Auditory masking patterns in bottlenose dolphins (Tursiops
truncatus) with natural, anthropogenic, and synthesized noise
Brian K. Branstetter
National Marine Mammal Foundation, 2240 Shelter Island Drive, No. 200, San Diego, California 92106
Jennifer S. Trickey, Kimberly Bakhtiari, Amy Black, and Hitomi Aihara
G2 Software Systems Inc., 4250 Paciﬁc Highway, Suite 125, San Diego, California 92110
James J. Finneran
U.S. Navy Marine Mammal Program, Space and Naval Warfare Systems Center Paciﬁc, Code 71510, 53560
Hull Street, San Diego, California 92152
(Received 8 November 2012; revised 8 January 2013; accepted 16 January 2013)
Auditory masking occurs when one sound (usually called noise) interferes with the detection, dis-
crimination, or recognition of another sound (usually called the signal). This interference can lead
to detriments in a listener’s ability to communicate, forage, and navigate. Most studies of auditory
masking in marine mammals have been limited to detection thresholds of pure tones in Gaussian
noise. Environmental noise marine mammals encounter is often more complex. In the current study,
detection thresholds were estimated for bottlenose dolphins with a 10 kHz signal masked by natural,
anthropogenic, and synthesized noise. Using a band-widening paradigm, detection thresholds
exhibited a pattern where signal thresholds increased proportionally to bandwidth for narrow band
noise. However, when noise bandwidth was greater than a critical band, masking patterns diverged.
Subsequent experiments demonstrated that the auditory mechanisms responsible for the divergent
masking patterns were related to across-channel comparison and within-valley listening.
C2013 Acoustical Society of America. [http://dx.doi.org/10.1121/1.4789939]
PACS number(s): 43.80.Lb, 43.66.Dc, 43.66.Gf, 43.66.Nm [WWA] Pages: 1811–1818
Contrary to the title of Jacques-Yves Cousteau’s 1956
documentary “The Silent World,” we know the oceans are not
now and never have been silent. Long before the industrial rev-
olution gave rise to increasing levels of anthropogenic noise
(McDonald et al., 2006), sounds from weather, seismic, and
biological sources ﬁlled ocean environments. Aquatic animals
had the relative luxury of gradually evolving coping mecha-
nisms in response to these noisy environments. However, the
rapid and recent onset of anthropogenic noise has been abrupt
on an evolutionary time scale. Auditory adaptations suited for
an ancestral acoustic environment may be ill-equipped to cope
with today’s rising ocean noise. Consequently, marine mam-
mals may experience behavioral alterations, auditory masking,
physiological damage, and potentially death.
Auditory masking occurs when one sound interferes
with a listener’s ability to hear another sound. Auditory
masking can interfere with an animal’s ability to effectively
forage, detect predators, navigate, and communicate. Studies
of masking in marine mammals have, for the most part, been
limited to simple stimuli such as Gaussian (G) noise and
pure tone signals. This approach has been useful for mapping
parameters related to auditory peripheral ﬁlters [e.g., critical
bandwidths (Johnson, 1968;Au and Moore, 1990;Southall
et al., 2003)] and auditory ﬁlter shapes (Finneran et al.,
2002), as well as for extrapolating the effects of masking
noise in the marine environment (Clark et al., 2009). A ﬁnd-
ing consistent across mammalian species, including marine
mammals, is that the auditory periphery behaves as a series
of continuously overlapping bandpass ﬁlters (Greenwood,
1990). When detecting a tonal signal in noise, the noise
within a band centered on the signal has the most masking
effect (Fletcher, 1940). In theory, an energy detector at the
output of an auditory ﬁlter registers an increase in energy
when a signal is present (Branstetter et al., 2007). If the
increase in energy in that ﬁlter is sufﬁcient, a decision is
made that a signal is present. These ﬁndings represent a spe-
cial case of auditory masking in which the instantaneous am-
plitude of the noise is sampled from a G distribution.
