Download Psychoacoustic Consequences of Compression in the Peripheral

Psychological Review
1998, Vol. 105, No. 1, 108-124
Copyright 1998 by Ihe American Psychological Association, Inc.
0033-295X/98/S3.00
Psychoacoustic Consequences of Compression
in the Peripheral Auditory System
Brian C. J. Moore
Andrew J. Oxenham
University of Cambridge
Institute for Perception Research
Input-output functions on the basilar membrane of the cochlea show a strong compressive nonlinearity at midrange levels for frequencies close to the characteristic frequency of a given place. This
article shows how many different phenomena can be explained as consequences of this nonlinearity,
including the "excess" masking produced when 2 nonsimultaneous maskers are combined, the
nonlinear growth of forward masking with masker level, the influence of component phase on the
effectiveness of complex forward maskers, changes in the ability to detect increments and decrements
with level, temporal integration, and the influence of component phase and level on the perception
of vowellike sounds. Cochlear hearing loss causes basilar-membrane responses to become more
linear. This can account for loudness recruitment, linear additivity of nonsimultaneous masking,
linear growth of forward masking, reduced temporal resolution for sounds with fluctuating envelopes,
and reduced temporal integration.
For many years, processing in the peripheral auditory system
has been modeled as a bank of overlapping bandpass filters,
which were assumed to be roughly linear (Fletcher, 1940; Patterson & Moore, 1986). This approach was partly motivated
by the fact that many aspects of simultaneous masking can be
described using such a model. For example, the threshold for
detecting a sinusoidal signal in a broadband noise masker corresponds to approximately a constant signaltmasker ratio, over a
Schematic illustrations of input-output functions are given
in Figure 1. The solid line shows the type of function that would
be observed for a tone at the characteristic frequency of the
place being studied. Note that for midrange levels the output
grows by less than 10 dB for each 10-dB increase in input level,
very wide range of masker levels (Hawkins & Stevens, 1950).
However, recent physiological studies of the mechanics of the
basilar membrane within the cochlea indicate that the response
to sinusoidal tones can be highly compressive. Input-output
functions show marked compression at medium input levels
(40-80 dB sound pressure level [SPL]) and become more linear at very low and very high levels (Rhode, 1971; Robles,
Ruggero, & Rich, 1986; Yates, 1990). The compression occurs
only for tones that are reasonably close to the characteristic
frequency for the place at which the response is being measured
(Robles et al., 1986). The characteristic frequency is the frequency giving the largest response at that place on the basilar
membrane.
indicating compression. This function can be approximated for
midrange levels (30-70 dB) by a power function; the shortdashed line shows a power function with an exponent, p, of
0.25, meaning that a 10-dB range of input levels is compressed
into a 2.5-dB range of response levels. The long-dashed line
shows the type of function that would be observed for a tone
with frequency well below the characteristic frequency. In this
case, the response grows in a linear manner (p = 1); each 10dB increase in input level gives rise to a 10-dB increase in
response.
It is widely believed that the compression on the basilar membrane depends on an "active" physiological mechanism within
the cochlea; it does not depend simply on the passive mechanical
properties of the basilar membrane and surrounding structures
(Yates, 1995). Running along the length of the basilar membrane are two types of receptor cells, which are called hair cells
because of the hairlike stereocilia at their tips. The inner hair
cells lie closest to the inside of the spiral-shaped cochlea and
they form a single row. They are thought to be responsible for
the detection of mechanical vibrations on the basilar membrane
and for transforming these vibrations into action potentials in
the auditory nerve. In other words, the inner hair cells act as
transducers. Most, if not all, of the information carried by the
auditory nerve is conveyed via the inner hair cells.
Brian C. J. Moore, Department of Experimental Psychology, University of Cambridge, Cambridge, England; Andrew J. Oxenham, Institute
for Perception Research, Eindhoven, The Netherlands.
This work was supported by the Medical Research Council (United
Kingdom), the European Community (Telematics and Informatics for the
Disabled and Elderly project), the Anglo-German Academic Research
Collaboration Programme, and a Wellcome Trust Travelling Research
Fellowship (0044215/Z/95/Z). We thank Joseph Alcantara, Brian
The outer hair cells lie closest to the outside of the spiralshaped cochlea and they form three to five rows. The outer hair
cells appear to play a crucial role in the active mechanism.
They are able to change their length in response to electrical or
chemical stimulation, and they appear to apply forces to the
basilar membrane in such a way as to enhance the sharpness of
tuning and to amplify the response to weak sounds. In other
Glasberg, Inga Holube, Robert Peters, and Chris Plack for their collaboration in the research described in this article. We also thank Joseph
Alcantara and Brian Glasberg for assistance in the preparation of figures.
Correspondence concerning this article should be addressed to Brian
C. J. Moore, Department of Experimental Psychology, University of
Cambridge, Downing Street, Cambridge CB2 3EB, England. Electronic
mail may be sent via Internet to [email protected].
108
PSYCHOACOUSTIC
CONSEQUENCES
100
90
80
70
/P-l
60
50
0
10
20
30
40
50
60
70
80
90
100
Input level. dB
Figure L The solid line shows a schematic illustration of an inputoutput function on the basilar membrane for a tone with frequency close
to the characteristic frequency. The short-dashed line shows that a power
function with exponent, p, equal to 0.25 provides a reasonable approximation to the input-output function for midrange sound levels. The
dashed line shows a linear input-output function (p = 1). This is typical
of what might be observed for a tone with frequency well below the
characteristic frequency.
words, the outer hair cells play a motor role. Exactly how they
do this is not fully understood, but many experiments have
shown that damage to the outer hair cells causes dramatic
changes in the mechanical responses of the basilar membrane
to low-level sounds (Ruggero, 1992; Ruggero & Rich, 1991;
Ruggero, Rich, Robles, & Recio, 1996); the tuning is broadened, and the sensitivity is decreased.
Of particular interest for us in this article is that the compression on the basilar membrane depends on the active mechanism,
mediated by the outer hair cells. The active mechanism amplifies
the response to weak sounds but applies progressively less amplification as the sound level increases (Yates, 1995). The ordinate in Figure 1 has been chosen so that, for the tone at the
characteristic frequency, an input level of 100 dB gives an output
of 100 dB, consistent with the idea that the active mechanism
plays little role at high levels (Yates, 1995). The "gain" for
input levels around absolute threshold is about 55 dB, consistent
with physiological estimates of the gain provided by the active
mechanism at medium frequencies (Patuzzi, Yates, & Johnstone,
1989; Yates, 1990, 1995).
The nonlinear processing probably does not have clear effects
in simultaneous masking of a tone by broadband noise because
the components of the noise that dominate the masking fall in
the same frequency region as the signal and are subject to the
same compression as the signal. In this case the compression
does not change the effective signal:masker ratio, and this is the
quantity that is usually assumed to determine the threshold in
simultaneous masking (Fletcher, 1940; Patterson & Moore,
1986). However, the compression may influence the ability to
detect changes in level (Moore, Peters, & Glasberg, 1996a),
and it may affect simultaneous masking when the masker and
signal frequencies are very different (Nelson & Schroder, 1996;
Oxenham & Moore, 1995). Compression may have large effects
when the signal and masker are presented nonsimultaneously,
as in forward masking or backward masking. This is based on
the assumption that the compression is fast acting, so that for
nonsimultaneous stimuli each stimulus is independently com-
OF COMPRESSION
109
pressed, even when they are close together in time; this is consistent with the physiological data (Yates, 1995). The compression
may also have substantial effects on the internal representation
of sounds, thus influencing timbre perception.
Cochlear hearing loss, also sometimes called sensory hearing
loss or sensorineural hearing loss, is usually associated with
damage to the hair cells in the cochlea. It appears that the outer
hair cells are generally more susceptible to damage than the
inner hair cells. Hence, it is relatively common to find cases in
which the damage is mainly confined to the outer hair cells, but
it is rare to find cases in which the damage is confined to the
inner hair cells; often damage to both inner and outer hair cells
occurs, with the latter predominating (Borg, Canlon, & Engstrom, 1995; Liberman, Dodds, & Learson, 1986; Moore,
1995b). Damage to the outer hair cells results in a loss of
sensitivity and in a linearization of the basilar-membrane transfer
function; essentially, the active mechanism appears to be rendered inoperative (Ruggero, 1992; Ruggero & Rich, 1991; Ruggero et al., 1996). Hence, psychoacoustic phenomena observed
for normal-hearing people that depend on the compression on
the basilar membrane should be different or absent for people
with cochlear hearing loss. Indeed, the differences between normal-hearing people and people with cochlear hearing loss can
give insight into the importance of the compressive nonlinearity.
In this article we consider some of the psychoacoustic consequences of compression on the basilar membrane for normalhearing people and the consequences of loss of the compression
in people with cochlear hearing loss. We show that many apparently disparate phenomena can be explained in terms of this
single underlying mechanism. In the work described below on
people with cochlear hearing loss, it is assumed that a substantial
part of the hearing loss is due to outer hair cell damage and the
consequent loss of basilar-membrane compression. The people
tested always demonstrated loudness recruitment, and this
almost certainly is associated with outer hair cell damage
(Moore & Glasberg, 1997).
The Additivity of Nonsimultaneous Masking
According to an energy detection model (Green & Swets,
1974), if two different maskers are equally effective (i.e., each
produces the same amount of masking), then the combination
of the two maskers should result in a doubling of the signal
energy required for threshold; this corresponds to an increase
in the signal level at threshold of 3 dB. In simultaneous masking,
this prediction sometimes holds, but often the combined masker
produces more masking than predicted (Lutfi, 1983; Moore,
1985). The amount by which signal thresholds exceed the energy summation prediction is referred to as excess masking.
Various mechanisms have been put forward to account for the
excess masking; for a review see Moore (1995a). However, the
fact that excess masking does not always occur suggests that it
cannot be explained in terms of a peripheral physiological process like compression.
