short-duration signal is presented near the onset (for-
ward fringe) or offset (backward fringe) of a longer-duration
masker. Most often, the greatest amount of masking occurs in
these fringe conditions (masking overshoot).
As has already been described, the nonlinear properties
of auditory transduction can have several psychophysical
consequences. The existence of aural harmonics and differ-
ence tones is one such consequence. It is also probably the
case that there are suppressive effects that are a function of
some form of nonlinearity. That is, the masker may sup-
press or inhibit the excitatory effect of the signal under
different conditions. The separation of the signal and
masker in temporal masking conditions allows one to po-
tentially isolate these suppressive effects. The fact that psy-
chophysical tuning measured in forward masking generates
measures of narrower tuning (smaller critical bands) than
that obtained in simultaneous masking may be consistent
with such suppressive effects existing in the simultaneous
conditions (Moore, 1986).
Nonlinear peripheral processing is a compressive nonlin-
earity in which neural output is compressively related to sound