Listener weighting of cues for lateral angle: the duplex theory of sound localization revisited.

Abstract

The virtual auditory space technique was used to quantify the relative strengths of interaural time difference (ITD), interaural level difference (ILD), and spectral cues in determining the perceived lateral angle of wideband, low-pass, and high-pass noise bursts. Listeners reported the apparent locations of virtual targets that were presented over headphones and filtered with listeners' own directional transfer functions. The stimuli were manipulated by delaying or attenuating the signal to one ear (by up to 600 micros or 20 dB) or by altering the spectral cues at one or both ears. Listener weighting of the manipulated cues was determined by examining the resulting localization response biases. In accordance with the Duplex Theory defined for pure-tones, listeners gave high weight to ITD and low weight to ILD for low-pass stimuli, and high weight to ILD for high-pass stimuli. Most (but not all) listeners gave low weight to ITD for high-pass stimuli. This weight could be increased by amplitude-modulating the stimuli or reduced by lengthening stimulus onsets. For wideband stimuli, the ITD weight was greater than or equal to that given to ILD. Manipulations of monaural spectral cues and the interaural level spectrum had little influence on lateral angle judgements.

Full text

Page 1

Listener weighting of cues for lateral angle:
The duplex theory of sound localization revisiteda)
Ewan A. Macphersonb)and John C. Middlebrooks
Kresge Hearing Research Institute, University of Michigan, 1301 East Ann Street, Ann Arbor,
Michigan 48109-0506
?Received 24 September 2001; accepted for publication 26 February 2002?
The virtual auditory space technique was used to quantify the relative strengths of interaural time
difference ?ITD?, interaural level difference ?ILD?, and spectral cues in determining the perceived
lateral angle of wideband, low-pass, and high-pass noise bursts. Listeners reported the apparent
locations of virtual targets that were presented over headphones and filtered with listeners’ own
directional transfer functions. The stimuli were manipulated by delaying or attenuating the signal to
one ear ?by up to 600 ?s or 20 dB? or by altering the spectral cues at one or both ears. Listener
weighting of the manipulated cues was determined by examining the resulting localization response
biases. In accordance with the Duplex Theory defined for pure-tones, listeners gave high weight to
ITD and low weight to ILD for low-pass stimuli, and high weight to ILD for high-pass stimuli. Most
?but not all? listeners gave low weight to ITD for high-pass stimuli. This weight could be increased
by amplitude-modulating the stimuli or reduced by lengthening stimulus onsets. For wideband
stimuli, the ITD weight was greater than or equal to that given to ILD. Manipulations of monaural
spectral cues and the interaural level spectrum had little influence on lateral angle
judgements. © 2002 Acoustical Society of America. ?DOI: 10.1121/1.1471898?
PACS numbers: 43.66.Qp ?LRB?
I. INTRODUCTION
The locations of sound sources are not mapped directly
at the sensory periphery. Instead, locations must be derived
by combining acoustical cues that result from the interaction
of incident sound waves with the external ears, head, and
upper body. The acoustical cues for sound localization were
explored as early as the end of the 19th century by the Brit-
ish physicist Lord Rayleigh ?Strutt, 1907?, among others.
Rayleigh worked primarily with pure-tone stimuli produced
by vibrating tuning forks. He determined that the primary
cue to the lateral positions of sources with frequencies ?500
Hz was the interaural difference in sound pressure levels
?ILDs? resulting from acoustic shadowing by the head. At
lower frequencies, however, the wavelength of sound is
much larger than the diameter of the head, and ILDs are
negligible. Through the ingenious use of a pair of mistuned
low-frequency tuning forks, Rayleigh demonstrated compel-
lingly that human listeners are sensitive to interaural differ-
ences in the ongoing phase of low-frequency sounds and,
thus, that interaural time differences ?ITDs? could provide
cues to the lateral positions of low-frequency sources. Ray-
leigh’s understanding of the localization of tones in the lat-
eral dimension has come to be known as the ‘‘Duplex
Theory’’ of sound localization and has been substantiated in
numerous psychophysical and physiological studies. Ray-
leigh also appreciated that subjects could not discriminate
the front-versus-back locations of pure-tone stimuli, but that
such front/back discrimination was possible for ‘‘sounds of
other character’’ ?i.e., sounds with broader bandwidths?. Re-
search in the past few decades has filled in some understand-
ing of the cues for front/back and vertical localization, re-
vealing the importance of spectral-shape cues provided by
the direction-dependent filtering of broadband sounds by the
external ears.
The Rayleigh Duplex Theory is quite satisfactory to ex-
plain left/right localization of tonal stimuli. Nevertheless, in
the real world most sounds have bandwidths of several oc-
taves, and a listener rarely is exposed to a pure tone. Three
sets of observations led us to revisit the Duplex Theory in the
context of complex ?i.e., broadband? sounds.
First, lateralization studies have shown than listeners are
sensitive to ITDs in high-frequency complex sounds. Sensi-
tivity to ongoing time differences in simple sounds such as
pure tones is limited to low frequencies by the loss of phase-
locking in the auditory nerve at high frequencies and likely
also by a lower-frequency cutoff in the binaural system. In-
deed, psychophysical studies show that sensitivity to ongo-
ing ITD is limited to frequencies below ?1.3 kHz for pure
tones ?Zwislocki and Feldman, 1956?. Nevertheless, the au-
ditory system can extract timing information from the enve-
lopes of higher-frequency sounds that contain multiple fre-
quency components, and listeners can detect ITDs in high-
frequency complex sounds presented through headphones
?e.g., Henning, 1974; Leakey et al., 1958; McFadden and
Pasanen, 1976?.
Second, Wightman and Kistler ?1972, 1997b? have in-
vestigated localization cues using virtual auditory space
?VAS? techniques, which permit more-or-less independent
manipulation of ITD, ILD, and spectral-shape cues. They
demonstrated that ITDs dominate listeners’ judgments of the
location of broadband sound sources that contain low-
a?Portions of this work were previously presented in poster form at the 24th
Annual Midwinter Research Meeting of the Association for Research in
Otolaryngology in February 2001.
b?Electronic mail: emacpher@umich.edu
2219J. Acoust. Soc. Am. 111 (5), Pt. 1, May 20020001-4966/2002/111(5)/2219/18/$19.00© 2002 Acoustical Society of America

Page 2

frequency components ?Wightman and Kistler, 1992?. In
some cases, the influence of ITDs persisted even when
stimulus spectra were limited to high frequencies, although
the effect of high-pass filtering varied widely among listen-
ers. Also, imposition of an interaural imbalance ?i.e., an ILD
of 10–20 dB? had surprisingly little impact on lateral loca-
tion judgments by some listeners ?Wightman and Kistler,
1997b?.
Third, spectral-shape cues, which provide essential cues
to the vertical and front/back location of broadband sounds,
also vary with source azimuth and might contribute to the
judgment of lateral localization. For instance, some congeni-
tally monaural listeners, whose only cues to sound-source
location derive from direction-dependent filtering by one ex-
ternal ear, show reasonably accurate localization in the hori-
zontaldimension
?Slattery
Whether normal binaural listeners similarly use spectral cues
in determining lateral angle is unknown.
These three sets of observations raise questions about
the applicability of the Rayleigh duplex model to localization
of naturally occurring broadband sounds. Might ITDs con-
tribute to or even dominate localization of high-frequency
complex sounds? How salient are spectral-shape cues com-
pared to interaural difference ?i.e., ITD and ILD? cues? Al-
though ILDs have an obvious impact on lateralization of
high-frequency tones, do ILDs have any influence on local-
ization of broadband sounds?
We addressed these questions by measuring localization
of broadband, low-pass, and high-pass sounds presented us-
ing VAS techniques. We quantified the degrees to which cues
addressed by the Duplex Theory ?such as low-frequency ITD
and high-frequency ILD? and other cues ?such as spectral
cues and high-frequency, envelope-based ITD? contributed to
the lateral component of listeners’ localization judgments.
Apart from some influence of envelope-based ITDs in high-
pass signals, we found that the localization of complex
sounds in the lateral dimension accorded nicely with a more-
than-century-old theory based on the localization of vibrating
tuning forks.
andMiddlebrooks,1994?.
II. METHODS
A. Subjects
Thirteen paid listeners ?five female and eight male, ages
18–35 years, including the first author? were recruited from
the student body of the University of Michigan and the staff
of the Kresge Hearing Research Institute. All had extensive
experience in free-field localization experiments. Only one
?the first author, S18? had previous experience in localizing
virtual auditory stimuli presented over headphones, but the
others received several hours of practice in the virtual free-
field localization task described in Sec. IID prior to the be-
ginning of the present study. These practice sessions in-
volved the localization of virtual broadband noise-burst
targets. No feedback was provided. All listeners had normal
hearing as defined by standard audiometric testing. Ten of
the listeners ?not including S18? participated in Experiments
I and II-A. Two of these ten plus S18 and two additional
listeners participated in Experiment II-B. Three of the ten
plus S18 participated in Experiment III, and three from Ex-
periment II-B participated in Experiment IV.
B. Directional transfer function measurements
In order to compute the filters necessary to synthesize
the virtual free-field stimuli, measurements were made of
each listener’s directional transfer functions ?DTFs?. The de-
tails of this procedure are given by Middlebrooks ?1999a?.
Briefly, 512-point, 50-kHz Golay codes ?Zhou et al., 1992?
were presented from a loudspeaker positioned 1.2 m from
the listener’s head at 400 locations approximately evenly dis-
tributed in space around the listener’s head. The responses to
these excitation signals were recorded simultaneously by two
miniature electret microphones ?Knowles, model 1934? in-
serted approximately 5 mm into the listeners’ ear canals.
Head-related transfer functions ?HRTFs? were extracted by
cross-correlation of excitation and response, transformation
to the frequency domain, and division by the previously
measured loudspeaker transfer function. For each ear, DTFs
were computed from the HRTFs by dividing by a complex
common component computed for the set of HRTFs for that
ear ?Middlebrooks, 1999a?. This nondirectional component
was a combination of the ear canal resonance and the
diffuse-field average response at the ear canal entrance. The
DTFs were transformed into the time domain, yielding a set
of directional impulse responses ?DIRs? for each listener.
The DIRs were used in the synthesis of the virtual free-field
targets as described below.
C. Stimulus synthesis
Stimulus noise waveforms were computed on an Intel-
based desktop personal computer using custom MATLAB
scripts ?The Mathworks?. An inverse-Fourier-transform
method was used to produce flat-spectrum, random-phase
noise waveforms with the desired passband and duration
sampled at 50 kHz. Raised-cosine ?i.e., cos2? ramps of 1, 20,
or 50 ms duration ?depending on the stimulus condition?
were applied to the onsets and offsets. The resulting wave-
form was convolved with the right- and left-ear DIRs corre-
sponding to the desired target location. The stimuli were pre-
sented over ‘‘diffuse-field
headphones ?Sennheiser HD 265? at a sampling rate of 50
kHz using digital-to-analog converters, attenuators, and
headphone amplifiersfrom Tucker-Davis Technologies
?models DD1, PA4, and HB6, respectively?. The stimulus
level on each trial was equivalent to that of a free-field
source at a sound pressure level of approximately 65 dB.
We did not attempt a rigorous equalization of the head-
phone response. Rather, the headphone response itself re-
stored an approximation of the diffuse-field component re-
moved in the computation of the DIRs, and the listener’s
own ear canal restored the ear canal resonance. We have
discussed this approach previously and have shown that lis-
teners can localize accurately in the VAS generated by this
method ?Middlebrooks, 1999b?.
In Experiments I and II, the interaural time difference
?ITD? and interaural level difference ?ILD? cues naturally
equalized’’circumaural
2220J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002 E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 3