Recent marine mammal masking studies have demon-
strated that auditory masking is not limited to noise within a
single auditory ﬁlter if the noise is non-G (Branstetter and
Finneran, 2008;Erbe, 2008;Trickey et al., 2011). These
ﬁndings are consistent with previous studies with humans
(Hall et al., 1984) and a few other animal listeners (Klump
et al., 2001;Pressnitzer et al., 2001). For comodulated (CM)
noise, in which amplitude ﬂuctuations are correlated across
auditory ﬁlters, a release from masking occurs, a result
known as comodulation masking release (CMR) (Hall et al.,
1984). The magnitude of CMR can be deﬁned as the differ-
ence in thresholds estimated in G and CM noise (Schoone-
veldt and Moore, 1989). Natural ocean-borne noise [e.g.,
snapping shrimp (SS)] can also be CM, leading to a release
from masking (Trickey et al., 2011). Before understanding
can be gained about the auditory mechanisms that govern
masking in complex noise, basic masking studies using
Author to whom correspondence should be addressed. Electronic mail:
J. Acoust. Soc. Am. 133 (3), March 2013 V
C2013 Acoustical Society of America 18110001-4966/2013/133(3)/1811/8/$30.00
Downloaded 30 Apr 2013 to 22.214.171.124. Redistribution subject to ASA license or copyright; see http://asadl.org/terms
realistic noise types are warranted. The goals of this study
were to (1) measure auditory masking patterns for a broad
range of noise types and (2) if divergent masking patterns
exist (threshold differences as a function of noise type), con-
duct successive experiments to determine the auditory mech-
Two female Atlantic bottlenose dolphins (Tursiops trun-
catus) participated in the study: SAY (age 36) and APR (age
27). The dolphins had normal hearing at the frequencies tested
as determined via auditory evoked potential testing (Houser
and Finneran, 2006). One of the dolphins (SAY) had prior ex-
perience with psychoacoustic tasks while APR was new to
psychophysical testing. They were housed in 9 m 9m or
9m18 m ﬂoating netted enclosures (pens) located in San
Diego Bay, California. The study followed a protocol
approved by the Institutional Animal Care and Use Commit-
tee of the Biosciences Division, Space and Naval Warfare
Systems Center Paciﬁc, and all applicable U.S. Department of
Defense guidelines for the care of laboratory animals.
B. General procedure
Data were collected using the same procedures and ap-
paratus as Trickey et al. (2011). A brief description of the
procedures is presented here.
1. Signal presentation
Experimental sessions were conducted in a 3 m 9m
testing pen, and the average ambient noise spectral density
level was generally 75 dB re: 1 lPa
/Hz at 10 kHz. The dol-
phin was required to station on an underwater bite plate and
then phonate in response to a tone (tone trial) or remain silent
if no tone was present (catch trial). During each dive, the dol-
phin completed between 2 and 15 trials, as determined by a
random schedule. A one-down, one-up adaptive staircase pro-
cedure was used to adjust the level of the signal and to esti-
mate thresholds at the 50% correct detection level (Levitt,
1971). The initial step size of the signal was 5 dB, followed
by a 2 dB step size after the ﬁrst miss/reversal. The order of
tone trials and catch trials was pseudo-randomized (Geller-
man, 1933), where 50% of the trials were “signal present”
and the other 50% were “signal absent” catch trials. A session
continued until at least 11 reversals were reached. The ﬁrst re-
versal (associated with the 5 dB step size) was always
excluded when determining the threshold, which was calcu-
lated by averaging the last ten reversals. Thresholds where the
false alarm rate exceeded 20% were rejected from analysis.