A different argument can be made for nonsimultaneous masking. Two types of nonsimultaneous masking can be distinguished; in forward masking the signal is presented just after
the masker, whereas in backward masking it is presented just
before the masker. Combining two equally effective nonsimulta-
110
MOORE AND OXENHAM
neous maskers (one forward and one backward) consistently
results in excess masking, usually of 7-12 dB at moderate sound
levels (Cokely & Humes, 1993; Wilson & Carhart, 1971). This
excess masking can be explained if it is assumed that each
stimulus (the forward masker, signal, and backward masker) is
subjected to a compressive nonlinearity before the effects of the
stimuli are combined in a linear temporal integrator (Penner,
1980; Penner & Shiffrin, 1980), as illustrated in the model of
temporal resolution shown in Figure 2.
In the mode], there is an initial stage of bandpass filtering,
reflecting the action of the basilar membrane; each point on the
basilar membrane behaves like a bandpass filter. For simplicity,
only one filter is shown; in reality there is an array of parallel
channels, corresponding to different places along the basilar
membrane. Each filter is followed by a compressive nonlinearity
that mimics the compression occurring on the basilar membrane.
The output of the compressive nonlinearity is fed through a
"smoothing" device often referred to as a temporal integrator.
This sums the output of the compressive nonlinearity over a
certain time interval or "window." Usually, the window is characterized as a weighting function such as an exponential or a
pair of exponentials (Moore, Glasberg, Plack, & Biswas, 1988;
Oxenham & Moore, 1994; Plack & Moore, 1990). However,
the duration of the window can conveniently be described by
its equivalent rectangular duration (ERD; Moore et al., 1988),
which is typically about 7-11 ms. The window is assumed to
slide in time so that the output of the temporal integrator is like
a running average of the input. This has the effect of smoothing
rapid fluctuations while preserving slower ones. Usually, the
smoothing device is thought of as occurring after the auditory
nerve; it is assumed to reflect a relatively central process. The
output of the smoothing device is fed to a decision device. The
decision device may use different "rules" depending on the
task required. For example, if the task is to detect a brief temporal gap in a signal, the decision device might look for a ' 'dip''
in the output of the temporal integrator (Moore, 1997). For
high-frequency signals, the response of the initial bandpass filter
is very fast, and so the characteristics of the temporal integrator
play a major role in limiting temporal resolution. For example,
a gap in a signal may be smoothed by the temporal integrator
leading to only a very small dip in the output of the temporal
integrator.
Consider now the application of this model to excess masking.
If two equally effective nonsimultaneous maskers (one forward
and one backward) are presented together, then they will be
compressed independently (as they do not overlap in time), and
then their effects will be summed in the temporal integrator.
The integrator itself is a linear device, and so the internal effect
evoked by the two maskers is simply double the effect evoked
by either alone. Thus, to reach threshold, the level of the signal
Bandpass
f 11 tar
Figure 2.
has to be increased relative to the level required for a single
masker. In fact, to reach the signal threshold, the internal effect
of the signal must also be doubled. This requires more than a
3-dB increase in signal threshold because the signal itself is
independently compressed.
This theoretical approach was taken by Oxenham and Moore
(1994). In their work, the compression was modeled by a simple power function (as illustrated by the short-dashed line in
Figure 1). A good fit to their forward and backward masking
data was obtained when the rectified stimulus amplitude was
raised to a power between 0.5 and 0.7 (depending on the participant). This corresponds to raising intensity to a power between
0.25 and 0.35. If, for example, the stimulus intensity is raised
to the power 0.3, then a tenfold increase in power (corresponding to 10 dB) would be needed to double the internal effect of
the signal. Thus, for two equally effective maskers, one forward
and one backward, excess masking of 7 dB is predicted.
If excess masking is due to compression on the basilar membrane, and if cochlear hearing loss is usually accompanied by
a loss of compression, then hearing-impaired participants should
show less excess masking than normal-hearing participants. This
hypothesis was evaluated in an experiment of Oxenham and
Moore (1995, 1997). They tested 3 normal-hearing and 3 hearing-impaired participants. The latter were 73-77 years old and
were diagnosed as having a moderate cochlear hearing loss with
loudness recruitment; there was no evidence for any conductive
component to their hearing loss. The hearing losses were about
30-40 dB at low frequencies, increasing to 55-65 dB at the 4kHz test frequency.
Participants were tested in forward and backward masking
conditions. The signal was a 4-kHz sinusoid with a 2-ms steadystate portion, gated with 2-ms raised-cosine ramps. The maskers
(both forward and backward) consisted of low-pass filtered
white noise with a cutoff frequency of 9 kHz. The maskers had
a steady-state duration of 200 ms and were gated with 1-ms
raised-cosine ramps.
Initially, the threshold of the signal in quiet was measured for
each participant. Then the levels of the forward and backward
maskers were varied to individually produce 5, 10, 15, 20, and
25 dB of masking. Forward masker levels were determined for
masker-signal intervals of 5, 10, and 25 ms (defined as the
interval between the 0-voltage points of the envelope of the
electrical signal), and backward masker levels were determined
for signal-masker intervals of 1 and 5 ms. Once the masker
levels for each participant had been established, participants
were presented with pairs of equally effective maskers (one
forward and one backward), and signal thresholds in these combined masking conditions were measured.
The mean data, pooled across subject and condition, are
shown in Figure 3. Mean thresholds in the presence of two
N Coipresaive
nonlinearity
\
y
Temporal
integrator
N
Decision
y
A schematic model of temporal resolution in the auditory system.
device
PSYCHOACOUSTIC CONSEQUENCES OF COMPRESSION
111
or no excess masking at any level. These results are consistent
with the idea that excess masking results from compression in
•
5
35
Normal
the cochlea and that loss of compression reduces or abolishes the
excess masking. Note that the results for the hearing-impaired
participants show roughly linear additivity of masking when
O Impaired
intensity-like quantities are used. In other words, when compression on the basilar membrane is small or absent, the central
temporal integrator appears to integrate an intensity-like
30
•Q
quantity.
1 25
n
£
5 ao
The Growth of Masking in Forward Masking
c_
01
In forward masking, increments in masker level often do not
% 15
ID
e
produce equal increments in amount of forward masking. For
g 10
example, if the masker level is increased by 10 dB, the masked
threshold may only increase by 3 dB (lesteadt, Bacon, & Lehman, 1982; Moore & Glasberg, 1983). This contrasts with si-
5
multaneous masking, for which, at least for wideband maskers,
5
10
15
SO
25
30
threshold usually corresponds to a constant signal:masker ratio.
This effect can be quantified by plotting the signal threshold (in
decibels) as a function of the masker level (in decibels). The
Single-masker threshold (dB SL)
Figure 3. Mean thresholds in the presence of combined forward and
backward maskers. The solid line represents no change in threshold, the
short-dashed line represents a 3-dB increase, and the long-dashed line
is a fitted function, which is described in the text. SL = sensation level.
resulting function is called a growth-of-masking function. In
simultaneous masking such functions would have slopes close
to one. In forward masking the slopes are less than one, and
the slopes decrease as the time delay between the masker and
signal increases. This is illustrated in Figure 4 (data are from
Moore & Glasberg, 1983). The masker was a broadband noise,
equally effective maskers are plotted against thresholds in the
presence of a single masker (in decibels sensation level [SL]).
Thresholds for the hearing-impaired group are denoted by the
open circles, and thresholds for the normal-hearing group are
denoted by the filled circles. The error bars represent plus-andminus one standard deviation. The solid line indicates no change
and the signal was a brief 4-kHz sinusoid presented at various
times following the end of the masker.
Oxenham and Moore (1995) have suggested that the shallow
slopes of the growth-of-masking functions can be explained, at
least qualitatively, in terms of the compressive input-output
function of the basilar membrane. Such an input-output func-
in threshold in the presence of two maskers, and the shortdashed line indicates a 3-dB increase, as predicted by an energy
17.5
detection model. Thus, the amount of excess masking is given
as the amount by which the data points lie above this shortdashed line.
For the normal-hearing participants, the amount of excess
30
masking increases with level, reaching 9 dB at levels of 20 and
25 dB SL. The long-dashed curve, which fits the data well, was
calculated using a model of the type proposed by Humes and
27.5
'
-
37.5
Jesteadt (1989). They modified a simple power function by
20
including a constant to account for the absolute threshold. When
applied to the case of two equally effective maskers, the equation
becomes
X2 = 10 log [2(10*"°)' - I]1
(1)
E
10
where Xi is the signal threshold (in decibels SL; that is, decibels
relative to the absolute threshold) in the presence of two maskers: X is the signal threshold in the presence of a single masker,
and p is the compressive exponent. A value of/>= 1 corresponds
to linear additivity (energy summation), and excess masking is
predicted if p < 1. The best fitting value of p for the normalhearing participants was 0.2, corresponding to a compression
ratio of 5:1.
The data from the hearing-impaired participants showed little
0
10
20
30
40
50
Masker spectrum level. dB
figure 4. Thresholds for detecting a 4-kHz sinusoidal signal following
a broadband noise masker. The signal threshold is plotted as a function
of the masker spectrum level. The parameter is the delay time from the
end of the masker to the end of the signal.
112
MOORE AND OXENHAM
tion is shown schematically in Figure 5. It has a shallow slope
for medium input levels, but a steeper slope at very low input
levels. Assume that, for a given time delay of the signal relative
to the masker, the response evoked by the signal at threshold
is directly proportional to the response evoked by the masker.
Assume, as an example, that a masker with a level of 50 dB
produces a signal threshold of 12 dB. Consider now what happens when the masker level is increased by 20 dB. The increase
in masker level, denoted by AM in Figure 5, produces a relatively small increase in response, AO. To restore the signal to
threshold, the signal has to be increased in level so that the
response to it increases by AO. However, this requires a relatively small increase in signal level, AS, as the signal level falls
in the range in which the input-output function is relatively
steep. Thus, the growth-of-masking function has a shallow
slope.