present in the stimulus were manipulated by imposing a
whole-waveform delay or attenuation on the signal at one of
the ears. We refer to this procedure as biasing the stimulus
and to the amount of ITD or ILD offset as the imposed bias.
In Experiments III and IV, the stimuli were manipulated by
modifying the DTF spectrum at one or both ears in order to
bias the interaural level spectrum ?ILS? or the DTF of the ear
nearest the source. The cue manipulations for Experiments
I–IV are summarized in Table I.
D. Localization procedure
The localization procedure was similar to that described
by Middlebrooks ?1999b?. Listeners stood in the center of a
darkened anechoic chamber, and at the beginning of each
trial oriented towards a light-emitting diode ?the centering
LED? positioned at eye level 2 m directly in front of the
listener. A trial was initiated by pressing a hand-held button.
The centering LED was extinguished, and the listener’s ini-
tial head position was measured by a head-mounted electro-
magnetic tracking device ?Polhemus FASTRAK?. Following a
delay of 500 ms, the stimulus was presented over head-
phones. After hearing the virtual free-field stimulus, the lis-
tener oriented towards its perceived location, at which time a
second button press triggered a second measurement of head
orientation. If the initial head orientation deviated by more
than 10 degrees in either azimuth or elevation from the cen-
tering LED, the trial data were discarded. Depending upon
the condition, runs consisted of 79, 99, or 111 trials, and the
longest were typically completed in 10–12 min. Listeners
rested after every two to three runs, and typically completed
six to eight runs in one session.
E. Data analysis
1. Lateral-polar coordinate system
Target and response location data were converted from
the vertical polar ?azimuth and elevation? coordinate system
to the horizontal polar ?lateral- and polar-angle? coordinate
system ?Fig. 1?. Only the lateral angle data were analyzed in
detail. Sample lateral and polar angle response data for wide-
band, ITD-biased stimuli are shown in Fig. 2. In this ex-
ample, the ITD was biased by ?600 ?s ?top row?, 0 ?s
?middle row?, or ?600 ?s ?bottom row?. Note that for these
stimuli, positive ?right-leading? and negative ?left-leading?
ITD biases shifted the listener’s responses consistently to-
wards the right and left sides, respectively. For this listener,
applying ITD bias increased the rate of front/back confusions
?more responses fell in the upper-left or lower-right quad-
FIG. 1. Horizontal polar coordinate system. Lateral angle is the angle be-
tween the location and the median sagittal plane; positive values are to the
listener’s right. Polar angle combines elevation and front/back position. ?90
degrees: below; 0 degrees: front; ?90 degrees: overhead; ?180 degrees:
rear.
FIG. 2. Sample lateral- ?left column? and polar-angle ?right column? re-
sponse data for ITD-biased wideband targets for listener S66. ITD bias:
?600 ?s ?top?, 0 ?s ?middle?, ?600 ?s ?bottom?. Unbiased lateral angle
gain, a, was computed as the slope of a linear fit to the unbiased stimulus
lateral-angle responses ?solid line?. In the polar angle plots, accurate re-
sponses fell near the positive diagonal, and front/back reversed responses
fell near the negative diagonal.
TABLE I. Summary of cue manipulations for Experiments I–IV. The cues
referred to are interaural time difference or interaural phase spectrum ?ITD/
IPS?, interaural level difference or interaural level spectrum ?ILD/ILS?, and
the directional transfer functions for the ears nearer to and farther from the
source ?DTFnearand DTFfar?. Symbols indicate whether the cue corre-
sponded to the original target location ???, to a biased location ???, or ?in
the case of the artificial DTFs used in Experiments III and IV? to no actual
location ???.
Cue Experiments I and II
Experiment III
ILS bias
Experiment IV
DTFnear, bias
ITD/IPS
ILD/ILS
DTFnear
DTFfar
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
2221 J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002 E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 4

rants?, but the elevation component of the listener’s polar-
angle judgments remained reasonably accurate ?i.e., re-
sponses fell near the positive or negative diagonals?. The
effect of bias on polar-angle responses is discussed further in
Sec. IIIB2.
2. Unbiased lateral angle gain
In order to reveal any detrimental effect on lateral angle
localization caused by the restriction of stimulus bandwidth,
an unbiased lateral angle gain ?Hofman and Van Opstal,
1998; Macpherson and Middlebrooks, 2000? was computed
for responses to the filtered, but unmanipulated ?unbiased?,
targets in each stimulus set. The unbiased lateral angle gain
was simply the slope of a linear fit to the target and response
lateral angle data ?Fig. 2?.
3. ITD and ILD bias weights
We wished to assess the weighting or salience of the
manipulated cue ?ITD or ILD? in a manner which would
permit meaningful comparisons between weights. Neither
the classical time-intensity trading ratio ?in units of ?s/dB?
nor ratios between angular displacements of judgments and
magnitudes of imposed cue bias ?yielding values in
degrees/?s and degrees/dB? are useful in isolation in describ-
ing the relative effectiveness of ITD and ILD in localization.
To use such values to make comparisons between the
weighting of different cues, reference must also be made
both to the naturally occurring relations between the physical
cues and to the spatial disposition of the cues.
Our strategy was to derive dimensionless weights relat-
ing bias in responses to imposed bias in the underlying cue
by ?a? measuring the correspondence between the physical
cues and lateral angle, and ?b? using this relation to convert
any shift in the angular response to a quantity expressed in
the units of the manipulated cue itself.
As an example, consider a target location for which the
natural ILD is 5 dB. If an ILD bias of 10 dB is imposed, the
ILD in the presented stimulus will be 15 dB. If the listener
responds at a location for which the natural ILD were also 5
dB ?even if that were at a different lateral angle?, we would
conclude that the imposed bias had no effect, the observed
bias would be 0 dB, and we would estimate the perceptual
weighting of ILD to be close to 0. Conversely, if the listener
responds at a location for which the natural ILD were 15 dB,
we would conclude that the imposed bias was fully effective
in shifting the perceived lateral angle, the observed bias
would be 10 dB ?equal to the imposed bias?, and we would
estimate the ILD bias weight to be close to 1. Responses at
locations with intermediate ILDs would yield intermediate
weights.
This conversion of bias in response location to bias in an
underlying physical cue and the resulting derivation of a di-
mensionless weight permits the computation and comparison
of perceptual weights on the ITD and ILD cues without con-
cern about the physical correspondence of, or the auditory
system’s differential sensitivity to, the ITD and ILD cues. In
addition, by incorporating the relation between cue and lat-
eral angle into the procedure, the exact ?and possibly non-
monotonic? relation between the interaural cues and lateral
angle was rendered unimportant.
The following procedure was used to compute the bias
weights. First, the value of the physical interaural cue at each
of the 400 measured DTF locations was computed for each
listener using procedures similar to those of Gaik ?1993?. For
ILD, the energies in the right and left ear directional transfer
functions were integrated over the stimulus passband ?wide-
band, 0.5–16 kHz; low-pass, 0.5–2 kHz; high-pass, 4–16
kHz; see Secs. IIC and IIIA? and their ratio represented in
dB such that positive ILDs corresponded to higher intensity
at the right ear.
To compute ITD, the DIR for each ear was passed
through a gammatone filter bank ?Slaney, 1994? with low-
frequency channels at 600, 700, and 800 Hz, and high-
frequency channels at 4, 4.5, and 5 kHz. These center fre-
quencies were chosen because they produced the smoothest
ITD-versus-azimuth functions across listeners. For the low-
frequency channels, the ITD was taken from the lag of the
peak in the cross-correlation of the left and right ear signals.
For the high-frequency channels, the envelopes of the filter
outputs were extracted using a Hilbert transform prior to
cross-correlation in order to extract the group delay. This
paralleled the loss of phase-locking and the onset of enve-
lope following in the auditory nerve at high frequencies. If
multiple peaks appeared in the cross-correlation, the one
closest to the predicted ITD based on a spherical head model
?Kuhn, 1987? was chosen. If no peak was found within 250
?s of the predicted value, the mean of the computed ITDs
for neighboring locations was used. The median of the ITDs
in the low-frequency channels was used as the ITD cue for
low-pass and wideband stimuli, and the median of the high-
frequency channel ITDs was used as cue for high-pass
stimuli. Note that this process was intended as a means of
measuring the physical ITD, not an attempt to model the
extraction of ITD information by the auditory system.
Next, having associated a frequency-dependent ITD and
ILD with each DTF location, an observed cue bias ?ITD or
ILD? was computed for each localization response in the
manner illustrated in Fig. 3. The natural cue, cnat?in ?s or
dB?, was that present in the unmodified DTFs used in syn-
thesizing the spatialized stimulus. The observed cue, cobs,
was that associated with the DTF location closest to the lis-
tener’s response location. The observed cue bias was the dif-
ference between these values; bias?cobs?cnat. Trials in
which the manipulated cue exceeded the range of the listen-
er’s measured ITD or ILD values were eliminated from the
analysis. Overall, 10%–15% of trials were discarded for this
reason.
Finally, the listener’s weighting of the manipulated in-
teraural cue was computed as the slope of the linear regres-
sion between the observed cue bias and the imposed cue bias
?the magnitude of the added ITD, in ?s, or ILD, in dB?, as
illustrated in Fig. 4. We refer to these dimensionless values
as the ?s/?s and dB/dB weights. The standard error of the
regression coefficient was taken as a measure of the uncer-
tainty in the computed weight. If the manipulated cue had
little influence on the listener’s response, the response loca-
2222J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002 E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 5