III. EXPERIMENT I. CRITICAL BANDS IN NATURAL,
ANTHROPOGENIC, AND SYNTHESIZED NOISE
A. Stimuli and procedure
Signals and noise were calibrated and generated using
the same procedure as in Trickey et al. (2011). The signal
was a 500 ms, 10 kHz pure tone with 50 ms onset–offset lin-
ear ramps to reduce spectral splatter. Bandpass noise was cen-
tered on the signal frequency and was continuously projected
from the same transducer as the signal. Noise was saved as
.wav ﬁles, and was looped continuously during each experi-
mental session. The bandwidth of the noise was an independ-
ent variable and ranged between 100 and 8000 Hz. The noise
was bandpass ﬁltered (32nd order Butterworth ﬁlter, Cool
Edit Pro 2.0) to produce a ﬂat pressure spectral density level
of 95 dB re: 1 lPa
/Hz. Three noise categories were used for
this experiment: Natural, anthropogenic, and synthesized
noise. Synthesized noise types included G noise and CM
noise, and were the same noise ﬁles as in Branstetter and Fin-
neran (2008). Natural and anthropogenic noise were obtained
via ﬁeld recordings and included natural biological sounds
(SS, humpback whale song, pinniped chorus, whale social
sounds), natural non-biological sounds [rain (RN), ice squeaks
(IS)], and anthropogenic sounds [pile saw (PS), motor boat
(BT) noise, C-tractor tug boat sound, vibratory hammer, seis-
mic airgun, vessel echo-sounder]. Of these sounds, the hump-
back whale song, pinniped chorus, whale social sounds, and
C-tractor sounds were rejected because most of the spectral
energy was below the signal frequency of 10 kHz. Transient
sounds (seismic airgun, vessel echo sounder, and vibratory
hammer) were also rejected because their short durations
would only mask a small temporal portion of the 500 ms sig-
nal. The remaining noise types used in the experiment are pre-
sented in Fig. 1and Table I. See the Appendix for details of
how each noise type was acquired or generated. All noise
types were ﬁltered (FFT ﬁlter, Cool Edit Pro 2.0, Syntrillium
Software Corp., Phoenix, AZ) to compensate for the transmis-
sion voltage response of the Reson ITC 1001 transducer
(Slangerup, Denmark). The dolphin SAY participated in this
Masking patterns for the different noise types are found
in Fig. 2. Assuming the dolphin’s auditory ﬁlter is approxi-
mately 1 kHz wide, masking patterns for within-channel noise
are similar, suggesting that the auditory system processes
these sound types in a similar manner. However, masking pat-
terns diverge for bandwidths greater than 1 kHz, suggesting
that different auditory mechanisms operate on different noise
types. Diverging masking patterns can also be seen in Fig. 3,
in which the standard deviation of the data from Fig. 2is plot-
ted as a function of bandwidth. Again, a sharp increase in the
standard deviation for noise bands greater than an auditory ﬁl-
ter bandwidth suggests that different mechanisms govern
masking patterns for wideband noise.
Critical ratios (CRs) derived from wideband G noise
maskers are often used to predict masking with real environ-
mental noise. To determine how well G noise CRs predict
the current masking results, CRs were calculated for the
thresholds masked in the 8 kHz noise bands for each noise
type. In dB, the CR can be deﬁned as
is the level of the signal at threshold (re: 1 lPa),
and Nis the noise pressure spectral density (re: 1 lPa
1812 J. Acoust. Soc. Am., Vol. 133, No. 3, March 2013 Branstetter et al.: Dolphin auditory masking
Downloaded 30 Apr 2013 to 126.96.36.199. Redistribution subject to ASA license or copyright; see http://asadl.org/terms
A 22 dB range resulted, with G noise having the median CR
(Fig. 4). Tukey’s honestly signiﬁcant difference was used to
compare thresholds in G noise to thresholds in the other
noise types resulting in a signiﬁcant difference for CM
(P¼0.001), SS (P¼0.009), and IS (P¼0.003) noise. Mask-
ing by RN, PS, and BT noise was not signiﬁcantly different
than G noise masking.