According to this explanation, the shallow slope arises from
the fact that the signal level is lower than the masker level, so
the masker is subject to more compression than the signal. The
difference in compression applied to the masker and signal increases with increasing difference in level between the masker
and signal. Hence the slope of the growth-of-masking function
should also decrease with increasing difference in level between
the masker and signal. This can account for the progressive
decrease in the slopes of the growth-of-masking functions with
increasing time delay between the signal and masker (see Figure
4); longer time delays are associated with greater differences
in level between the signal and masker.
A prediction of this explanation is that the growth-of-masking
function for a given signal time delay should increase in slope
if the signal level is high enough to fall in the highly compressive
region of the input-output function. Such an effect can be seen
in the growth-of-masking function for the shortest delay time
in Figure 4; the function steepens for the highest signal level.
The effect is also apparent, but less clearly so, for the longer
time delays of the signal.
Oxenham and Plack (1997) have investigated forward masking for a 6-kHz sinusoidal masker and a signal of the same
100
90
BO
70
0
10
20
30
40
50
60
70
80
90
100
Input level, dB
Figure 5.
The curve shows a schematic input-output function on the
basilar membrane. When the masker is increased in level by AM, this
produces an increase in response of AO. To restore signal threshold,
the response to the signal also has to be increased by AO. This requires
an increase in signal level, AS, which is markedly smaller than AAf.
100
Normal hearing
O 3-kHz Masker
90
• 6-KHz Masker
BO
60
50
40
40
SO
60
70
Masker level
BO
90
100
(dB SPL)
Figure 6. Data are from Oxenham and Plack (1997) for normal-hearing participants. Thresholds for detecting a 6-kHz signal following a 3kHz or 6-kHz masker are shown. The signal level was fixed, and the
masker level was varied to determine the threshold. Symbols represent
the mean thresholds of 3 normal-hearing participants. Error bars represent plus-and-minus one standard error of the mean. They are omitted
where they would be smaller than the relevant data point. SPL = sound
pressure level.
frequency. They showed that under certain conditions, if the
signal is made very brief and the time delay between the masker
and signal is very short, the level of the signal at threshold is
approximately equal to the masker level. Under these conditions,
the growth-of-masking function has a slope of one; each 10-dB
increase in masker level is accompanied by a 10-dB increase in
signal level. This is illustrated in Figure 6 (filled symbols), and
it is consistent with the explanation offered earlier. When a
masker frequency well below the signal frequency was used (3
kHz instead of 6 kHz), the growth-of-masking function had a
slope much greater than one; a 10-dB increase in masker level
was accompanied by a 40-dB increase in signal level, as shown
by the open symbols in Figure 6. This can be explained in the
following way: The signal threshold depends on the response
evoked by the masker at the signal characteristic frequency. The
growth of response on the basilar membrane for tones well
below the characteristic frequency is linear (see the long-dashed
line in Figure 1). Thus, the signal is subject to compression
whereas the masker is not (essentially the opposite of the situation illustrated in Figure 5). This gives rise to the steep growthof-masking function. The slope value of around 4 corresponds
to a compressive power exponent of about 0.25, which is in
good agreement with the estimates of Oxenham and Moore
(1994, 1995).
If the compression on the basilar membrane is lost as a consequence of cochlear hearing loss, then the growth-of-masking
functions in forward masking (in decibels per decibels) should
113
PSYCHOACOUSTIC CONSEQUENCES OF COMPRESSION
have slopes close to unity, except when the signal is very close
to its absolute threshold. Furthermore, the slope should remain
close to unity, regardless of the relative frequencies of the masker
and signal, as all frequencies should be processed linearly. Empirical data have confirmed these predictions (Oxenham &
Moore, 1995, 1997; Oxenham & Plack, 1997). This is illustrated in Figure 7, which shows individual data from 3 participants with moderate cochlear hearing loss in the same conditions
as those used for the normal-hearing participants in Figure 6.
In contrast to Figure 6, all 3 hearing-impaired participants in
Figure 7 show linear growth-of-masking functions for both the
6-kHz masker and the 3-kHz masker. This is consistent with the
view that cochlear damage results in a loss of basilar-membrane
compression.
The Influence of Component Phase on the Effectiveness
of Complex Forward Maskers
The envelope of a sound resembles a kind of "outline" of
the sound. It may be determined graphically by drawing a
smooth curve through the peaks in the fine structure of the
waveform of the sound; for a more rigorous way of denning
the envelope see Glasberg and Moore (1992). By manipulating
the phases of the components in a complex sound, it is possible
to create waveforms that differ markedly in their envelopes,
even though the sounds have identical power spectra. Sounds
differing in their envelopes may be processed differently by the
auditory system, especially if the waveforms are subjected to
fast-acting compression on the basilar membrane. However, in
this case, the important factor is not the envelopes of the stimuli
themselves but the envelopes after filtering on the basilar
membrane.
100
90
Impaired hearing
MV
AR JK
D
> O 3 kHz
•
^ + 6 kHz
e
Figure 8. Waveforms of harmonic complex tones with components
added in Schroeder-positive phase or Schroeder-negative phase.
80
n
70
Kohlrausch and Sander (1995) used complex tones containing many equal-amplitude harmonics of a 100-Hz fundamental. The phase, (?„, of the nth harmonic was determined
according to equations described by Schroeder (1970):
= +jr«(n
= -im(n + l ) / N ,
(2)
40
where N is the number of harmonics in the complex sound. In
this equation, if a plus sign is used ( Schroeder-positive phase),
40
50
60
70
BO
90
100
Masker level WB SPU
Figure 7. Data are from Oxenham and Plack (1997) for 3 participants
with cochlear hearing loss. Individual data from participants M.V., A.R.,
and J.K. are plotted. Error bars represent plus-and-minus one standard
error of the mean. SPL = sound pressure level.
the phase increases progressively with increasing harmonic
number; whereas if a minus sign is used (Schroeder-negative
phase), the phase decreases progressively with increasing harmonic number. Both of these phases give waveforms with almost
flat envelopes; one waveform is a time-reversed version of the
other; see Figure 8. The Schroeder-negative wave is like a repeated
upward frequency
sweep (up chirp), whereas the
114
MOORE AND OXENHAM
Schroeder-positive wave is like a repeated downward frequency
sweep (down chiip).
Kohlrausch and Sander (1995) presented evidence that the two
waveforms give very different responses on the basilar membrane
at a place tuned to 1100 Hz; the Schroeder-positive phase leads
to a very "peaky" waveform, whereas the Schroeder-negative
phase leads to a waveform on the basilar membrane for which
the envelope remains nearly flat. They showed this by using the
complex tones as maskers and by measuring the threshold for a
brief 1100-Hz sinusoidal signal presented at various times during
the 10-ms period of the masker. For the Schroeder-negative phase,
the threshold of the signal varied little with the time of presentation. For the Schroeder-positive phase, the threshold varied over
a considerable range, indicating that the internal representation
of the masker was fluctuating within one period.
Carlyon and Datta (1997) took this experiment a stage further.
They reasoned that if the waveforms were subjected to fastacting compression, then the more peaky waveform (Schroederpositive phase) should lead to lower average excitation in the
auditory system than the less peaky waveform (Schroeder-negative phase). To test this idea, they used the two complex sounds
as forward maskers. They reasoned that the amount of forward
masking should provide an index of the average excitation
evoked by each masker. For medium-to-high masker levels, the
masker with Schroeder-negative phase gave more masking than
the masker with Schroeder-positive phase, consistent with their
prediction. However, at low levels, for which the input-output
function on the basilar membrane is less compressive, the difference was absent.
These results confirm that the compression on the basilar
membrane is fast acting, being able to influence the effective
waveform within each 10-ms period of the stimulus. This is
consistent with other psychoacoustic data (Moore, Wojtczak, &
Vickers, 1996b); see also the experiments on vowel perception
described later. Indeed, physiological data (Ruggero, 1992; Ruggero et al., 1996) suggest that the compression can act over
time scales much shorter than 10 ms, operating on individual
cycles of tones close to the upper frequency limit of hearing.
(1995) and Nelson and Schroder (1996). To detect the change
in level, participants may monitor places on the basilar membrane tuned well above the signal frequency, a process known
as off-frequency
listening. The background noise in the experiment of Moore et al. would have prevented off-frequency listening, forcing the participants to monitor a place on the basilar
membrane tuned close to the signal frequency.
Moore et al. (1996a) found that performance improved with
increasing frequency for all decrement and increment durations,
and it also tended to improve with increasing level—especially
at 2000 and 4000 Hz. The results obtained for 4 participants at
4000 Hz are shown in Figure 9. The results were analyzed using
a four-stage model similar to that shown in Figure 2. It consisted
of a bandpass filter .centered on the signal frequency, a compressive nonlinearity, a sliding temporal integrator, and a decision
mechanism. It was assumed that threshold was reached when
the increment or decrement produced a change (at the output
of the temporal integrator) that equaled a criterion value, AO.
Initially, the results were analyzed by assuming a simple
power-law compression; in such a model, the amount of compression does not decrease at high levels (see the short-dashed
line in Figure 1). The improvement in performance with increasing level could, in principle, be accounted for by assuming
that the equivalent rectangular duration (ERD) of the temporal
integrator decreases with level, the value of AO decreases with
level, or both (Moore et al., 1996a; Peters, Moore, & Glasberg,
1995). However, Moore et al. (1996a) found that the effects of
level could be accounted for even better by assuming that the
compressive nonlinearity was similar in form to the basilar
membrane input-output function (as illustrated by the solid
line in Figure 1). In this case, a very good fit to the data could
be obtained when both the ERD of the temporal integrator and
the value of AO were assumed to be invariant with level (although they could vary across frequency).