tion was close to the original target location, the observed
cue bias was close to zero in all trials, and the cue weight,
WITDor WILD, was also close to zero ?Fig. 4, upper-right and
lower-left panels?. Conversely, if the listener derived the
judgment of lateral angle primarily from the manipulated
cue, then the response was expected to lie at a location for
which the natural cue was similar to the stimulus cue value.
In such cases, the cue weight was close to 1 ?Fig. 4, upper-
left and lower-right panels?.
III. EXPERIMENT I: WEIGHTING OF ITD AND ILD
CUES IN LOW-PASS, HIGH-PASS, AND WIDEBAND
NOISE
A. Stimuli and locations
In Experiment I, we measured listeners’weighting of the
ITD and ILD cues to lateral angle under three passband con-
ditions: wideband, 0.5–16 kHz; low-pass 0.5–2 kHz; and
high-pass, 4–16 kHz. The target stimuli were 100-ms noise
bursts with 1-ms raised-cosine onsets and offsets. Each
stimulus set contained interleaved stimuli from four classes:
?1? unfiltered ?i.e., wideband? noise bursts, ?2? filtered, un-
manipulated ?i.e., no imposed ITD or ILD bias? noise bursts,
?3? filtered noise bursts with medium imposed cue bias
??300 ?s ITD or ?10 dB ILD?, and ?4? filtered noise bursts
with large imposed cue bias ??600 ?s ITD or ?20 dB ILD?.
Some listeners also completed an additional set of ILD-bias
conditions in which the medium and large biases of 10 and
20 dB were replaced by 4- and 8-dB biases, respectively.
Analysis of these data showed that the computed weights
were insensitive to which range of ILD biases was used, and
the 4- and 8-dB ILD-bias data were not included in the fol-
lowing analysis.
Unfiltered targets, 36 in all, were placed at 10-degree
increments in azimuth from ?170 to ?180 degrees with
elevations of ?30 or ?30 degrees. Of the filtered, unbiased
targets, 24 were placed at azimuths 0, ?20, ?160, and 180
degrees with elevations of ?20 and ?40 degrees ?i.e., on or
near the median plane, both in the front and rear, and above
and below the horizontal plane?. An additional 18 were
FIG. 3. Computation of observed cue bias. Small symbols show the mea-
sured low-frequency ITD for listener S92 as a function of lateral angle. Each
small symbol represents a distinct DTF measurement location ?unique lat-
eral angle and polar angle combination?. The large filled circle indicates the
lateral angle and natural ITD of an example original target location. An ITD
bias of ?600 ?s was applied. The large open circle indicates the lateral
angle and natural ITD of the DTF measurement location closest to the
listener’s response ?i.e., the observed ITD?. The observed cue bias is the
signed difference between the observed ITD and the natural target ITD.
FIG. 4. Illustration of cue bias weight computation. Ob-
served ITD and ILD cue biases are plotted against im-
posed cue bias for listener S77 in the low-pass ?0.5–2
kHz? and high-pass ?4–16 kHz? passband conditions.
Upper-left:low-passITD-bias
?300, ?600 ?s?; upper-right: low-pass ILD bias con-
dition ?bias?0, ?10, ?20 dB?; lower-left: high-pass
ITD-bias condition; lower-right: high-pass ILD bias
condition. The cue bias weight was the slope of a linear
fit to these data. The ?30 degrees scale near the vertical
axis indicates the natural variation of the manipulated
cue within 30 degrees of the midline.
condition
?bias?0,
2223J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002 E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 6

placed in 20-degree increments of azimuth around the hori-
zontal plane, for a total of 42 locations. To minimize the
presentation of stimuli with ITD or ILD cues outside of the
physiological range, the medium bias was not applied to tar-
gets more than 40 degrees away from the median plane on
the side toward which the bias was applied, leaving 38 loca-
tions. Similarly, the large bias was not applied to targets
more than 20 degrees from the median plane on the biased
side, leaving 34 locations. In total there were six stimulus
sets ?3 passbands?2 cue-bias types? of 222 targets each. The
targets in each set were presented twice in a shuffled order
over the course of four blocks of 111 trials each. Blocks from
different stimulus sets were intermixed within an experimen-
tal session but trials from different sets were not intermixed
within blocks.
B. Results
1. Lateral angle responses
For each listener, the unbiased lateral angle gain was
computed for the unmanipulated stimuli of each passband,
and the cue weight was computed for each bias and passband
condition. The lateral angle gain was unaffected by stimulus
passband, and was close to unity for all listeners except for
listener S69, who showed a mild decrement in lateral gain in
the low-pass condition. Similar, near-unity, unbiased lateral
angle gains were obtained for all other listeners in all condi-
tions of Experiments I–IV. This result indicated that low-
pass or high-pass filtering did not impair listeners’ ability to
judge accurately the lateral angle of the unbiased virtual free-
field targets, and it increased our confidence that, in the bi-
ased conditions, the listeners also had access to usable cues
to lateral angle in all passband conditions.
The cue bias weights are shown in the scatter plot of
Fig. 5, in which ILD-bias weight is plotted against ITD-bias
weight for each of the ten listeners in each passband condi-
tion.
The computed weights for the ITD and ILD cues were
always greater than 0, and only in two cases exceeded 1 ?for
S68, WITD,LP?1.01; for S92, WITD,WB?1.01). Weights less
than 1 were expected because the manipulated cue was
placed in opposition to all the unmanipulated cues, which
corresponded to a zero-bias target location. The unmanipu-
lated cues thus opposed or diluted the effect of the biased
cue. For clarity, the computed standard errors of the weights
are not shown in Fig. 5, but they ranged from 0.01 to 0.04 for
both ITD and ILD. The median standard error was ?0.025.
The relation between the ITD and ILD weights de-
pended on the stimulus passband condition. In the wideband
condition ?open circles?, ITD was weighted more heavily
than ILD for all listeners ?mean WITD,WB?0.82, mean
WILD,WB?0.52),although the
weights was small for listeners S67 and S86, and close to
zero for S74. In the low-pass condition ?filled squares?, ITD
weights were ?0.8 for eight of the ten listeners, and were
substantially higher than the ILD weights for all listeners
?mean WITD,LP?0.88, mean WILD,LP?0.24). In the high-pass
condition ?filled circles?, the relation between ITD and ILD
weights was reversed from that observed in the low-pass
case: ITD was weighted less heavily than ILD for all listen-
ers except S92 ?mean WITD,HP?0.24,
?0.82). For all listeners, ITD weights were much lower than
those observed in the wideband and low-pass conditions.
There were, however, marked individual differences in
the weight given to the ITD cue in the high-pass passband
condition. Six of the ten listeners ?S68, S69, S74, S77, S86,
S91? had small ??0.20? high-pass ITD weights, whereas the
other four ?S66, S67, S76, S92? had larger weights ?0.24 to
0.60?. Our high-pass stimuli were limited to frequencies
above 4 kHz, and because the auditory system has virtually
no representation of waveform fine structure at frequencies
this high ?Palmer and Russell, 1986?, we presume that any
ITD information utilized by these listeners must have been
derived from the envelopes of the signals. We investigated
the relative influence of the ITD cues provided by the onset
and ongoing portions of the envelope in Experiment II.
differencebetween the
mean WILD,WB
2. Polar angle responses
We inspected the polar angle components of the listen-
ers’ responses under the conditions of restricted stimulus
passband and imposed ITD or ILD bias. With no imposed
bias, polar angle responses were similar to those observed in
other free-field and virtual localization studies. In the wide-
band, unbiased condition, most listeners responded accu-
rately in polar angle, but with rates of front/back confusion
that varied among listeners. Errors of elevation and front/
back position were most common for targets high above and
behind the listener. The rate of front/back confusions in-
creased for some listeners in the high-pass condition. Listen-
ers’ ability to localize the virtual stimuli in the vertical plane
demonstrated that our stimulus synthesis was of sufficient
quality to deliver accurate spectral cue information. We
therefore presume that any influence of these spectral cues
on apparent lateral angle ?see Secs. V and VI? was similar to
that occurring in real-world auditory environments. In the
low-pass condition, no listeners were able to judge accu-
rately the polar angle of the targets, and all made responses
near the horizontal plane in either the front or rear hemi-
FIG. 5. Measured ITD- and ILD-bias weights for Experiment I. ILD-bias
weight ?vertical axis? is plotted against ITD-bias weight ?horizontal axis? for
each listener in the wideband, low-pass, and high-pass passband conditions.
2224 J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002 E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 7