IV. EXPERIMENT II. EFFECTS OF ACROSS-CHANNEL
PHASE DELAY ON SIGNAL DETECTION
According to the power spectrum model of masking
(PSM), only noise energy within an auditory ﬁlter centered
on a signal contributes to the masking of that signal (Moore,
1993). However, the masking release (MR) observed with
CM and SS noise, the masking increase observed with IS
noise, as well as CMR from previous studies (Branstetter
and Finneran, 2008;Trickey et al., 2011) demonstrate that
noise beyond an auditory ﬁlter can have an effect on the dol-
phin’s ability to detect tonal signals. One proposed mecha-
nism to explain CMR is that the auditory system extracts the
temporal envelope from a bank of auditory ﬁlters and com-
pares AM patterns across auditory ﬁlters (Hall and Grose,
1988;Branstetter and Finneran, 2008). If a sinusoidal signal
is added, the ﬁlter centered on the signal will be decorrelated
relative to the ﬂanking bands, providing a temporal cue that
a signal is present. Hypothetically, this cue will not work for
G noise because across-channel envelopes are already
TABLE I. Noise types used in Experiment I. Each noise type was saved as a
Microsoft.wav ﬁle with a speciﬁed duration and sampling rate. For source
information, please see the Appendix.
Noise type Duration Sampling rate Source
BT 60 s 44.1 kHz 1
PS 30 s 96.0 kHz 2
IS 5 s 100.0 kHz 3
RN 15 s 40.0 kHz 4
SS 3 min 96.0 kHz 1
G 3 min 44.1 kHz 5
CM 3 min 44.1 kHz 5
FIG. 2. Detection thresholds as a function of masker bandwidth for each
noise type. (G ¼Gaussian, CM ¼comodulated, PS ¼pile saw, BT ¼boat
noise, RN ¼rain, IS ¼ice squeaks, SS ¼snapping shrimp). Each data point
represents the average of four thresholds. Error bars were removed for
clarity but the standard deviation was less than 3 dB for each data point.
FIG. 3. Signal threshold standard deviations as a function of masker band-
width. Each data point represents the standard deviation of thresholds for all
noise types combined. Standard deviations are low and consistent up to
1 kHz, at which point the standard deviation increases. The 1 kHz breakpoint
represents the limit of the auditory ﬁlter resulting in processing transition.
FIG. 1. (Color online) Spectrograms of noise types used in Experiment I.
All noise types displayed are 8 kHz wide (6 to 14kHz). Gaussian (G), como-
dulated (CM), snapping shrimp (SS), rain (RN), boat noise (BT), pile saw
(PS), ice squeaks (IS). All noise types had ﬂat averaged spectral densities of
95 dB re: 1 lPa
/Hz (40 averages, 500 ms samples).
J. Acoust. Soc. Am., Vol. 133, No. 3, March 2013 Branstetter et al.: Dolphin auditory masking 1813
Downloaded 30 Apr 2013 to 188.8.131.52. Redistribution subject to ASA license or copyright; see http://asadl.org/terms
decorrelated. Branstetter and Finneran (2008) demonstrated
that when amplitude modulated noise was decorrelated
across channels, CMR was disrupted. How this mechanism
works has not been investigated in dolphins. In Experiment
II, stimuli were created to systematically decorrelate across-
channel envelopes by delaying the temporal envelope of a
band of noise centered on the signal (signal band) compared
to two ﬂanking bands.
A. Stimuli and procedure
Noise samples used in Experiment I (G and CM, 8 kHz
bandwidth) were bandpass ﬁltered (12th order Butterworth
ﬁlter, MATLAB R2007a) into three bands: A signal band (9.5
to 10.5 kHz) and two ﬂanking bands (6 to 9 kHz and 11 to
14 kHz). The signal band was chosen to be 1 kHz wide,
reﬂecting the processing transition seen in Figs. 2and 3,
which is assumed to represent the width of the auditory ﬁlter
centered at 10 kHz. The signal band relative to the ﬂanking
bands was delayed in time, with the delays ranging between
0 and 1000 ms (Fig. 5). Delays were restricted by the recip-
rocal of the sampling rate (44.1 kS/s). Therefore, actual
delays were integer multiples of 22.67 ls. The individual
noise bands were then digitally added together and saved
as.wav ﬁles. The 10 kHz signal and all threshold estimation
procedures were identical to Experiment 1. Dolphin SAY
participated in this experiment.