The goodness of fit of the model obtained in this way was
10
Detection of Increments and Decrements in Sinusoids
The compression on the basilar membrane can also influence
the way that the ability to detect small changes in level is affected by overall level and frequency. Moore et al. (1996a)
measured thresholds for the detection of brief decrements in
level of sinusoidal signals as a function of decrement duration,
level (70, 80, and 90 dB SPL), and frequency (250, 500, 1000,
2000, and 4000 Hz). Thresholds for detecting a 10-ms increment in level were also measured. When the level of a sinusoid
is changed abruptly, this introduces spectral components at frequencies away from the nominal frequency, an effect called
spectral splatter. To prevent detection of the splatter, these authors presented the sinusoids in a background noise that would
mask components away from the signal frequency. The noise
also served a second purpose. When a sinusoid is changed in
level, the response of the basilar membrane changes more at
places tuned well above the signal frequency than at places
tuned close to the signal frequency. This is another manifestation
of cochlear nonlinearity; see Figure 1 and Oxenham and Moore
5
0
-5
-10
-15
-20
4
B
IB Inc
Duration of decrement (ms)
Figure 9.
Decrement detection thresholds plotted as a function of dec-
rement duration and increment (Inc) detection thresholds for the single
duration used. Durations of the decrements are specified at half-amplitude points. Each symbol shows results for a different overall level. The
signal frequency was 4 kHz.
PSYCHOACOUSTIC CONSEQUENCES OF COMPRESSION
generally markedly better than obtained when a simple powerlaw nonlinearity was assumed and the ERD and AO were allowed to be different for each level. This happened despite the
fact that the power-law analysis used five free parameters (one
value of the ERD for each level and two parameters defining
the variation of AO with level), whereas the analysis that was
based on the basilar-membrane input-output function used only
two (one value of the ERD and one value of AO). To fit the
data at 250 Hz, Moore et al. (1996a) had to assume that the
nonlinearity was somewhat less compressive than at higher center frequencies. This is consistent with physiological data
(Neely, 1993; Patuzzi & Robertson, 1988; Patuzzi et al., 1989;
Yates, 1990, 1995). However, the ERD was found to be about
7 ms, regardless of center frequency.
In summary, changes in increment and decrement detection
with level and frequency can be accounted for using a relatively
simple model in which the characteristics of the temporal integration process are invariant with level and frequency. The
model uses a form for the compressive nonlinearity that is based
on the basilar-membrane input—output function.
Temporal Integration
The absolute threshold for detecting a sound depends on the
duration of the sound. For durations up to a few hundred milliseconds, the intensity required for threshold decreases as the
duration increases. For durations exceeding about 500 ms, the
sound intensity at threshold is roughly independent of duration
(Florentine, Fasti, & Buus, 1988). Many researchers have investigated the relation between threshold and duration for tone
pulses, over a wide range of frequencies and durations. The
early work of Hughes (1946) and Gamer and Miller (1947)
indicated that over a reasonable range of durations, the ear
appears to integrate the intensity of the stimulus over time in
the detection of short-duration tone bursts. This is often called
temporal integration. In other words, the threshold corresponds
to a constant energy rather than to a constant intensity.
It seems very unlikely that the auditory system would actually
integrate stimulus energy; it is almost certainly neural activity
that is integrated (Pennei; 1972; Zwislocki, 1960). It may also
be the case that the auditory system does not actually perform
an operation analogous to integration. Rather, it may be that the
threshold intensity decreases with increasing duration partly
because a longer stimulus provides more detection opportunities
(more chances to detect the stimulus through repeated sampling). This idea is sometimes called multiple looks (Viemeister & Wakefield, 1991).
For people with cochlear damage, the change in threshold
intensity with signal duration is often, but not always, smaller
than it is for normal-hearing people. If the thresholds are plotted
on decibel versus log—duration coordinates, the slopes are usually much less in absolute value than the typical value of —3
dB per doubling of signal duration found for normal-hearing
people. This is often described as reduced temporal integration
(Carlyon, Buus, & Florentine, 1990; Chung, 1981; Elliott, 1975;
Gengel & Watson, 1971; Hall & Fernandes, 1983; Pedersen &
Elberling, 1973). There is a trend for higher absolute thresholds
to be associated with flatter slopes. In other words, the greater
the hearing loss, the more reduced is the temporal integration.
115
A number of explanations have been advanced to account for
reduced temporal integration in people with cochlear damage
(Florentine et al., 1988). One of the more plausible of these is
that it results from a reduction or complete loss of the compressive nonlinearity on the basilar membrane (Moore, 1995b). This
leads to steeper rate-versus-level functions in the auditory nerve.
According to the models of temporal integration proposed by
Zwislocki (1960) and by Penner (1972), this will automatically
lead to reduced temporal integration. Figure 10 illustrates schematically two rate-versus-level functions: The left-hand curve
shows a typical function for a low-threshold neuron in a normal
auditory system; the right-hand curve shows a typical function
for a neuron in an auditory system with cochlear damage. The
curve is shifted to the right, reflecting a loss of sensitivity, and
is steeper, reflecting loss of the compressive nonlinearity on the
basilar membrane. Note that it is assumed here that there is
some residual compression on the normal basilar membrane at
levels close to threshold; although the input-output function
steepens at low levels (as illustrated in Figure 1), it does not
become completely linear.
Assume that, to a first approximation, the threshold for detecting a sound requires a fixed number of neural spikes to be
evoked by that sound. Assume also that the neurons involved
in detection at absolute threshold are relatively homogeneous in
terms of their input-output functions. The lower dashed horizontal line in Figure 10 indicates the number of neural spikes
per second, N\, needed to achieve absolute threshold for a longduration sound; in practice of course, the absolute threshold
depends on the activity of many neurons, but if they are all
similar, then the argument can be illustrated by considering just
one. If the duration of the sound is decreased by a factoi; R,
then the level has to be increased to restore the total number of
spikes evoked by the sound. Assume that the higher spike rate
needed for the shorter duration sound is AT2, where N2 — R X
Sound level
Figure JO. Schematic illustration of rate-versus-level functions in single neurons of the auditory nerve for a normal ear (left curve) and an
impaired ear (right curve). The horizontal dashed lines indicate the
mean firing rate needed for threshold for a long-duration sound (lower
line, rate /Vl) and a short-duration sound (upper line, rate AC) on the
assumption that threshold corresponds to a fixed total number of spikes.
See text for explanation of AL.
116
MOORE AND OXENHAM
M. For example, if the duration is halved, the spike rate has to
be increased by a factor of two to achieve the same total spike
count. The increase in level, AL, needed to achieve this increased rate is greater for the normal than for the impaired ear
50 d8
95
because the rate-versus-level function is steeper for the impaired
ear, and this explains the reduced temporal integration.
If changes in basilar-membrane compression can affect tem-
90
poral integration, then temporal integration should be dependent
on level in normal-hearing participants, as basilar-membrane
compression is thought to be maximal at medium sound levels.
85
This idea was tested by Oxenham, Moore, and Vickers (1997)
by measuring thresholds for detecting a 6.5-kHz sinusoid in
the presence of a simultaneous noise masker. Consistent with
predictions, maximum integration (i.e., the greatest rate of
change of threshold with duration) was found at a medium
55
masker level. However, the effect of level was only observed
for signal durations less than 20 ms. This may mean that true
integration occurs only over fairly short durations, up to 10 or
60
20 ms. By using the model described in the previous section,
these authors found that the data could best be fitted by assuming
that the basilar membrane was about four times as compressive
at medium levels as at very high or low levels. Assuming that
55
the basilar-membrane response is almost linear at very high and
low levels, they derived a power exponent of 0.25 at medium
levels, which is in good agreement with estimates of the maximum basilar-membrane compression obtained using other techniques, as described earlier (Oxenham & Moore, 1994, 1995;
-H-
35
Oxenham & Plack, 1997).
Oxenham et al. (1997) used the model described in the previous section to fit their data for signal durations up to 10 ms.
30
The mean data (symbols), together with the model predictions
(solid lines), are shown in Figure 11. To achieve this fit, stimuli
at the lowest and highest masker levels were assumed to be
processed linearly while the stimuli at the medium masker level
25
were assumed to be compressed using a power-law exponent of
0.25 before being integrated within the sliding temporal integrator. The model fits the data rather well; all points but one lie
within one standard error of the mean data.
This approach is successful in accounting for changes in temporal integration at short durations. However, differences in temporal integration between normal-hearing and hearing-impaired
participants over longer signal durations could not be described
with the same model parameters. It is quite likely that there is
2
3 4 5 6 78910
Signal duration (ms)
Figure 11. Data from Oxenham, Moore, and Vickers (1997) showing
thresholds for detecting a 6.5-kHz sinusoid in a broadband noise as a
function of signal duration. The noise spectrum level is given in each
panel. Mean data from 4 normal-hearing participants (symbols) and
model predictions (solid lines) are compared. Error bars represent plusand-minus one standard error of the mean. SPL = sound pressure level.
some residual compression at very low and high levels in normal-hearing participants, which produces greater integration
than for hearing-impaired participants. However, this aspect has
not yet successfully been accounted for in quantitative terms.
Alcantara, Holube, and Moore (1996) examined the role of
component phase and level on the identification of vowellike
sounds. Normal-hearing participants were asked to identify
Under some circumstances it appears that the compression
which of six possible vowellike harmonic complexes was presented on each trial. The stimuli were complex tones containing
the first 35 harmonics of a 100-Hz fundamental. All of the
on the basilar membrane can have a marked effect on the perception of the timbre of complex sounds. This is especially the case
harmonics below 3000 Hz were equal in amplitude except for
three pairs of successive harmonics, at frequencies correspond-
when the waveforms of the sounds have high "peak factors"
(the peak factor is the ratio of the peak value of the waveform
ing to the first three formants of six vowels, which were incremented in level relative to the background harmonics by 1,2,
4, 8, and Ifi dB; the amount of increment was referred to as
Perception of Timbre in Vowellike Sounds
to its root-mean-square value). If the waveform remains
"peaky" after filtering on the basilar membrane, then the compression on the basilar membrane has the effect of increasing
the salience of low-level portions of the waveform.
spectral contrast. The spectra of two of the stimuli are illustrated
schematically in Figure 12.