sphere for such filtered targets. This is a response pattern
typically observed for low-pass stimuli ?for example, Carlile
et al., 1999; Morimoto and Aokata, 1984?.
The change in polar angle responses to ITD- or ILD-
biased targets varied among listeners, but in general, the
medium-magnitude biases ??300 ?s ITD or ?10 dB ILD?
produced little change from the unbiased condition. For ap-
proximately half of the listeners, large-magnitude biases
??600 ?s ITD or ?20 dB ILD? substantially increased the
rate of front/back confusions ?as seen, for example, in Fig. 2?
and caused mild compression of polar angle responses to-
wards the horizontal plane. Wightman and Kistler ?1992?
have reported similar effects on vertical-plane localization
for ITD-biased virtual stimuli. For the other listeners, even
the large biases had little effect on polar responses. Sensitiv-
ity of polar response patterns to imposed ITD or ILD bias did
not appear to be correlated with the weight given by indi-
vidual listeners to the biased cue.
C. Discussion
1. Previous lateralization studies
Much of our knowledge about the auditory system’s pro-
cessing of interaural difference comes from so-called lateral-
ization studies, in which stimuli are delivered over head-
phones without DTF filtering and their apparent positions lie
inside the listener’s head. In many respects, our derived
weights for the ITD and ILD cues are consistent with the
results of lateralization experiments employing both pure-
tone and noise stimuli. We observed moderate to large
weights on both ITD and ILD for wideband targets, although
ITD was usually weighted more strongly; large ITD and
small ILD weights for low-pass targets; and small ITD and
large ILD weights for high-pass targets.
The intracranial position of diotic wideband noise can be
displaced from the midline by the application of either ITD
or ILD, and can be fully shifted to one side by either cue;
similar sensitivities to ITD and ILD exist for low-frequency
tones and low-pass noise ?Blauert, 1997; Pinheiro and Tobin,
1969?. ILD can displace the images of high-frequency tones
?Fedderson et al., 1957? and bands of noise ?Simon and Ale-
ksandrovsky, 1997?, but high-frequency tones above ?1.3
kHz cannot be lateralized on the basis of ongoing ITD
?Zwislocki and Feldman, 1956?. It is not clear whether the
decline in interaural phase sensitivity above this frequency is
caused by loss of phase-locking in the auditory nerve or to a
lower-frequency cutoff particular to the binaural system.
Phase-locking at frequencies up to ?4 kHz has been ob-
served in the squirrel monkey auditory nerve ?Rose et al.,
1967?, but whether this is representative of the human audi-
tory system is unknown.
In principle, the auditory system should be able to ex-
ploit envelope fluctuations in such signals to extract ITD
information, but thresholds for detecting ITD changes in
amplitude-modulated high-frequency tones are typically
much larger than those for low-frequency tones ?e.g., Mc-
Fadden and Pasanen, 1976?. Also, several studies have
shown that the extent of lateralization possible based on
high-frequency envelope ITD is in general small and
listener-dependent ?Blauert, 1982; Henning, 1974; Trahiotis
and Bernstein, 1986?. This parallels the typically small, but
occasionally substantial, high-frequency ITD weights de-
rived in our experiments I and II. ?See Sec. IVD to follow.?
The one aspect in which our results do not correspond to
those obtained in lateralization studies is the low weight
given to ILD in the low-pass condition of our Experiment I
?mean weight, 0.24; see Fig. 5?. When presented over head-
phones without DTF filtering, the intracranial position of
low-pass noise is sensitive to ILD. Pinheiro and Tobin ?1969?
found that the image of broadband noise and noise low-pass
filtered at 1.2 kHz could be shifted fully to one side by an
ILD of 9 dB. Although naturally occurring ILDs are smallest
in the low-frequency regime, this 9-dB value does not ex-
ceed the physiological range. For all of our listeners, ILD
computed over the 0.5–2-kHz band approached or exceeded
10 dB at lateral angles of 50–60 degrees. Similar ILDs were
observed in a 0.8–1-kHz band by Wightman and Kistler
?1997a, Fig. 6?.
The difference between the low-pass lateralization result
and our own might be related to the externalization of our
stimuli. Our unbiased lateral angle gain measure indicated
that listeners were able to judge accurately the lateral angle
of these low-pass targets, and, anecdotally, none reported in-
the-head-localization for biased or unbiased low-pass targets.
Presumably the 2-oct bandwidth and DTF filtering of these
targets was sufficient to create externalized auditory images
and perhaps to engage a different mode of cue processing, in
which ILD plays a much reduced role in the determination of
perceived lateral angle. The role of low-frequency ILDs in
near-field distance perception is discussed in Sec. IIIC3.
The relative potency of ITD and ILD has been explored
in many lateralization studies in which a time-intensity trad-
ing ratio was measured. The trading ratio measured using a
‘‘centering’’ method describes the amount of time difference
favoring one ear that is required to center the image of a
signal presented with a level difference favoring the opposite
ear. In the ‘‘pointer’’ method, the listener adjusts the ITD of
a broadband noise to match the lateral location of the experi-
mental stimulus presented with some combination of ITD
and ILD ?e.g., Moushegian and Jeffress, 1959?. Although
prone to intersubject differences and level dependencies, a
typical result is that the trading ratio for low-frequency
stimuli is lower than that for high-frequency stimuli, which
indicates that ITD is more potent relative to ILD at low fre-
quencies. For example, Harris ?1960? reported trading ratios
of ?25 ?s/dB and ?60 ?s/dB for low-passed and high-
passed clicks, respectively.
Stimuli with images centered using conflicting ITD and
ILD cues can be readily discriminated from diotic stimuli
?Hafter and Carrier, 1969? perhaps because narrow-band
stimuli presented over headphones with conflicting ITD and
ILD cues can generate multiple intracranial images ?e.g.,
Whitworth and Jeffress, 1961?. The location of the so-called
‘‘time’’ image is determined primarily by ITD ?and thus ex-
hibits a very low trading ratio?, while the ‘‘intensity’’ image
location is controlled by both ILD and ITD ?higher trading
ratio?. Hafter and Jeffress ?1968? found that trading ratios for
both types of images were higher for high-passed clicks than
2225J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 8