As predicted, the phase delay of the signal band had no
effect on thresholds for G noise (Fig. 6). However, the phase
delay did have the effect of disrupting CMR for CM noise.
Thresholds were relatively ﬂat for delays less than 2 ms and
then increased from 2 to 10 ms at a rate of 0.85 dB/ms (Fig.
7). After 10 ms, the disruption to CMR was complete.
This experiment was repeated with a second dolphin
(APR). However, only two conditions were chosen (0 and
1000 ms delays) to serve as a parsimonious test to determine
whether across-channel coherence was a mechanism that
inﬂuenced masked thresholds in a greater variety of noise
types. In addition to G noise, which served as a baseline con-
dition, the three noise types (from Experiment I) that
produce signiﬁcantly different masked thresholds were used
(CM noise, SS, and IS). Thresholds for each noise condition
are plotted in Fig. 8. Each bar represents the average of four
thresholds and error bars represent standard deviations.
FIG. 4. CRs for the seven noise types. (CM ¼comodulated, SS ¼snapping
shrimp, RN ¼rain, G ¼Gaussian, PS ¼pile saw, BT ¼boat noise,
IS ¼ice squeaks).
FIG. 5. (Color online) Spectrograms of masking noise used in Experiment
II. Maskers were composed of a signal band (9.5 to 10.5 kHz) and two ﬂank-
ing bands (6 to 9 kHz and 11 to 14 kHz). (A) Masking noise with no delay
(correlated) to the signal band and (B) the same masking noise with the sig-
nal band delayed (decorrelated) by 10 ms.
FIG. 6. Thresholds as a function of the phase delay of the signal band in G
noise and CM. Each data point represents the average of four thresholds and
error bars represent standard deviations. The phase delay had no effect on
Gaussian noise. Thresholds for CM noise increased with phase delays
between 2 and 10 ms.
1814 J. Acoust. Soc. Am., Vol. 133, No. 3, March 2013 Branstetter et al.: Dolphin auditory masking
There was a signiﬁcant difference between the 0 ms delay
and 1000 ms delay for both CM and SS noise. Delaying the
signal band did not affect thresholds for G or IS noise.
V. EXPERIMENT III. EFFECTS OF WITHIN-CHANNEL
PROCESSING ON MASKED SIGNAL THRESHOLDS
In Fig. 6, the difference in thresholds between G and
CM noise with no delay was greater than 17 dB. This differ-
ence can be expressed as
where CMR is comodulation masking release and th
are masked thresholds for G and CM noise, respec-
tively. By decorelating the envelopes across channels, a dis-
ruption in the CMR effect was demonstrated but the average
effect between the two dolphins was only 6.8 dB. This sug-
gests that at least one mechanism in addition to across-
channel comparisons is responsible for the remaining release
from masking. One hypothesis is that the listener could
exploit “quiet” temporal loci within the amplitude modulated
noise in order to listen for the presence of the tonal signal.
This strategy is often referred to as “valley” listening or
“dip” listening (Buus, 1985). To test this hypothesis, the
depth of AM of CM noise was adjusted between 0% and
100% modulation depth. At greater modulation depths, the
valleys are relatively “deep,” which should facilitate a
within-valley listening strategy. However, when the depth of
modulation decreases, the CM noise will begin to approxi-
mate G noise and thresholds should increase. To prevent the
dolphins from using the across-channel comparison mecha-
nism from Experiment II, the noise was bandpass ﬁltered to
a bandwidth of 1 kHz. This presumably is the approximate
bandwidth of the auditory ﬁlter centered at 10 kHz, and in
theory restricts any masking processes to an auditory ﬁlter
centered on the signal.
A. Stimuli and procedure
CM noise was created in which the depth of AM was an
independent variable. To accomplish this, CM noise was ﬁrst
synthesized by multiplying G noise by low-pass ﬁltered
noise (5th order Butterworth low-pass ﬁlter with a cutoff fre-
quency of 100 Hz; MATLAB 2007, Mathworks, Natick, MA).