The components in the harmonic complexes were added in
117
PSYCHOACOUSTIC CONSEQUENCES OF COMPRESSION
four different starting phase relationships: cosine, random,
Schroeder-positive, and Schroeder-negative phases. For cosine
phase, all of the components have a starting phase of 90° or
•7T/2 radians. Schroeder phase (Schroeder, 1970) was described
earlier; see Equation 2. These phases were chosen to give several
distinct temporal patterns at the outputs of the basilar-membrane
filters (i.e., the temporal pattern at a given place on the basilar
membrane would be different for each of the four phases). The
stimuli were presented at three overall levels: 85, 65, and 45
dB SPL.
The mean results (averaged across subjects and vowels) are
shown in the upper panels of Figure 13. Performance was similar
for the random and Schroeder-negative phases and did not vary
as a function of level. Performance for the cosine and Schroederpositive phase conditions was better than for the other two phase
conditions but decreased as the level was reduced. Performance
for all four phase conditions was equivalent for the lowest level.
Alcantara et al. (1996) argued that the good performance
observed for the cosine-phase stimuli and Schroeder-positivephase stimuli at high levels occurred because these stimuli produce waveforms on the basilar membrane with distinct lowamplitude portions. This is illustrated in Figure 14, which shows
outputs from a basilar-membrane model (see below for details).
The left column shows responses for "reference" cosine-phase
stimuli, which have zero spectral contrast. When formant harmonics are incremented in level, this produces changes in the
low-amplitude portions of the waveforms at points on the basilar
membrane tuned close to the formant frequencies, as illustrated
in the middle column of Figure 14.
The results were analyzed using a model with the following
stages: (a) a bank of nonlinear basilar-membrane filters with
level-dependent amplitude characteristics, a compressive inputoutput function, and a realistic phase response (Strube, 1985,
1986); (b) a threshold device (having an effect similar to rectification); (c) a square law device (as Oxenham and Moore,
40
1995, had found that the temporal integrator operates linearly
on an intensity-like quantity); (d) a sliding temporal integrator;
(e) an across-channel comparator; and (f) a decision device.
It was assumed that performance depended on the difference in
response between channels (corresponding to different places
on the basilar membrane) responding mainly to the incremented
harmonics and channels responding mainly to the background
harmonics. This difference is shown in the right column of
Figure 14. The maximum value of the difference was called the
maximum internal contrast, and it corresponds to the peak values in the right column of Figure 14. If performance depends
on the maximum internal contrast, then the results for all of the
different conditions should collapse onto a single function if
they are expressed as percentage correct versus maximum internal contrast. Figure 15 shows that this was indeed the case. The
lower panels of Figure 13 show the predictions of this model
for mean performance as a function of spectral contrast, overall
level, and phase. The predictions correspond very well with the
data in the upper panels.
Alcantara et al. (1996) found that the effects of level could
not be accounted for using a linear basilar-membrane model as
the initial stage; it was important to use a model with a realistic
nonlinear input-output function. This suggests that the compressive nonlinearity on the basilar membrane plays an important role in producing the observed phase and level effects.
Loudness Recruitment
Most, if not all, people with cochlear hearing loss show a
phenomenon referred to as loudness recruitment (Fowler, 1936;
Steinberg & Gardner, 1937). The phenomenon may be described
as follows. The absolute threshold is higher than normal. However, when a sound is increased in level above the absolute
threshold, the rate of growth of loudness level with increasing
sound level is greater than normal. When the level is sufficiently
40
(h) ee (d)
(h) er (d)
CD
TD
CD
CD
_J 20
20
CD
•t-\
-P
CD
,—i
CD
DC
100
1000
Frequency (Hz)
3500
50
100
1000
Frequency (Hz)
Figure 12. Schematic spectra of two vowellikc sounds, showing the background and formant harmonics.
The formant harmonics are indicated by dashed lines.
3500
118
MOORE AND OXENHAM
100
85 dBSPL
45 dBSPL
65 dBSPL
90
80
70
60
50
30
20
Observed scores
10
O
D
O
•*
Cosine
Random
Schroeder +ve
Schroeder -ve
O
D
O
•#
Cosine
Random
Schroeder +ve
Schroeder -ve
0
100
90
80
70
60
50
c
CD
U
C.
01
a
30
30
Predicted scores
10
0
16
1
2
4
8
16
Spectral contrast level (dB)
8
16
Figure 13. The upper panels show results of Alcantara, Hohibe, and Moore (1996). Percentage correct
scores are plotted as a function of spectral contrast level (1, 2, 4, 8, and 16 dB) for the four phase conditions
(cosine, random, Schroeder positive, and Schroeder negative). Results are shown for three different overall
levels. Error bars indicate plus-and-minus one standard error across subjects. Error bars are omitted when
they would be smaller than the symbol used. The lower panels show predictions of a model that are described
in the text. SPL = sound pressure level; +ve = positive; —ve = negative.
high, usually around 90-100 dB SPL, the loudness reaches its
"normal" value; the sound appears as loud to the person with
impaired hearing as it would to a normal-hearing person. With
further increases in sound level above 90-100 dB SPL, the
loudness grows in an almost normal manner
Two factors have been suggested to contribute to loudness
recruitment. The first is a reduction in, or toss of, the compressive nonlinearity in the input-output function of the basilar
membrane; this is usually associated with damage to the outer
hair cells. In an ear where the damage is confined largely to the
outer hair cells, with inner hair cells intact, the transformation
from basilar-membrane velocity or amplitude to neural activity
(spike rate) probably remains largely normal. However, in an
ear with damage to the inner hair cells as well, the transduction
process may also be affected.
If the input-output function on the basilar membrane is
steeper (less compressive) than normal in an ear with cochlear
damage, this would be expected to lead to an increased rate of
growth of loudness with increasing sound level. However, at
high sound levels, around 90-100 dB SPL, the input-output
function becomes almost linear in both normal and impaired
ears. The magnitude of the basilar-membrane response at high
sound levels is roughly the same in a normal and an impaired
ear (Ruggero & Rich, 1991). This could explain why the loudness in an impaired ear usually "catches up" with that in a
normal ear at sound levels around 90-100 dB SPL.
Another factor that may contribute to loudness recruitment
is reduced frequency selectivity; the filtering on the basilar membrane is less sharp in an impaired than in a normal cochlea
(Sellick, Patuzzi, & Johnstone, 1982), and this can be measured
behaviorally in humans as a broadening of the auditory filters
(Glasberg & Moore, 1986; Pick, Evans, & Wilson, 1977). For
a sinusoidal stimulus, reduced frequency selectivity leads to an
excitation pattern that is broader (spreads over a greater range
of characteristic frequencies) in an impaired than in a normal
ear. Kiang, Moxon, and Levine (1970) and Evans (1975) suggested that this might be the main factor contributing to loudness
recruitment. They suggested that, once the level of a sound
exceeds threshold, the excitation in an ear with cochlear damage
spreads more rapidly than normal across die array of neurons,
119
PSYCHOACOUSTIC CONSEQUENCES OF COMPRESSION
Smoothed Difference
85 dB SPL
0.16
65 dB SPL
0.16
45 dB SPL
0.16
0.14
Figure 14, Temporal waveforms at the output of a nonlinear basilar-membrane filter with a characteristic
frequency of around 2150 Hz for a cosine-phase reference stimulus (left column) and a test stimulus with
4-dB spectral contrast (middle column). The right-hand column shows the variation in internal contrast as
a function of time for two periods of the stimulating waveforms. The time axes are aligned with respect to
one another. Each row shows results for one level. Note that the amplitude scale in each of the rows is
linear and varies across rows. SPL - sound pressure level.
and this leads to the abnormally rapid growth of loudness with
increasing level. However, experiments using filtered noise to
mask signal-evoked excitation at characteristic frequencies remote from the signal frequency indicate that reduced frequency
selectivity is not a major contributor to loudness recruitment
(Hellman, 1978; Hellman & Meiselman, 1986; Moore, Glasberg, Hess, & Birchall, 1985; Zeng & Turner, 1991); such noise
has only a small effect on the perceived loudness of the signal.
Although spread of excitation may play a small role, the loss
of compressive nonlinearity on the basilar membrane seems to
be a more dominant factor.
This conclusion is reinforced by a model of loudness perception applied to cochlear hearing loss, as proposed by Moore and
Glasberg (1997). The model attempts to partition the overall
hearing loss between the loss due to outer hair cell damage and
the loss due to inner hair cell damage. The former is associated
with loss of compression and reduced frequency selectivity,
whereas the latter is associated simply with reduced sensitivity.
The model accounts well for measures of loudness recruitment
in people with unilateral cochlear hearing loss. It also accounts
for the fact that changes in loudness with stimulus bandwidth
(Zwicker, Flottorp, & Stevens, 1957) are smaller in hearingimpaired than in normal-hearing people.
Temporal Resolution for Sounds With Fluctuating
Envelopes in People With Cochlear Hearing Loss
For certain types of sounds, the temporal resolution of people
with cochlear hearing loss is worse than normal, even when the
stimuli are well above threshold and when all of the components
of the stimuli fall within the audible range. This happens mainly
for stimuli that contain slow, random fluctuations in amplitude,
such as narrow bands of noise. For such stimuli, people with
cochlear damage often perform more poorly than normal in tasks
120
MOORE AND OXENHAM
100
BO
70
60
50
40
30
20
C85
r85
10
S+B5 ^ s+65 •(> 3+45
9-85
O.S
1
2
© C65 O C45
S r6S
D P45
© s-65
4
B
O
3-45
16
Maximum internal contrast (dB)
Figure 15. Results of Alcantara, Holube, and Moore (1996) showing
the percentage correct for each condition (a specific combination of
spectral contrast level, phase, and overall level) plotted as a function of
the maximum internal contrast derived from the model. The phases are
indicated as follows: c = cosine; r = random; s+ = Schroeder positive;
s— = Schroeder negative.
such as gap detection (Buus & Florentine, 1985; Fitzgibbons &
Wightman, 1982; Florentine & Buus, 1984; Glasberg, Moore, &
Bacon, 1987). However, gap detection is not usually worse
than normal when the stimuli are sinusoids, which do not have
inherent amplitude fluctuations (Moore, 1997; Moore & Glasberg, 1988). It seems likely that the poor temporal resolution for
sounds with fluctuating envelopes is a consequence of a loss
of, or reduction in, the compressive nonlinearity on the basilar
membrane (Glasberg et al., 1987; Moore & Glasberg, 1988).