for 500-Hz tone pips, again indicating the reduced potency
of ITD at high frequencies. Gaik ?1993? examined the con-
ditions under which narrow bands of noise were most likely
to produce multiple or noncompact intracranial images.
Natural combinations of ITD and ILD in each frequency
band were identified using measured directional impulse re-
sponses from a human subject. Results showed that the lat-
eralized images were most likely to be unitary and compact
when the imposed ITD and ILD were close to a natural pair-
ing of these cues. It is not known whether similar image
splitting occurs for wideband noise signals with conflicting
binaural difference cues. Our data ?see Sec. IIIC2? and those
of Wightman and Kistler ?1992, 1997b? do not provide
strong evidence for such an effect in localization experi-
ments.
2. Localization or lateralization?
Although our results are substantially in agreement with
many lateralization studies describing the relative strengths
of ITD and ILD cues and the perception of binaural stimuli
with conflicting time and level cues, we believe that ours was
not a lateralization experiment. First, our stimuli consisted of
multi-octave, DTF-filtered bands of noise, whereas a major-
ity of trading ratio, binaural discrimination, and extent-of-
lateralization studies have used stimuli of restricted band-
width. Our stimuli thus provided listeners with the
opportunity to integrate binaural difference information
across frequency, to use monaural spectral cues, and to ex-
perience externalized auditory images.
Second, several pieces of evidence make us confident
that our listeners perceived externalized ?rather than intrac-
ranial? images and were localizing them as they would real
free-field targets. It was not practical to combine distance
estimation with our localization response method, but in a
VAS study using ILD-biased stimuli very similar to ours
?Wightman and Kistler, 1997b?, listeners reported external-
ized images even for extreme ILD biases. None of our lis-
teners reported trouble making orienting responses to the bi-
ased stimuli, which might have been expected were the
images not externalized. S18 ?the first author?, S04 and S93
?laboratory colleagues? all described the images as well ex-
ternalized and noticed no obvious differences between biased
and unbiased stimuli.
Third, our data provide no indication that listeners were
responding to multiple images produced by the biased
stimuli. Narrow-band signals presented in a lateralization
paradigm with conflicting time and amplitude cues often pro-
duce multiple intracranial images ?i.e., ‘‘time’’ and ‘‘inten-
sity’’ images?. Had this occurred frequently with our VAS
stimuli, we would expect to observe bimodality in lateral
angle responses and in plots of observed versus imposed
bias. This would have been particularly evident in the ILD-
biased conditions, for which the location of the ‘‘time’’ im-
age would be highly insensitive to the manipulation. Bimo-
dal response patterns were not observed ?for example, Fig.
4?. Even if multiple images were generated, we are content
that our derived weights describe the contributions of ITD
and ILD to the image that dominated the percept and drove
the listeners’ orienting responses. Scatter in listeners’ lateral
angle responses did not increase markedly as ITD or ILD
bias was imposed.
Finally, the strongest evidence that the biased stimuli
were localized rather than lateralized comes from the polar
angle response data. Imposition of large ITD or ILD biases
did produce an increase in front/back confusions for some
listeners, but, as in the example shown in Fig. 2, listeners
continued to respond accurately to the original elevation of
the target. This accuracy would seem unlikely if the orienting
responses were derived from a nonexternalized image.
3. Previous localization studies
Our work is not the first to use a localization task in
addressing the relative roles of ITD and ILD in spatial hear-
ing. Sandel et al. ?1955? used a loudspeaker array to produce
natural and unnatural combinations of interaural phase and
intensity for pure tones, and concluded that ITD was the
dominant lateral angle cue for frequencies below 1.5 kHz.
Other researchers have used VAS techniques and noise
stimuli to explore the binaural cue weighting. Wightman and
Kistler ?1992? presented virtual free-field targets in which the
interaural phase spectrum was manipulated to correspond to
a lateral angle of 90 degrees, rather than to the natural lateral
angle of the target. For wideband noise stimuli, listeners’
lateral angle judgments agreed with the manipulated interau-
ral phase cue, but with frequencies below 2.5 kHz removed
by high-pass filtering, the influence of the fixed ITD cue was
almost eliminated for most, but not all, listeners. This result
demonstrated the dominance of low-frequency ITD informa-
tion over other interaural cues and the lack of influence of
high-frequency ITD cues. In order to prevent listeners from
learning the individual spectral characteristics of their loud-
speakers, Wightman and Kistler scrambled the spectra of
their noise targets from trial-to-trial in
might have obscured or weakened the salience of the veridi-
cal spectral cues competing with the manipulated ITD cue.
For some of the listeners in our Experiment I, the wideband
ITD weight was lower than the low-pass ITD weight, which
might indicate increased influence of high-frequency spectral
or ILD cues in our unscrambled, wideband noise targets.
Wightman and Kistler did not seek to differentiate between
the roles of DTF spectra and ILD as the salient cues to lateral
angle for high-pass stimuli. The results of our Experiment IV
?Sec. VI?, however, suggest that DTF spectra have little in-
fluence on lateral angle judgments.
Wightman and Kistler ?1997b? also obtained results in
agreement with ours in a virtual free-field localization con-
dition almost identical to the wideband-ILD condition of our
Experiment I. Broadband ?0.2–14 kHz? noise targets were
presented with an attenuation of 0-40 dB applied to the left-
ear signal. For level imbalances ?equivalent to our ILD bias?
of 10 or even 20 dB, the listeners’ responses were generally
unperturbed, and corresponded to the position indicated by
the unmanipulated ITD and spectral cues. This result was in
agreement with our finding that ILD is a weak cue for lateral
angle in wideband noise stimuli.
Our results are in agreement with those of Wightman
and Kistler ?1992, 1997b? that ITD is the dominant lateral
1
3-oct bands. This
2226 J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 9

angle cue for stimuli containing low-frequency components.
We note, however, that this dominance has been detected
only in experiments conducted in anechoic or virtual free-
field environments, and might not apply in reverberant envi-
ronments. Hartmann and Constan ?1998? found that the bin-
aural coherence of low-frequency signals presented in rooms
is lower than that required to support lateralization on the
basis of ITD. Virtual auditory space ?VAS? studies incorpo-
rating synthesized reverberation would be useful in deter-
mining the weighting of cues under nonanechoic conditions.
Low-frequency ILDs for lateral sources in the near-field
??1 m? are much larger than those observed for distant
sources, but low-frequency ITDs are not strongly distance-
dependent ?Brungart and Rabinowitz, 1999?. Lateral angle
localization judgments remain accurate in the near-field even
as low-frequency ILDs diverge from their far-field values
?Brungart et al., 1999?. These results suggest that, in accord
with our finding from the low-pass passband condition of
Experiment I, apparent lateral angle for low-pass sources is
determined on the basis of ITD ?distance-independent? and is
independent of ILD ?distance-dependent?. Low-frequency
ILDs have been shown to be important cues for near-field
apparent distance ?Brungart, 1999?. We hypothesize that low-
frequency ILDs have little effect on apparent source direc-
tion because they are reserved as cues for near-field distance
perception.
IV. EXPERIMENT II: WEIGHTING OF ONSET AND
ONGOING ENVELOPE-BASED ITD CUES IN HIGH-
PASS NOISE
A. Motivation
Experiment I revealed marked individual differences in
the perceptual weighting of ITD in high-pass noise stimuli.
Because the peripheral auditory system cannot transduce the
fine structure of these stimulus waveforms, listeners who
placed substantial weight on the high-pass ITD cue must
have derived interaural timing information from the enve-
lopes of the signals. Information might be extracted by inter-
aural processing of stimulus onsets or offsets. Moreover, in-
formation might be extracted from ongoing fluctuations in
the envelopes of the noise waveform, although Middlebrooks
and Green ?1990? have hypothesized that modulation depth
in the outputs of high-frequency auditory filter channels is
too small for robust extraction of ongoing ITD cues from
noise stimuli.
In experiment II, we explored the relative influence of
ITD cues from the onset and ongoing portions of the enve-
lope in high-pass noise stimuli. We attempted to weaken the
transient envelope ITD cues by lengthening the duration of
the stimulus onset and offset ramps. In a separate condition,
we attempted to strengthen the ongoing cues by amplitude-
modulating the target noise bursts.
B. Stimuli and locations
In Experiment II-A, the high-pass ITD and ILD bias
manipulations of Experiment I were repeated with identical
sets of target location and bias combinations, but the tempo-
ral characteristics of the target noise bursts were altered. The
onset and offset ramps of the noise bursts were lengthened
from 1 to 20 ms in order to weaken the onset/offset ITD cue.
To preserve the 98-ms plateau duration present in the 1-ms-
ramped stimuli, the length of the noise bursts was increased
to 138 ms. All ten listeners from Experiment I participated in
Experiment II-A.
In experiment II-B, alterations of the onset/offset and
ongoing portions of the envelope were combined factorially.
Onset/offset ramps were either short ?1 ms? or long ?50 ms,
198 ms total duration?, and the envelope of the ongoing por-
tion was either flat or modulated. In the modulated case, the
ongoing portion of the signal consisted of 4-ms segments of
silence alternating with 6-ms noise bursts with 1-ms onset
and offset ramps.1The long onsets were intended to reduce
the salience of the onset ITD cue, and the amplitude modu-
lation was intended to enhance the salience of the ongoing
envelope ITD cue. Data were collected for the high-pass
passband condition with either ITD or ILD biases and for all
four combinations of the ramp and modulation parameters.
Each of the eight stimulus sets ?1 passband?2 cue bias
types?2 onsets?2 modulations? was presented only once.
All other details of the stimulus sets and their presentation
were identical to those of Experiment I, including the block
size of 111 trials. In the 1-ms-onset/flat-envelope condition
for listeners S91 and S92, the data from the equivalent con-
dition of Experiment I were used.
C. Results
1. Experiment II-A
The high-pass ITD and ILD weights for the 20-ms onset
stimuli are compared to those for the 1-ms onset targets in
Fig. 6. The four listeners whose 1-ms ITD weights were
?0.24 in Experiment I all exhibited reduction of the ITD
weight in the 20-ms onset condition. This suggests that they
had been relying on the onset ITD cue. The ITD weights for
two of these four listeners remained fairly high, however,
suggesting either that onset ITD information was still avail-
able with the 20-ms onset or that these listeners could exploit
FIG. 6. Effect of onset ramp duration on high-pass ITD and ILD bias
weights. The scatter-plot shows high-pass ITD ?filled symbols? and ILD
?open symbols? bias weights for each of the ten listeners in Experiment II-A.
Horizontal axis: cue bias weights for 1-ms onset stimuli; vertical axis: cue
bias weights for 20-ms onset stimuli.
2227 J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 10