Next, the Hilbert envelope was extracted from the CM noise
where h(t) is the Hilbert transform of the CM noise wave-
form f(t). The depth of AM was scaled and then multiplied
by G noise by
where ais AM depth and n(t) was G noise, and CM
resulting CM noise with a speciﬁed AM depth. Different
noise ﬁles were created with different AM depths between 0
and 1. Each ﬁle had only one modulation depth. CM
then bandpass ﬁltered (12th order Butterworth bandpass ﬁl-
ter with start and stop frequencies of 9.5 and 10.5 kHz,
respectively). This bandwidth was chosen to represent the
approximate width of a dolphin critical band centered at
10 kHz (Branstetter and Finneran, 2008). All noise ﬁles had
a sampling rate of 44.1 kS/s and were 3 min in duration. The
signal for this study was the same 10 kHz tone used in the
previous experiments. Calibrations and threshold estimation
procedures were identical to Experiment I. Dolphins APR
and SAY participated in this experiment.
Threshold patterns for CM noise with AM depths rang-
ing between 0 and 0.9 were relatively ﬂat (Fig. 9). However,
AM depths above 0.9 resulted in a threshold decrease of
5.5 and 7.3 dB for dolphins APR and SAY, respectively.
The average release from masking for both dolphins was
6.4 dB. Although the masking patterns for both dolphins are
similar, thresholds for SAY are on average 5.8 dB lower than
FIG. 7. Data from Fig. 5with a linear ﬁt to the increasing segment of the
threshold function. Thresholds increased at a rate of 0.85 dB/ms.
FIG. 8. Signal thresholds for the phase delay experiment with different noise
types. The noise types were G, CM, SS, and IS, and the numbers that follow
the abbreviations (0 and 1) represent either 0 s delay or 1 s delay of the sig-
nal band relative to the ﬂanking band. For example, CM1 represents como-
dulated noise with a 1 s delay to the signal band relative to the ﬂanking
band. The p-values are shown for CM noise and SS noise in which the phase
delay resulted in signiﬁcant differences in thresholds.
J. Acoust. Soc. Am., Vol. 133, No. 3, March 2013 Branstetter et al.: Dolphin auditory masking 1815
VI. GENERAL DISCUSSION
A. Masking patterns
The PSM predicts that thresholds increase as a function
of noise bandwidth but only up to a “critical-bandwidth” that
reﬂects the limits of a hypothetical auditory ﬁlter centered on
the signal. Spectral components of noise that fall outside of
the auditory ﬁlter should have no effect on signal detectabil-
ity. In Fig. 2, making patterns for RN, G, PS, and BT noise
types are consistent with the PSM. For these noise types, there
was a general threshold increase up to a 1 kHz bandwidth, and
then threshold patterns were generally asymptotic. The
remaining noise types also display the general trend of a
threshold increase up to 1 kHz. However, beyond 1 kHz, CM
and SS thresholds decrease while IS thresholds continue to
increase. When noise exceeds the width of an auditory ﬁlter,
additional mechanisms are required to describe the resulting
masking patterns for CM, SS, and IS noise types. Like human
listeners, the ability of the dolphin’s auditory system to com-
pare or integrate time-domain information across widely
spaced auditory ﬁlters appears to be one of the mechanisms
that leads to CMR. The masking pattern for IS noise is unlike
the other noise types in that thresholds displayed a monotonic
increase as a function of bandwidth well beyond a critical
band. Future experiments are needed to determine the mecha-
nism behind this masking pattern.
When noise is AM and coherent across frequency chan-
nels, CMR occurs (Hall et al., 1984). The speciﬁc conditions
that govern CMR in dolphins are as follows.