When compression is reduced, the inherent amplitude fluctuations in a narrowband noise would result in larger-than-normal
loudness fluctuations from moment to moment (Moore et al.,
1996b) so that inherent dips in the noise might be more confusable with the gap to be detected. It should be noted that most
everyday sounds, including speech, fluctuate in level from moment to moment, so the poor temporal resolution found for such
sounds probably leads to difficulties in everyday life.
To assess the idea that steeper input—output functions lead to
impaired gap detection for stimuli with fluctuating envelopes,
Glasberg and Moore (1992) processed the envelopes of narrow
bands of noise so as to modify the envelope fluctuations. The
envelope was processed by raising it to a power, A'. For stimuli
of constant amplitude (e.g., sinusoidal tones), the level (in decibels) of stimuli processed in this way is a linear function of the
level of the unprocessed stimuli, with slope N. This reproduces
one of the features of loudness recruitment seen in people with
unilateral hearing loss; for levels below about 90-100 dB SPL,
the level of sound needed in the normal ear to match the loudness
of a sound in the impaired ear is roughly a linear function of
the level in the impaired ear.
The effect of the processing on the envelope of a narrowband
noise is shown in Figure 16 for values of N of 2 (squared
envelope), 0.5 (square root of envelope), and 1 (unprocessed,
normal envelope). If Nis greater than unity, this has the effect
of magnifying fluctuations in the envelope, thus simulating the
effects of loss of compression or, equivalently, loudness recruitment; higher powers correspond to greater losses of compression. If N is less than unity, fluctuations in the envelope are
reduced. This represents a type of signal processing that might
be used to compensate for recruitment; it resembles the operation of a fast-acting compressor or automatic gain control
(AGC) system.
Values of N used in the experiment were 0.5, 0.66, 1.0, 1.5,
and 2. A value of N of 2 simulates the type of recruitment
typically found in cases of moderate-to-severe cochlear damage,
for which, for example, a 50-dB range of stimulus levels gives
the same range of loudness as a 100-dB range of stimulus levels
in a normal ear. To prevent the detection of spectral splatter
associated with the gap or with the envelope processing, the
stimuli were in a continuous background noise. The spectrum
of the noise was chosen so that it would be as effective as
possible in masking the splatter while minimizing its overall
loudness.
Figure 17 shows an example of results obtained using a participant with unilateral hearing loss of cochlear origin, aged 74.
The hearing loss in his impaired ear was reasonably uniform
across frequency and was 55 dB at the 1-kHz test frequency.
The stimuli were presented at 85 dB SPL, a level that was well
above the absolute threshold for both the normal and impaired
0
Squared envelope
-10
CD -20
•o
..- -30
-10
-20
Square root of envelope
-30
-10
£ -20
-30
-40
0.1
0.2
0.3
Time (sees)
Figure 16. Examples of the envelopes of noise bands, illustrating the
effect of raising the envelope to a power, N. Values of N were 1 (unprocessed, bottom), 0.5 (middle), and 2 (top). The noise bandwidth was
10 Hz. The envelope magnitudes are plotted on a decibel scale.
121
PSYCHOACOUSTIC CONSEQUENCES OF COMPRESSION
200
5
-
0.5
1.0
1.5
2.0
0.5
1.0
1.5
2.0
Power to which the envelope is raised
Figure 17.
Thresholds for detecting a gap in a noise band for which the envelope had been processed to
enhance or reduce fluctuations. Gap thresholds are plotted as a function of the power to which the envelope
was raised, with noise bandwidth (BW) as parameter; the higher the power, the greater the fluctuations.
Results are shown for each ear of a participant with unilateral cochlear hearing loss. Data are from Glasberg
and Moore (1992).
ears (although the sensation level was lower in the impaired
ear). The results for the normal ear were very similar to those
of 3 normal-hearing participants who were also tested. Gap
thresholds increased significantly with decreasing noise bandwidth. This is as expected because the inherent fluctuations in
the noise are slower, and more confusable with the gap to be
detected, when the bandwidth is narrow (Moore, 1997;
Shailer & Moore, 1983).
For all noise bandwidths, gap thresholds increased as N increased. This effect was particularly marked for the smaller
noise bandwidths. There was a significant interaction between
bandwidth and N, reflecting the fact that changes in gap threshold with N were greater for small bandwidths. This supports the
idea that fluctuations in the noise adversely affect gap detection;
greater fluctuations lead to worse performance, especially when
the fluctuations are slow.
Gap thresholds were larger for the impaired than for the normal ear. The overall geometric mean gap threshold was 12.8 ms
for the normal ear and 27.2 ms for the impaired ear. Performance
for the normal ear with N = 2 was roughly similar to performance for the impaired ear with unprocessed noise bands (N
- 1); geometric mean gap thresholds were 26.9 ms for the
former and 26.5 ms for the latter. Thus, the simulation of loss
of compression in the normal ear was sufficient to produce
impaired gap detection, comparable to that actually found in
the impaired ear.
Reduction of fluctuations, by raising the envelope to a power
less than 1, produced a small improvement in performance for
the normal ear. Values of N less than 1 gave a marked improvement for the impaired ear. Performance for the impaired ear
with the envelope raised to the power 0.5 was comparable to,
or even slightly better than, that for the normal ear with unprocessed stimuli; geometric mean thresholds were 11.6 ms for the
former and 12.5 ms for the latter. Thus, the impaired detection
of gaps in noise bands occurring for an impaired ear can be
restored to normal by appropriate compression of fluctuations
in the envelopes of the stimuli.
For both normal-hearing and hearing-impaired participants,
the effects of changing N decreased with increasing noise bandwidth. One reason for this is that slow fluctuations can be followed by the auditory system, whereas rapid fluctuations are
smoothed to some extent by the central temporal integration
process described earlier. The rapid fluctuations of the wider
noise bands are smoothed in this way, thus reducing their influence on gap detection.
The results suggest that for most people with cochlear damage, a reduction in the peripheral compressive nonlinearity may
provide a sufficient explanation for increased gap thresholds.
Thus, it is not usually necessary to assume any abnormality in
temporal processing occurring after the cochlea.
Conclusions
The basilar membrane in the normal cochlea shows a strong
compressive nonlinearity for tones close to the characteristic
122
MOORE AND OXENHAM
frequency at midrange levels. In this article we have reviewed
some of the psychoacoustic consequences of this nonlinearity,
and of the loss or reduction of the nonlinearity in people with
cochlear hearing loss.
Many of the properties of nonsimultaneous masking may be
accounted for by the compressive nonlinearity. When two
equally effective nonsimultaneous maskers are combined, the
resultant masking is usually markedly greater than would be
predicted from a simple intensity summation of the effects of
the individual maskers, an effect referred to as excess masking.
This can be accounted for by assuming that stimuli are compressed before being combined in a linear temporal integrator. In
hearing-impaired people, excess masking is reduced or absent,
probably because the compression on the basilar membrane is
absent.
The nonlinear growth of forward masking can partially be
accounted for by assuming that the masker usually falls in a
range of levels where the basilar-membrane response is strongly
compressive, whereas the signal usually falls in a lower range
where the basilar-membrane response is less compressive. The
growth of masking becomes more linear when the masker and
signal levels are nearly equal, as can be achieved by using very
brief signals, close in time to the masker. The amount of forward
masking produced by a harmonic complex masker can be altered
by manipulating the phases of the components, holding the
power spectrum constant. This can be explained by assuming
that the waveform on the basilar membrane is subjected to fastacting compression. Phases leading to peaky waveforms lead
to less average excitation and less forward masking than phases
leading to waveforms with relatively flat envelopes.
The results of tasks measuring temporal resolution, such as
the detection of brief decrements and increments, can be accounted for using a nonlinearity resembling the form of the
input-output function on the basilar membrane. The variation
in performance with level and frequency can be explained in
this way without assuming any variation with level of the central
processes involved.
The influence of component phase and level on the perception
of vowellike sounds can be accounted for using a basilar membrane model incorporating a compressive nonlinearity. More
generally, the timbre of sounds with strongly fluctuating envelopes (on the basilar membrane) is influenced by the compression, as the low-amplitude portions are made more salient.
The loss of compressive nonlinearity in people with cochlear
hearing loss may be responsible for many of the perceptual
difficulties that they encounter in everyday environments. It appears to be the main cause of loudness recruitment. It also
contributes to reduced changes in loudness with stimulus bandwidth and to reduced temporal integration. Finally, it can also
result in impaired temporal resolution for sounds with fluctuating envelopes.
loss—Literature review and experiments in rabbits. Scandinavian
Audiology, 14 (Suppl. 40), 1-147.
Buus, S., & Florentine, M. (1985). Gap detection in normal and impaired listeners: The effect of level and frequency. In A. Michelsen
(Ed.), Time resolution in auditory systems (pp. 159- 179). New "fork:
Springer-Verlag.
Carlyon, R. P., Buus, S., & Florentine, M. (1990). Temporal integration
of trains of tone pulses by normal and by cochlearly impaired listeners.
Journal of the Acoustical Society of America, 87, 260-268.
Carlyon, R. P., & Datta, A. J. ( 1997). Excitation produced by Schroederphase complexes: Evidence for fast-acting compression in the auditory
system. Journal of the Acoustical Society of America, 101, 36363647.