intrinsic envelope fluctuations to obtain ongoing ITD infor-
mation. Further evidence for the reduced salience of the on-
set ITD in the 20-ms condition is provided by the increase in
the ILD weights obtained for the longer-onset stimuli. In ILD
bias conditions, the ITD cue was consistent with the unbi-
ased target location. Thus weakening of the ITD cue should
have reduced the influence of cues competing with the biased
ILD.
2. Experiment II-B
ITD and ILD weights for the stimuli of Experiment II-B
are shown for five listeners in Fig. 7. The unbiased lateral
angle gain was consistent across envelope condition for all
listeners, which indicates that the envelope manipulations
did not affect the listeners’ localization accuracy. In the
1-ms-onset/flat-envelope condition, the weights for listeners
S18, S04, and S93 ?0.19, 0.24, and 0.13, respectively? were
consistent with the distribution of weights observed in the
high-pass ITD bias condition of Experiment I.
We consider first the effect of the onset-time manipula-
tion on ITD weight. In the flat-envelope conditions, the ef-
fect of lengthening the signal onset and offset from 1 to 50
ms was small ?mean ITD weight reduction, 0.03? except for
listener S92, whose larger ITD weight decreased from 0.60
to 0.50. Thus, as found in experiment II-A, weakening the
onset ITD cue had little impact on the listeners whose ITD
bias weights were already small in the 1-ms-onset condition
?S91, S18, S04, and S93?, likely because of a simple floor
effect. For S91 and S04, the ITD bias weights actually in-
creased slightly ?by only 0.03 and 0.04, respectively?. In the
modulated-envelope conditions, lengthening the onset had
little effect on the ITD weight for any listener ?mean ITD
weight reduction, 0.04?.
In contrast, the envelope modulation manipulation had a
pronounced effect in both the 1- and 50-ms onset conditions.
Four of the five listeners exhibited an increase in the weight-
ing of the ITD cue in both onset-length conditions when
envelope modulation was added ?for these four, mean ITD
weight increase over both onset lengths, 0.16?. For these four
listeners, adding envelope modulation also produced modest
decreases in the ILD weight ?mean ILD weight reduction,
0.09?. A similar inverse relation between ITD and ILD
weights was observed in Experiment II-A. Listener S92, for
whom the high-pass ITD weight in Experiment I was already
high ?0.60?, maintained a consistently high weight on the
ITD cue but did not exhibit an increase in the weight in the
modulated-envelope conditions.
Together, the results of Experiments II-A and II-B sug-
gest that both onset and ongoing envelope ITD cues play a
role in the sensitivity of some listeners to high-frequency
ITD. It appears, however, that the highest high-frequency
ITD bias weights are obtained when the listener is able to
process ongoing envelope ITD cues, and that under such
circumstances, the onset cue is of reduced importance.
D. Discussion: Role of high-frequency envelope ITD
cues
As noted previously, complex, high-frequency stimuli
such as amplitude-modulated tones, tone complexes, and
bands of noise provide the auditory system with ongoing
envelope ITD information that is absent from high-frequency
pure-tone stimuli. ITD discrimination thresholds measured
for such sounds are typically found to be two to ten times
larger than those for low-frequency tones—as large as a few
hundred ?s ?e.g., Bernstein and Trahiotis, 1994; Blauert,
1982; Henning, 1974; McFadden and Pasanen, 1976; Nuet-
zel and Hafter, 1981?. Although easily detectable, ITDs of
physiologically plausible magnitude have been found to be
remarkably ineffective in displacing the intracranial images
of high-frequency, complex stimuli away from the midline in
lateralization experiments, and both detection thresholds and
lateral displacement sensitivity display marked individual
differences ?Trahiotis and Bernstein, 1986?.
FIG. 7. Effect of onset ramp duration and envelope modulation on high-pass
ITD and ILD bias weights. ITD bias weight ?black bars? and ILD bias
weight ?open bars? are plotted for five listeners for each of the four combi-
nations of onset duration ?1 or 50 ms? and noise envelope modulation ?flat or
100 Hz, 100% modulation depth?. Error bars show the standard error of the
regression coefficient ?i.e., of the cue bias weight?.
2228 J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002 E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 11

Our results from Experiment I paralleled these findings
?in a localization rather than lateralization task? both with
respect to the generally low weights accorded to high-
frequency ITD and to the variation of those weights among
listeners.
The results of Experiments II-A and II-B show that ?a?
some listeners with substantial weights on high-frequency
ITD derive this information primarily from the onset ITD
cue, ?b? others are able to extract ongoing ITD from intrinsic
noise envelope fluctuations, and ?c? if robust envelope modu-
lation is present, weights on high-frequency ITD increase
and the strength of the onset cue has little effect. These re-
sults are consistent with the findings that listeners vary in
their ability to extract envelope-based, ongoing ITD informa-
tion, and that onset cues are salient only when the normally
potent ongoing cues are ambiguous. The psychophysical lit-
erature contains a diversity of reports on the relative strength
and discriminability of onset and ongoing ITD cues ?for ex-
ample, Buell et al., 1991; Hafter and Dye, 1983; Tobias and
Schubert, 1959?. In a synthesis of these and other results,
Freyman and colleagues ?1997? concluded that results differ
because the contribution of the onset cue is strongly depen-
dent upon the spectral, temporal, and binaural characteristics
of the ongoing stimulus. In particular, ‘‘lateralization of a
spectrally dense signal with an unambiguous ongoing delay
is not subject to dominance by onsets even if the onset cue
itself is strong’’?Freyman et al., 1997?. In our Experiment II,
the high-pass noise signals satisfied the condition of spectral
density, and ambiguity in ongoing envelope ITDs was re-
duced by applying envelope modulation.
The results of a simulation of high-frequency, envelope-
based ITD discrimination ?see Appendix? suggest that intrin-
sic envelope fluctuations in unmodulated high-frequency
noise bands are not sufficiently large for useful discrimina-
tion of ongoing, envelope-based ITDs. This result supports
the hypothesis of Middlebrooks and Green ?1990?. Modula-
tion depth was found to be independent of differences in
listeners’ DTFs, and therefore individual differences in high-
pass ITD bias weights must be related to individual differ-
ences in envelope extraction processes or in information pro-
cessing strategies. Constan and Hartmann ?2001? have
shown that in reverberant environments such as those in typi-
cal rooms, the coherence of the two ear signals is often in-
sufficient to permit lateralization on the basis of high-
frequency envelope ITD. This may be another reason that
most listeners discount these cues.
For noise stimuli, intrinsic envelope fluctuations in dif-
ferent auditory filter channels are not correlated. That is, the
channel signals are not comodulated. Comodulation is
known to be an important factor in across-frequency integra-
tion of information ?e.g., Hall and Grose, 1990?, and it is
possible that, for noise stimuli, the lack of comodulation
among envelopes in different auditory filter bands reduces
the salience of the high-frequency envelope cue for most
listeners. The increased ITD bias weight we observed when
amplitude modulation was imposed on the noise targets in
Experiment II-B might have been a function of increased
interchannel comodulation as well as increased modulation
depth. Saberi ?1995? has shown that ITD discrimination
thresholds for stimuli composed of two spectrally separated,
high-frequency, narrow bands of noise are lower when the
temporal envelopes of the bands are identical than when they
are different. Similarly ?but in a low-frequency regime?, Tra-
hiotis and Stern ?1994? have found that complexes of spec-
trally separated SAM tones with consistent carrier ITDs pro-
duce compact unitary binaural images only when identical
modulators are applied to each carrier.
Eberle et al. ?2000? investigated the salience of
envelope-based ITD cues in a free-field localization task.
Listeners reported the apparent location of a high-frequency
octave band of noise ?7–14 kHz?, with and without an ap-
plied 20- 80- or 320-Hz, 100%, sinusoidal amplitude modu-
lation. This signal was similar to the modulated-envelope
targets used in Experiment II-B of the present study. The
authors found that the introduction of amplitude modulation
produced no reduction in the mean magnitude of errors in
listeners’lateral angle judgements. They inferred that the am-
plitude modulation did not facilitate the extraction of
envelope-based, high-frequency ITD cues. The results of our
Experiment II-B suggest that ITD salience is enhanced by
amplitude modulation, and thus are not in agreement with
this conclusion.
The discrepancy between our conclusions and those of
Eberle et al. ?2000? likely resulted from the fact that our
VAS technique permitted the elimination of a confound be-
tween ITD and ILD cues inevitably present in the free-field
situation. In our Experiments I and II, listeners were able to
judge accurately the lateral angle of the unmanipulated high-
pass targets despite their generally low sensitivity to the ITD
cue. The veridical judgements were, therefore, likely to have
been based on the ILD or on spectral cues ?but see Sec. VID
to follow?. Using free-field target presentation, it is not pos-
sible to present stimuli in which the various localization cues
are in conflict. If, similarly to our subjects, the listeners in
the Eberle et al. ?2000? study were able to judge lateral angle
accurately on the basis of the sufficient and veridical ILD
cues, then adding ITD information consistent with the ILD
via amplitude modulation would not have changed the appar-
ent location of the source. The consistently low lateral local-
ization error across conditions might reflect the resolution of
the motor response method rather than evidence of complete
discounting of the high-frequency ITD cue.
V. EXPERIMENT III: WEIGHTING OF THE INTERAURAL
LEVEL SPECTRUM CUE
A. Motivation
The frequency-independent ILD bias manipulation used
in Experiments I and II resulted in stimuli with unnatural
interaural level spectra ?ILS; i.e., patterns of ILD across
frequency?. As a wideband sound source is moved away
from the median plane, ILD grows more rapidly at high fre-
quencies than at low ones, and thus the frequency-
independent ILD created by applying ILD bias to a midline
source in Experiments I and II would never be observed
under natural conditions. In principle, the ILS could serve as
a robust cue to source elevation and front-back location for
wideband sources ?Duda, 1997?, because it is the difference
2229 J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 12