1. CMR increases as a function of bandwidth
When CM noise bandwidth exceeds a critical band,
thresholds decrease. More total noise power results in less
masking. This result may appear counterintuitive. However,
if the signal and noise are examined in the time domain in
addition to the spectral domain, the mechanism is less
elusive. A simple model of the mammalian auditory periph-
ery system consists of a bank of auditory ﬁlters (Finneran
et al., 2002) followed by half-wave rectiﬁcation and low-
pass ﬁltering (Viemeister, 1979;Berg, 1996;Branstetter
et al., 2007). This model is often referred to as a leaky inte-
grator or envelope detector (Viemeister, 1979). For CM
noise wider than a single auditory channel, the output of the
model will be multiple correlated envelopes. Adding a sinu-
soid to the signal channel will result in a decorrelated enve-
lope which may cue the presence of a signal. This
comparison can only be made if spectral components of
noise fall into two or more auditory ﬁlters and if the noise is
coherently modulated across frequency regions. As the band-
width of CM noise increases, more comparisons can be
made between the signal channel and additional ﬂanking
channels. Presumably, more comparisons result in increased
conﬁrmation that a signal is present, which then leads to
lower thresholds. In the case of G noise, which lacks any co-
herence across frequency regions, this mechanism cannot be
employed. As a result, signal thresholds masked by G noise
asymptote when noise bandwidths exceed a critical band.
2. Temporal resolution affects CMR
Temporal envelope comparisons appear to form the ba-
sis for CMR. Furthermore, the temporal resolution of the
dolphin’s auditory system governs the ﬁdelity of the physio-
logical representation of the temporal envelope. Although
hair cells on the basilar membrane are relatively fast, 8th
nerve ﬁbers are sluggish in comparison (Rhode and Smith,
1986). The result is that high frequency AM rates may not
be faithfully represented at higher levels of the auditory sys-
tem. Results from Branstetter and Finneran (2008) demon-
strate that CMR is greatest for CM noise with lower AM
rates, and this is likely due to the auditory system’s inability
to resolve high frequency AM rates. The temporal modula-
tion transfer function (TMTF) in human listeners has been
shown to be related to CMR (Berg, 1996). No psychophysi-
cal equivalent of a TMTF has been measured for dolphins,
although modulation rate transfer functions using evoked
potentials display a similar low-pass feature.
In Experiment II, delaying the signal band for G noise
had no effect (Fig. 6) because the bands lack any across-
channel correlation at any delay time. For CM noise, phase
differences less than 2 ms had little effect on thresholds,
which may be due to an unresolvable time difference. How-
ever, a disruption to CMR occurs for phase delays greater
than 2 ms and the effect is complete for delays greater than
10 ms (Figs. 6and 7). The neural circuitry for this effect
might resemble a coincidence detector (Jeffress, 1948)
where outputs from each auditory ﬁlter (envelope detector)
converge on a common target. When inputs are in phase,
summation will occur and activate the target. Target activa-
tion codes that the envelopes are coherent and are likely
from the same sound source. When the envelopes are out of
phase, summation will not occur. Experiments from the ﬁeld
of auditory scene analysis demonstrate that the human audi-
tory system segregates and groups sounds based on similar
acoustic features into sound events that can be related to
FIG. 9. Thresholds as a function of modulation depth. The masking pattern
for both dolphins display fairly ﬂat thresholds up to a modulation depth of
0.9, at which point thresholds decrease with increased modulation depth.
1816 J. Acoust. Soc. Am., Vol. 133, No. 3, March 2013 Branstetter et al.: Dolphin auditory masking
individual environmental sources (Bregman, 1990). One
acoustic feature used for grouping is the AM rate (Bregman,
1990; p. 261). In natural auditory scenes, frequency regions
that have a common AM pattern are likely generated from
the same source, and are thus grouped together (Hall et al.,
1984). Results from Experiment II are consistent with this
interpretation and a coincidence detector may be the neural
mechanism for grouping. The addition of a sinusoidal tone
to the signal band decorrelates the signal band from the
ﬂanking bands, thus inhibiting neural summation. Even a rel-
atively low amplitude sinusoid could have this effect, which
may explain why thresholds in CM noise are relatively low.