Chung, D. Y. (1981). Masking, temporal integration, and sensorineural
hearing loss. Journal of Speech and Hearing Research, 24, 514-520.
Cokely, C. G., & Humes, L. E. (1993). Two experiments on the temporal
boundaries for the nonlinear additivity of masking. Journal of the
Acoustical Society of America, 94, 2553-2559.
Elliott,L.L.(1975). Temporal and masking phenomena in persons with
sensorineural hearing loss. Audiology, 14, 336-353.
Evans, E. E (1975). The sharpening of frequency selectivity in the
normal and abnormal cochlea. Audiology, 14, 419-442.
Fitzgibbons, P. J., & Wightman, F. L. (1982). Gap detection in normal
and hearing-impaired listeners. Journal of the Acoustical Society of
America, 72, 761-765.
Fletchei; H. (1940). Auditory patterns. Reviews of Modem Physics, 72,
47-65.
Florentine, M., & Buus, S. (1984). Temporal gap detection in sensorineural and simulated hearing impairment. Journal of Speech and
Hearing Research, 27, 449-455.
Florentine, M., Fasti, H., & Buus, S. (1988). Temporal integration in
normal hearing, cochlear impairment, and impairment simulated by
masking. Journal of the. Acoustical Society of America, 84, 195-203.
Fowler, E. P. (1936). A method for the early detection of otosclerosis.
Archives of Otolaryngology, 24, 731—741.
Gamer, W. R., & Miller, G. A. (1947). The masked threshold of pure
tones as a function of duration. Journal of Experimental Psychology,
37, 293-303.
Gengel, R. W., & Watson, C. S. (1971). Temporal integration: 1. Clinical
implications of a laboratory study. II. Additional data from hearingimpaired subjects. Journal of Speech and Hearing Disorders, 36,
213-224.
Glasberg, B. R., & Moore, B. C. J. (1986). Auditory filter shapes in
subjects with unilateral and bilateral cochlear impairments. Journal
of the Acoustical Society of America, 79, 1020-1033.
Glasberg, B. R., & Moore, B. C. J. (1992). Effects of envelope fluctuations on gap detection. Hearing Research, 64, 81-92.
Glasberg, B. R., Moore, B. C. J., & Bacon, S. P. (1987). Gap detection
and masking in hearing-impaired and normal-hearing subjects. Journal of the Acoustical Society of America, 81, 1546-1556.
Green, D. M., & Swets, J. A. (1974). Signal detection theory and psychophysics. New York: Krieger.
Hall, J. W., & Fernandes, M. A. (1983). Temporal integration, frequency
resolution, and off-frequency listening in normal-hearing and cochlearimpaired listeners. Journal of the Acoustical Society of America, 74,
1172-1177.
References
Hawkins, J. E., Jr., & Stevens, S. S. (1950). The masking of pure tones
and of speech by white noise. Journal of the Acoustical Society of
Alcantara, J. I., Holube, I., & Moore, B. C. J. (1996). Effects of phase
America, 22, 6-13.
and level on vowel identification: Data and predictions based on a
Hellman, R. P. (1978). Dependence of loudness growth on skirts of
nonlinear basilar-membrane model. Journal of the Acoustical Society
excitation patterns. Journal of the Acoustical Society of America, 63,
of America, 100. 2382-2392.
Borg, E., Canlon, B., & Engstrom, B. (1995). Noise-induced hearing
Hellman, R. P., & Meiselman, C, H. (1986). Is high-frequency hearing
1114-1119.
123
PSYCHOACOUSTIC CONSEQUENCES OF COMPRESSION
necessary for normal loudness growth at low frequencies? 12th ICA,
normally hearing and hearing-impaired subjects.
Paper Bll-5.
Acoustical Society of America, 98, 1921-1935.
Journal of the
Hughes, J. W. (1946). The threshold of audition for short periods of
Oxenham, A.J., & Moore, B. C.J. (1997). Modeling the effects of
stimulation. Philosophical Transactions of the Royal Society of Lon-
peripheral nonlinearity in listeners with normal and impaired hearing.
don, B, 133, 486-490.
In W. Jesteadt (Ed.), Modeling sensorineural hearing loss (pp. 273-
Humes, L. E., & Jesteadt, W. (1989). Models of the additivity of masking. Journal of the Acoustical Society of America, 85, 1285-1294.
288), Hillsdale, NJ: Erlbaum.
Oxenham, A. J., Moore, B. C. J., & Vickers, D. A. (1997). Short-term
Jesteadt, W., Bacon, S. P., & Lehman, J. R. (1982). Forward masking
temporal integration: Evidence for the influence of peripheral com-
as a function of frequency, masker level, and signal delay. Journal of
pression. Journal of the Acoustical Society of America, 101, 3678-
the Acoustical Society of America, 71, 950-962.
3687.
Kiang, N. Y. S., Moxon, E. C., & Levine, R. A. (1970). Auditory nerve
Oxenham, A. J., & Plack, C. J. (1997). A behavioral measure of basilar-
activity in cats with normal and abnormal cochleas. In G. E. W. Wols-
membrane nonlinearity in listeners with normal and impaired hearing.
tenholme & J. J. Knight (Eds.), Sensorineural hearing loss (pp. 241268). London: Churchill.
Journal of the Acoustical Society of America, 101, 3666-3675.
Patterson, R. D., & Moore, B. C. J. (1986). Auditory niters and excita-
Kohlrausch, A., & Sander, A. (1995). Phase effects in masking related
tion patterns as representations of frequency resolution. In B. C. J.
to dispersion in the inner ear: IT. Masking period patterns of short
Moore (Ed.), Frequency selectivity in hearing (pp. 123-177). Lon-
targets. Journal of the Acoustical Society of America, 97, 1817-1829.
Liberman, M. C., Dodds, L. W., & Learson, D. A. (1986). Structurefunction correlation in noise-damaged ears: A light and electron-mi-
don: Academic Press.
Patuzzi, R., & Robertson, D. (1988). Tuning in the mammalian cochlea.
Physiological Reviews, 68, 1009-1082.
croscopic study. In R. J. Salvi, D. Henderson, R. P. Hamernik, & V.
Patuzzi, R. B., Yates, G. K., & Johnstone, B. (1989). Outer hair cell
Colletti (Eds.), Basic and applied aspects of noise-induced hearing
receptor current and sensorineural hearing loss. Hearing Research,
loss (pp. 163-176). New York: Plenum.
Lutfi, R. A. (1983). Additivity of simultaneous masking. Journal of the
Acoustical Society of America, 73, 262-267.
Moore, B. C. J. (1985). Additivity of simultaneous masking, revisited.
Journal of the Acoustical Society of America, 78, 488-494.
Moore, B. C. J. (1995a). Frequency analysis and masking. In B. C. J.
Moore (Ed.), Hearing (pp. 161-205). Orlando, FL: Academic Press.
Moore, B. C. J. (1995b). Perceptual consequences ofcochlear damage.
London: Oxford University Press.
Moore, B. C. J. (1997). An introduction to the psychology of hearing
(4th ed). San Diego, CA: Academic Press.
Moore, B. C. J., & Glasberg, B. R. (1983). Growth of forward masking
for sinusoidal and noise maskers as a function of signal delay: Implications for suppression in noise. Journal of the Acoustical Society of
America, 73, 1249-1259.
Moore, B. C. J., & Glasberg, B. R. (1988). Gap detection with sinusoids
and noise in normal, impaired and electrically stimulated ears. Journal
of the Acoustical Society of America, 83, 1093-1101.
Moore, B. C. J., & Glasberg, B. R. (1997). A model of loudness perception applied to cochlear hearing loss. Auditory Neumscience, 3, 289-
311.
Moore, B. C. J., Glasberg, B. R., Hess, R. F., & Birchall, J. P. (1985).
Effects of flanking noise bands on the rate of growth of loudness of
tones in normal and recruiting ears. Journal of the Acoustical Society
of America, 77, 1505-1515.
Moore, B. C. J., Glasberg, B. R., Plack, C. J., & Biswas, A. K. (1988).
The shape of the ear's temporal window. Journal of the Acoustical
Society of America, S3, 1102-1116.
Moore, B. C. J., Peters, R. W., & Glasberg, B. R. (1996a). Detection of
decrements and increments in sinusoids at high overall levels. Journal
of the Acoustical Society of America, 99, 3669-3677.
42, 47-72.
Pedersen, C. B., & Elberling, C. (1973). Temporal integration of acoustic energy in patients with presbyacusis. Acta Otolaryngologica, 75,
32-37.
Penner, M. J. (1972). Neural or energy summation inaPoisson counting
model. Journal of Mathematical Psychology, 9, 286-293.
Penner, M. J. (1980). The coding of intensity and the interaction of
forward and backward masking. Journal of the Acoustical Society of
America, 67, 608-616.
Penner, M. J., & Shiffrin, R. M. (1980). Nonlinearities in the coding of
intensity within the context of a temporal summation model. Journal
of the Acoustical Society of America, 67, 617-627.
Peters, R.W., Moore, B. C. J., & Glasberg, B. R. (1995). Effects of
level and frequency on the detection of decrements and increments in
sinusoids. Journal of the Acoustical Society of America, 97, 37913799.
Pick, G., Evans, E. F, & Wilson, J. P. (1977). Frequency resolution in
patients with hearing loss of cochlear origin. In E. F. Evans & J. P.
Wilson (Eds.), Psychophysics and physiology of hearing (pp.
273-
281). London: Academic Press.
Plack, C. J., & Moore, B. C. J. (1990). Temporal window shape as a
function of frequency and level. Journal of the Acoustical Society of
America, 87, 2178-2187.
Rhode, W. S. (1971). Observations of the vibration of the basilar membrane in squirrel monkeys using the Mossbauer technique. Journal of
the Acoustical Society of America, 49, 1218-1231.
Robles, L., Ruggero, M. A., & Rich, N. C. (1986). Basilar membrane
mechanics at the base of the chinchilla cochlea: I. Input—output functions, tuning curves, and response phases. Journal of the Acoustical
Society of America, SO, 1364-1374.