between the spectra at the left and right ears, and therefore
largely independent of irregularities in the source spectrum.
There is, however, little psychophysical evidence to support
the salience of ILS as a vertical-plane localization cue ?see
Discussion, Sec. VE?.
In Experiment III, we attempted to bias the perceived
location of virtual free-field targets without introducing un-
natural patterns of ILD across the frequency spectrum.
B. Stimuli and locations
The targets were 100-ms noise bursts with 1-ms raised-
cosine onsets and offsets. Stimuli with biased interaural level
spectra were generated as follows. Starting with an original
target location, A, we identified a second location, B, with
the same polar angle, but displaced in lateral angle from A
by ?30 or ?60 degrees.
For the ear on the same side as the target ?the near ear?,
we retained the DTF of location A, DTFnear(A,f ). For the
opposite ear ?the far ear?, we generated a new transfer func-
tion, X(f ), such that the resulting ILS matched that of loca-
tion B. That is,
?
DTFnear?A,f ?
X?f ? ??ILS?B,f ???
DTFnear?B,f ?
DTFfar?B,f ??.
It follows that
?X?f ????DTFnear?A,f ??DTFfar?B,f ?
DTFnear?B,f ??.
In practice, we computed X(f )?DTFfar(A,f )?H(f ),
where H(f ) was the zero-phase filter with transfer function
H?f ???
DTFnear?A,f ?
DTFfar?A,f ????
DTFnear?B,f ?
DTFfar?B,f ???ILS?A,f ?
ILS?B,f ?.
This operation preserved the natural near-ear DTF and
the interaural phase spectrum ?and hence the ITD? measured
at location A, but altered the ILS to correspond to that of
location B ?see Table I?. Positive ?rightward? ILS bias was
applied only to targets originally on or to the right of the
median plane. Similarly, negative ILS bias was applied only
to midline or left-hemisphere targets. Thus in the above ex-
pression for H(f ), the ear designated ‘‘near’’ remained the
same for location A and location B, as did the ear designated
‘‘far.’’
The wideband, low-pass, and high-pass passband condi-
tions used in Experiment I were employed. The groups of
target locations for Experiment III were similar to those for
Experiments I and II, but because ILS bias was applied only
to shift apparent position laterally away from the median
plane, the total number of biased targets was reduced from
144 to 80. In total there were three stimulus sets ?3
passbands?1 cue bias type? of 158 targets each. The targets
in each set were presented twice in a shuffled order over the
course of four blocks of 79 trials each. Blocks from different
passband conditions were intermixed within an experimental
session but trials from different sets were not intermixed
within blocks.
C. Analysis
Because the change in the ILS caused by the manipula-
tion was not a scalar quantity, as were the imposed ITD or
ILD biases of Experiments I and II, we first computed a
weight for the ILS cue by simply computing the slope of a
linear fit to the observed lateral angle response bias versus
imposed ILS bias data, both measured in degrees. This gave
a weight in units of degree/degree. We also analyzed the
responses as we did in Experiment I, treating gross ILD, the
overall interaural difference in energy integrated across the
stimulus passband, as the effective cue. In this case, the im-
posed bias was computed from the difference between the
interaural energy ratios present in the original and modified
DTFs, and the observed bias was computed as in Experiment
I ?Sec. IIE?. A linear regression between the observed and
imposed biases yielded an ILS bias weight in units of dB/dB.
D. Results
The cue weights derived from the responses to ILS-
biased stimuli are shown in Fig. 8, along with the corre-
sponding ILD weights from Experiment I ?or equivalent con-
FIG. 8. Interaural level spectrum and interaural level difference bias weights
for Experiment III. Each panel shows data for one listener. ILS-degree/
degree bias weight ?shaded/hatched bars?, ILS-dB/dB bias weight
?unshaded/hatched bars? and ILD bias weight ?open bars? are plotted for the
wideband, low-pass and high-pass passband conditions. Error bars show the
standard error of the regression coefficient ?i.e., of the cue bias weight?.
2230J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 13

ditions for S18?. In general the relation between ILS-degree/
degree weights
?shaded/hatched
condition was similar to that found for the ILD weights
?open bars? of Experiment I. That is, low weight was given
to the ILS cue in the low-pass condition ?three of four lis-
teners?, and a higher weight was given in the high-pass con-
dition. Curiously, small negative ILS-degree/degree and ILS-
dB/dB weights were observed for listener S18 in the low-
pass condition.
Several features of the data suggest that the detailed
shape of the interaural level spectrum itself is not a particu-
larly salient cue for lateral angle, but rather that the effective
interaural level cue is the overall energy difference between
the ears. For three of the four listeners, the ILS-degree/
degree weight was lower than the ILD weight from experi-
ment I in both the wideband and high-pass conditions. Also,
in the majority of cases in the wideband and high-pass con-
ditions, the ILS-dB/dB ?gross ILD? weight was higher than
the ILS-degree/degree weight computed from the same re-
sponses.
In some cases, a particular gross ILD produced more
response bias when it was produced by the ILS manipulation
than when it was applied as a flat attenuation. Evidence for
this can be found in the high-pass weights for S86 and S91.
For these listeners, the high-pass ILS-dB/dB weight ?Fig. 8,
hatched bars? was substantially higher than the high-pass
ILD weight from Experiment I ?open bars?.
bars?
and passband
E. Discussion: Interaural level spectrum and the
nature of ILD processing
The shape of the interaural level spectrum has been pro-
posed as a cue for vertical-plane localization for locations
both on ?Searle et al., 1975? and off ?Duda, 1997? the median
plane. The results of several psychophysical studies contra-
dict this proposal. First, vertical-plane localization of wide-
band sources can be disrupted by certain source-spectrum
irregularities which preserve the ILS ?e.g., Macpherson,
1996, 1998; Rakerd et al., 1999; Wightman and Kistler,
1997a?. Second, there is evidence that the shape of the spec-
trum at the ear contralateral to the target can be altered sub-
stantially without consequent degradation in vertical-plane
localization accuracy
?Humanski
Morimoto, 2001; Wightman and Kistler, 1999?, and that the
influence of the contralateral-ear DTF declines as the source
location is moved away from the median plane ?Morimoto,
2001?. These results suggest that the details of the ILS are
unimportant for human listeners’ vertical-plane localization,
and our results from Experiment III lead to a similar conclu-
sion about the lack of efficacy of the ILS as a cue to lateral
angle.
The effective cue for lateral angle in high-pass stimuli
seems to be better modeled as the overall difference in proxi-
mal stimulus energy between the ears, or, perhaps more re-
alistically, as an integration of the ILDs observed in discrete
frequency channels. This would parallel the auditory sys-
tem’s use of low-frequency ITD as a cue to lateral angle and
its apparent insensitivity to fine details in the interaural phase
spectrum ?Kulkarni et al., 1999?. Both the ITD and ILD find-
ings are consistent with the suggestion that ‘‘ . . . the system
andButler,1988;
does not evaluate every detail of the complicated interaural
dissimilarities, but rather derives what information is needed
from definite, easily recognizable attributes’’ ?Blauert, 1997,
p. 138?.
The results of Experiment III do, however, reveal some
effect of the distribution of ILD across frequency. As dis-
cussed earlier, we observed that a net ILD bias generated by
the ILS manipulation was more effective in biasing lateral
angle judgments than was an equivalent flat interaural at-
tenuation ?Fig. 8?. That is, the dB/dB weights for ILD ob-
tained in Experiment III were higher than those obtained in
Experiment I. A possible explanation for this is that listeners
placed greater weight on the gross ILD when the ILS was
more natural, or at least when it did not contain unnatural
low-frequency ILDs. This would be consistent with the pro-
posal of Wightman and Kistler ?1997a? that the naturalness
of observed localization cues plays a mediating role in their
salience.
VI. EXPERIMENT IV: WEIGHTING OF THE NEAR-EAR
SPECTRAL CUE
A. Motivation
Some have suggested that monaural spectral cues are
important in determining perceived lateral angle as well as
being the primary cues to sound source elevation and front/
back location ?for example Butler and Flannery, 1980?. In
Experiments I, II, and III of the present study, we could not
directly assess the influence of spectral cues on perceived
lateral angle because we did not manipulate these cues inde-
pendently of both of the binaural difference cues ?ITD and
ILD? simultaneously. In all cases, the spectral cue in the ear
on the side of the original target location corresponded with
the unmanipulated binaural difference cue. The results of Ex-
periments I and II suggest, however, that the spectral cue
contribution is minimal, as we discuss below in Sec. VID.
In Experiment IV, we biased the lateral angle corre-
sponding to the spectral cue in one ear independently of both
the ITD and ILD cues. The near-ear DTF ?as defined in Sec.
VB? was replaced by another corresponding to a new loca-
tion displaced in lateral angle from the original target loca-
tion. The opposite-ear impulse response was altered in order
to maintain the original interaural level and phase spectra.
We manipulated the near-ear DTF because it is thought to be
the most potent cue for vertical plane localization and be-
cause, as discussed earlier, it appears that the far-ear spec-
trum can be manipulated severely without affecting vertical
plane localization ?Humanski and Butler, 1988; Morimoto,
2001; Wightman and Kistler, 1999?.
B. Stimuli and locations
As in the previous experiments, the targets were 100-ms
noise bursts with 1-ms raised-cosine onsets and offsets.
Stimuli with biased near-ear DTFs were generated as fol-
lows. Starting with an original target location, A, we identi-
fied a second location, B, with the same polar angle, but
displaced in lateral angle from A by ?30 or ?60 degrees.
The interaural level- and phase-difference spectra for lo-
cation A were computed as
2231J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002 E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 14

ILS?A,f ???
DTFnear?A,f ?
DTFfar?A,f ??
and
IPS?A,f ????DTFnear?A,f ?????DTFfar?A,f ??.
For the ear on the target side, DTFnear(B,f ) was substi-
tuted for DTFnear(A,f ); this constituted the spectral cue bias.
In the far-ear channel, a new transfer function, X(f ), was
synthesized with magnitude
?X?f ????
DTFnear?B,f ?
ILS?A,f ??
and phase
??X?f ?????DTFnear?A,f ???IPS?A,f ?.
Thus the stimulus had the ITD and ILS cues of location
A, but the near-ear DTF of location B. This differed from
experiment III, in which the near-ear DTF corresponded to A
and the ILS to B ?see Table I?. Prior to computing the far-ear
DTF magnitude spectrum, ILS(A,f ) was smoothed by
frequency-domain convolution with a 1-kHz-wide rectangu-
lar filter. This reduced the creation of sharp resonances in the
synthesized DTF caused by notches in DTFnear(A,f ).
Biased targets were located at lateral angles of ?30 or
?60 degrees and at elevations of 0, ?20, and ?40 degrees
in the front and rear hemispheres. At each location there
were three possible biases that did not place locations A and
B on opposite sides of the median plane. For example, a
target at a lateral angle of ?30 degrees could be biased by
?60, ?30, or ?30 degrees without crossing the median
plane.
The wideband and high-pass passband conditions used
in Experiment I were employed. In total there were two
stimulus sets ?2 passbands?1 cue bias type? of 198 targets
each. Of these targets, 120 were filtered and biased, 40 were
filtered but unbiased, and the remaining 38 were wideband,
unmanipulated targets. The targets in each set were presented
once in a shuffled order over the course of two blocks of 99
trials each. Three listeners ?S04, S18, S93? who had partici-
pated in Experiment II-B participated in Experiment IV.
C. Results
We computed a bias weight for the ispilateral DTF cue
by simply computing the slope of a linear fit to the observed
lateral angle response bias data and the imposed DTF bias
data, both measured in degrees. This gave a weight in units
of degree/degree, as in our first analysis of the ILS bias data
in Experiment III ?Sec. V C?. Although our manipulation
preserved the ILS of the original target location, it did not
necessarily preserve the overall ILD because changing the
near-ear spectrum changed the spectral distribution of en-
ergy. Because we did not wish our results to be confounded
by an unintended ILD bias, trials for which the overall stimu-
lus ILD differed from that of the original target location by
more than 5 dB were excluded from the analysis. The pro-
portion of excluded trials varied between 4% and 10%
among listeners.
The obtained weights are shown in Table II. For all lis-
teners and in both passband conditions, the magnitudes of all
weights were ?0.1, and in a majority of cases were not sig-
nificantly different from 0. These results suggest that mon-
aural spectral cues play a negligible role in determining ap-
parent lateral angle even in the high-pass condition, in which
the strong conflicting low-frequency ITD cue was removed.
D. Discussion: Influence of spectral cues on lateral
localization
In Experiments I and II of the present study, we did not
manipulate monaural spectral cues independently of ITD and
ILD to assess their influence on perceived lateral angle, but
those results also suggest that the spectral cue contribution is
minimal. In each stimulus condition of these experiments,
the spectral cues at both ears corresponded to the original
target location prior to the ITD or ILD manipulation and
were in agreement with the unmanipulated interaural cue.
Thus, any reliance on the spectral cues should have resulted
in a reduction of the weight placed on the manipulated cue,
although, considering each condition in isolation, the effects
of spectral cues and the unmanipulated cue cannot be disam-
biguated. Our finding, however, that weights approaching
unity were found for one or other of the interaural cues in
each passband condition is evidence that spectral cues do not
contribute substantially to lateral angle localization even in a
binaural listening situation. A high weight on a biased inter-
aural cue indicates that it was possible to shift the apparent
position of the target to a location inconsistent with the natu-
ral spectral cues. In Experiment IV, we directly tested the
influence of monaural spectral cues by biasing the source-
side DTF independently of the natural ITD and ILS cues. We
found that the DTF bias had little or no influence on the
perceived lateral angle of the target.
These results are in agreement with and extend the work
of Slattery and Middlebrooks ?1994? and of Wightman and
Kistler ?1997b?, who found that normal-hearing listeners ren-
dered monaural by plugging one ear were incapable of accu-
rate lateral angle judgements. In such an acute monaural situ-
ation, listeners reported all localization targets originating
from a position opposite the unoccluded ear. Slattery and
Middlebrooks did find that approximately half of their
chronically monaural listeners were able to make use of
spectral information to determine source azimuth, but they
considered this to be the result of a long-term adaptation to
the listeners’ lack of access to binaural cues, rather than be-
ing indicative of the functioning of the normal auditory sys-
tem.
TABLE II. Measured DTF-bias weights for the wideband and high-pass
targets of Experiment IV.
Listener WidebandHigh-pass
S04
S18
S93
?0.01?0.03
?0.05?0.05
0.08?0.04
0.09?0.04
0.00?0.06
0.05?0.05
2232J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002 E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source

Page 15

VII. SUMMARY AND CONCLUSIONS
Our results suggest that, in broad outline, the duplex
theory does serve as a useful description of ?if not a prin-
cipled explanation for? the relative potency of ITD and ILD
cues in low- and high-frequency regimes. In Experiment I,
we found that listeners weighted the ITD cue strongly as a
cue for lateral angle for low-pass stimuli and ?with some
exceptions? weighted ITD weakly for high-pass stimuli. The
opposite pattern was observed for listeners’ ILD weights.
Even when substantial biases were introduced, ILDs were
generally ignored at low frequencies but were given high
weights for high-pass stimuli. For wideband targets, both
cues were given substantial weight, but ITD dominated for
most listeners. In Experiment II, we found that both onset
and ongoing envelope ITD cues contributed to the relatively
minor role of high-frequency ITD information, but that the
greatest high-frequency ITD weights were observed for lis-
teners who were able to make use of ongoing cues. In Ex-
periment III, we examined the role of the detailed shape of
the interaural level spectrum as a lateral angle cue. The re-
sults suggested that the precise shape of the ILS was not as
effective a cue as the overall energy difference between the
ears. In Experiment IV, we found that monaural spectral cues
had little or no influence on perceived lateral angle.
ACKNOWLEDGMENTS
The authors are grateful to Leslie Bernstein, Fred Wight-
man, and an anonymous reviewer for their constructive com-
ments on an earlier version of this paper. William Hartmann,
Zachary Constan, Douglas Brungart, Brian Mickey, and
Christopher Stecker also provided helpful suggestions.
Zekiye Onsan provided invaluable technical assistance. This
work was funded by NIH Grant Nos. R01DC00420 and
T32DC00011.
APPENDIX: CAN INTRINSIC ENVELOPE
FLUCTUATIONS IN WIDEBAND NOISE SUPPORT ITD
DISCRIMINATION AT HIGH FREQUENCIES?
1. Motivation
For most of our listeners in experiment I, ITD had very
little influence on judgments of lateral angle for high-pass
noise stimuli. One possible explanation for this result is
that the auditory system is unable to extract robust ITD in-
formation from such signals, as suggested by Middlebrooks
and Green ?1990?. Alternatively, the system might in prin-
ciple be able to obtain useful ITD information, but listeners
discount that information for some other reason. Psycho-
physical studies of ITD sensitivity in high-frequency noise
have tended to use narrowband stimuli ?e.g., Bernstein and
Trahiotis, 1994?, which might encourage listeners to adopt an
off-frequency listening strategy not possible with a broad-
band stimulus. Because psychophysical data are not avail-
able for the multi-octave high-pass stimuli we used in Ex-
periments I and II, we conducted a series of simulations in
order to determine whether high-frequency, envelope-based
ITD is indeed a viable lateral angle cue for wideband noise
stimuli.
To play a role in high-frequency ITD sensitivity, modu-
lations in the right- and left-ear signal envelopes must exhibit
two properties. First, the envelope fluctuations must be
highly coherent ?i.e., similar?, or else ITD cannot be defined.
For wideband noise sources off the midline, the strong asym-
metry between source-side and far-side DTFs leads to deco-
rrelation of the wideband right- and left-ear envelopes. This
occurs because the envelopes of the narrowband components
that compose the wideband noise are uncorrelated and the
overall envelope depends on the relative levels ?and phases?
of these components. Within individual auditory filter bands,
however, this decorrelation might not occur if the effect of
interaural DTF asymmetry is primarily the imposition of a
level difference, which would not alter the shapes of the
envelopes.
Second, the envelope modulations must be of sufficient
depth. In a study of interaural envelope delays, Middle-
brooks and Green ?1990? hypothesized that, because of the
low-pass limits of high-frequency envelope following, the
depth of intrinsic noise envelope modulations in high-
frequency auditory filter bands might be too small to allow
effective extraction of envelope-based ITD information from
noise signals. Reduction of modulation depth increases ITD
just-noticeable differences ?jnd’s; i.e., discrimination thresh-
olds? in sinusoidally amplitude-modulated ?SAM? tones and
beating two-tone complexes ?Henning, 1974; McFadden and
Pasanen, 1976; Nuetzel and Hafter, 1981?.
A metric which simultaneously captures the effects of
changes in modulation depth and interaural envelope coher-
ence is the normalized correlation of the left- and right-ear
signal envelopes. This is computed in the same manner as
the Pearson product-moment correlation ?or normalized co-
variance?, but the mean ?or d.c.? components of the enve-
lopes are retained. Using this metric, Bernstein and Trahiotis
?1996? have successfully accounted for the dependence of
ITD jnd’s on both SAM-tone modulation depth and two-tone
modulation depth as measured by Nuetzel and Hafter ?1981?
and Pasanen ?1976?, respectively.
In our simulations, we passed wideband Gaussian noise
signals through an auditory filter-bank model. We then esti-
mated ITD discrimination thresholds (?ITD) based on the
outputs of individual filter channels by determining the mini-
mum increment in ITD required to reduce the normalized
correlation by a criterion amount. We derived estimates of
minimum audible angle ?MAAs? from these ?ITDestimates.
We also examined the sensitivity of the jnd estimates to DTF
filtering of the noise signals and to the low-pass cutoff fre-
quency of the envelope-following process.
2. Methods
We passed 16 100-ms exemplars of wideband Gaussian
noise through a gammatone auditory filter-bank model
?Slaney, 1994? with binaural channels centered at 4, 6, 8, 10,
12, and 14 kHz. Envelopes for each ear in each frequency
band were extracted by half-wave rectification followed by
low-pass filtering at either 250 or 500 Hz ?fourth-order But-
terworth filter?. All processing was done at a sampling rate of
50 kHz.
2233J. Acoust. Soc. Am., Vol. 111, No. 5, Pt. 1, May 2002 E. A. Macpherson and J. C. Middlebrooks: Duplex theory of sound source