Ruggero, M. A. (1992). Responses to sound of the basilar membrane
Moore, B. C. J., Wojtczak, M., & Vickers, D. A. (1996b). Effect of
of the mammalian cochlea. Current Opinion in Neurobiology, 2, 449456.
loudness recruitment on the perception of amplitude modulation.
Ruggero, M. A., & Rich, N. C. (1991). Furosemide alters organ of Corti
Journal of the Acoustical Society of America, 100, 481-489.
Neely, S. T. (1993). A model ofcochlear mechanics with outer hair cell
motility. Journal of the Acoustical Society of America, 94, 137-146.
mechanics: Evidence for feedback of outer hair cells upon the basilar
membrane. Journal of Neuroscience, 11, 1057-1067.
Nelson, D. A., & Schroder, A. C. (1996). Release from upward spread
Ruggero, M. A., Rich, N. C., Robles, J., & Recio, A. (1996). The effects
of acoustic trauma, other cochlea injury and death on basilar mem-
of masking in regions of high-frequency hearing loss. Journal of the
brane responses to sound. In A. Axelsson, H. Borchgrevink, R. P.
Acoustical Society of America, 100, 2266-2277.
Hamernik, P. A. Hellstrom, D. Henderson, & R. J. Salvi (Eds.), Scien-
Oxenham, A. J., & Moore, B. C. J. (1994). Modeling the additivity of
nonsimultaneous masking. Hearing Research, 80, 105-118.
Oxenham, A. J., & Moore, B. C. J. (1995). Additivity of masking in
tific basis of noise-induced hearing loss (pp. 23-35). Stockholm:
Thieme.
Schroeder, M. R. (1970). Synthesis of low peak-factor signals and bi-
124
MOORE AND OXENHAM
nary sequences with low autocorrelation. IEEE Transactions on Information Theory, IT-16, 85-89.
Sellick, P.M., Patuzzi, R., & Johnstone, B. M. (1982). Measurement
of hasilar membrane motion in the guinea pig using the Mossbauer
technique. Journal of the Acoustical Society of America. 72, 131
141.
Shailer, M. J,, & Moore, E. C. J. (1983). Gap detection as a function
of frequency, bandwidth and level. Journal of the Acoustical Society
of America, 74, 467-473.
Steinberg, J. C., & Gardner, M. B. (1937). The dependency of hearing
impairment on sound intensity. Journal of the Acoustical Society of
America, 9. 11-23.
Strube, H. W. (1985). A computationally efficient basilar-inembrane
model. Acusfica, 58, 207-214.
Strube, H, W. (1986). The shape of the nonlinearity generating the
combination tone 2f,-f 2 . Journal of the Acoustical Society of
America, 79, 1511-1518.
Viemeister, N. E, & Wateneld, G. H. (1991). Temporal integration and
multiple looks. Journal of the Acoustical Society of America, 90,
858-865.
R* POSTAL SERVICE,*
(Required by 39 USC 3S85)
2. Publication Hunger
Psychological Review
A
4. Issue Frequency
8
8
3. Filing Date
8 0
0
4
Street
HE,
Washington,
DC
5180/InsL
Tdepfwne
(202)
First Street HE,
Washington,
DC
6, Paid anchor
Requested
Circulation
Actual No. Coptos of Single Issue
Avenge No. Copiel E*ch toiu*
During Preceding 1! Months
r> i. nt ..ml Nitura of Circulation
B. Total Number ol Copies (Net press run)
8,378
8,555
( 1 ) Sales Through Daalors and Carriers. Streol Vendors,
and Counter Sales (Not mailed)
(2) PaW of Requested Mail Subscriptions (tnduOe
-
6,218
5,634
6,218
5,634
147
178
336-5579
8. Complete Moling Address of Headquarters or Genera) Business Office of Publisher (Not primer)
750
July 1997
Review
October 1997
Contact Person
J. Rrodie
20002-4242
14, Issue Dalo for Circulation Data Below
Psychologica 1
IS.
S. Number ollssua s Published Annually
Quarterly
7. Complete Mailing Address ol Known Office ol Publication /Wo/ prinferjrsfreel, dry, county, state, an iZIP+4)
First
Received December 11, 1996
Revision received March 5, 1997
Accepted March 18, 1997 •
13. Publication Trtla
"'"~'"°
1. Publication Title
750
Wilson, R. H., & Carhart, R. (1971). Forward and backward masking:
Interactions and additivity. Journal of the Acoustical Society of
America, 49, 1254-1263.
Yates, O.K. (1990). Basilar membrane nonlinearity and its influence
on auditory nerve rate-intensity functions. Hearing Research, 50,
145^162.
Yates, G. K. (1995). Cochlear structure and function. In B. C. J. Moore
(Ed.), Hearing (pp. 41-73). San Diego, CA: Academic Press.
Zeng, E G., & Turner, C. W. (1991). Binaural loudness matches in unilaterally impaired listeners. Quarterly Journal of Experimental Psychology, 43A, 565-583.
Zwicker, E.. Flottorp, G., & Stevens, S.S. (1957). Critical bandwidth
in loudness summation. Journal of the Acoustical Society of America,
29, 548-557.
Zwislocki, J. J. (1960). Theory of temporal auditory summation. Journal of the Acoustical Society of America, 32, 1046-1060.
c. Total Paid and/of Requested Circulation
(Sum ol I5b(J) and JSbff))
20002-4242
<J. Free Distribution by Mail
(Samples, dunplmcnOry, and oilier freo)
0. Futl Names and Complete Malinp Addresses ol Publisher. Editor, and Managing Editor (Do no! le&vt Olar*}
Publisher (Name andeorrfhta mailing oddross)
.
o. FIDO Distribution Outside the Mail (Carriers or other means)
Anwrican Psychological Association, 750 First Street, BE, Washington, DC 20002-4242
i. Total Free Distribution (Sum ol tSd and tie)
^
147
178
6,365
5,812
2,190
2,566
Editor (Hamo andcomp/oto moiling address)
Robert A. Bjork, PI1.D., Dept of Psychology, Uni405 Hilgard Avenue, I,os Angeles, Ca 90024
;ity of California,
g. Total Distribution (Sum ol ISc and 151}
Manaoino Editor (Namo and complata mallina eddross)
^
mOffico Use. Leftovers. Spoiled
"SXS
Susan Knapp, American Psychological Association,
750 First Street, HE, Washington, DC 20002-4242
(2} Returns Irom News Agents
1 0. Owner (Do notfedvo blank, if the pyMcatton Is o v/ned by a corporation, give ths narr& art addiass ot tto corporal^ Irrnnoaiately lotlowed by the
names end addresses ol aH stocttniders owning or totting 1 poicent of more ol the total anwnfol stock. 11 not ovtrM by a corporation, give too
names and addresses ol the Individual owners. II owned by a partnership or other unirKXtrporated firrr^gwe Its rorno and address as wot! as those c(
each Individual owner. It the pubticaVon Is published by a nonprofit organization, give Its lame and address.)
1, Total {Sum ol 15g. ISh(l), and f St\(2)}
Full Name
Percent Paid and/c Requested Circulation
ftSc/ JSg x 100)
Complete Mailing AddtM*
American Psychological Association
750 First Street NE
Washington, DC
20002-4242
_
^
16. PuMcaton of Statement 01 uwnersmp
laniiarv 1
D Publication required. Will be prlntM in the January _L_
O Publication nol required.
Bfl cJ Editor. PuWishar. Business
Bus
Manager, or Owner
<&>f*ii
\-^' •—'•>•''
£&£
£-?j?_'-Yj£_
8,555
8,378
97.7
97.0
_ Issoa of this publication.
_/^//
J> S/CCC
7^>~
1^'Ci/7g'>'~
I
/ 7X / /
I certify thai aUlnlonraSontumlshed on IhcfefVls true and corr^tota. I umJer«^^tanyorw who lurrJshflslatee or rrWeadir^ Information on this form
w who omits rrutarbl cr Inhxmation reo^jeti^ ofi №• lorm may b« tub^ lo crlrri
(including rnuftlple damages and cfvil penalties).
11, Known Bondholders, Mortgagees, and Other Socurity Holders Ownftg or
HokSng 1 Percent or More ol Total Amount of Bonds, Mortgages, o/
RJHNjmt
Instructions to Publishers
Comploto Mailing Addrems
NONE
\. Complole and lile one copy of ihis form with your postmaster annually on or before October 1. Keep a copy of Ihe completed form (or
your records.
2. In cases where Ihe stockholder or secunty holder is a trustee, include in ilems 10 and 11 the nameol Ihe person or corporation for
whom Iho Irustee is acting. Also include the names and addresses ol Individuals who are slockholders who own or hold 1 percent or
more ol the lolal amount ol bonds, mortgages, or other securities of the publishing corporation. In ilem 11, it none, check the box. Use
3. Be sure lo iumish all circulation inlormaiion called lor in ilem 15. Free circulation must be shown in items 15d, e. and I.
4. 11 Ihe publication had second-class authorization as a general or requester publication, Ihis Statement of Ownership, Managomenl, and
Citcutalion musl be published; it musi be primed in any issue in October or. if Ihe publication is not published during October, the first
issue printed after October.
5. In item 16, indicate (he dale of ihe issue in which ihis Slaiement of Ownership will be published.
12. Tax Status (Fwcorr&otton try nonprofit oig^mt^autfK&rt to rr^ at i^^
The purpose, function, and nonprofit etatus of this organization and tho oxompt status tor federal Income tax purposes:
Q Has Not Changed During Preceding 1 z Months
Q Has Changed During Preceding 12 Months (Publisher must subml explanation of change vrith this statement)
PS Form 3526, September 1
fSoe /njiructfons on Raveisa)
6. Ilem 17 musl be signed.
Failure to file or publisn a statement ol ownership may lead lo suspension of second-class authorization.