Although such a mechanism is speculative, similar circuitry
has been found within animal auditory systems (Carr and
C. Within-channel masking
Results from Experiment III demonstrate that a release
from masking can occur in narrow band noise that is ampli-
tude modulated. Some authors have argued that this release
from masking is not true CMR because no across-channel
comparison is made (Schooneveldt and Moore, 1989). The
dip-listening mechanism, in which the auditory system
exploits temporal valleys where the signal-to-noise ratio is
small, is consistent with results from Experiment III. Large
modulation depths result in deep valleys and thus lower
thresholds. Wider valleys, which would result from lower
AM rates, should also result in lower thresholds.
Total MR in AM noise may be the cumulative effect of
across-channel and within-channel mechanisms (Table II).
Although the numbers do not add up for each individual dol-
phin, the average across-channel difference (6.8 dB) and
within-channel difference (6.4 dB) for both dolphins, when
added together, are less than 1 dB different from the average
D. Implications and future directions
Predictions of auditory masking based on the PSM or
metrics related to the noise spectrum (e.g., CRs, spectral
density, 1/3 octave noise measurements) may result in a
large amount of error. Different noise types in the current
study result in threshold differences within a 22 dB range
(Fig. 4). A more comprehensive model of auditory masking
that incorporates both spectral-domain and time-domain
metrics is needed. Detection of an auditory event is a prereq-
uisite for more complex auditory processing such as discrim-
ination and recognition abilities. Dolphins are hypothesized
to recognize stereotyped whistles and associate them with
speciﬁc group members (Janik et al., 2006). Whistle con-
tours and harmonic structures may also aid in maintaining
group cohesion (Lammers and Au, 2003). To accomplish
these tasks, whistle detection is not enough. The animal
must recognize the sound and associate it with something
meaningful if the sound is going to provide any advantage to
the individual. How noise affects recognition of sounds has
not been studied in odontocetes. Finally, the experiments
presented here represent a worst case auditory masking sce-
nario. The signal and noise were assumed to propagate from
the same location. In real auditory scenes, sound sources are
often separated in space, which leads to a spatial release
from masking (Holt and Schusterman, 2007).
We would like to thank the staff and interns of the U.S.
Navy Marine Mammal Program for their support. We also
thank Marc O. Lammers and Jennifer Miksis-Olds for gener-
ously providing us with ﬁeld recordings of natural noise. Finan-
cial support was provided by the Ofﬁce of Naval Research.
1. Source notes
1. A recording was made in San Diego Bay with a Reson
TC4032 transducer coupled to a Reson VP 1000 pream-
pliﬁer and a high pass ﬁlter of 100 Hz. Recordings were
digitized and saved as .wav ﬁles on a M-audio Microtrack
II. Boat noise was produced by dual outboard Evenrude
175 motors at half throttle. The hydrophone was posi-
tioned 5 m from the sound source in front of the boat,
which was moored to a dock.
2. A recording was made in San Diego Bay with a Reson
TC4032 transducer coupled to a Reson VP 1000 preampli-
ﬁer and a bandpass ﬁlter of 100 Hz to 100 kHz. The hydro-
phone was approximately 10 m from the sound source.
Recordings were digitized by a National Instruments PCI-
MIO-16 E-1 multifunction board (National Instruments,
Austin, TX) and saved as binary ﬁles and later converted
to a .wav ﬁle using MATLAB 2007a.
3. A recording was made with a custom hydrophone (re-
ceiver response 160 dB re: 1 mV/Pa) with a ﬂat response
between (100 Hz and 50 kHz). This recording was
acquired by Jennifer Miksis-Olds in the Arctic Ocean off
the coast of Alaska.
4. Recordings were made with an ecological acoustic recorder
(EAR) (Lammers et al., 2008) in 2.5 m depth. Recordings
were made in the Kure Atoll by Marc Lammers.
5. These .wav ﬁles were the same ﬁles used in Branstetter
and Finneran (2008).
Au, W. W. L., and Moore, P. W. B. (1990). “Critical ratio and critical band-
width for the Atlantic bottlenose dolphin,” J. Acoust. Soc. Am. 88, 1635–
TABLE II. Mechanisms that contribute to MR. Total MR represents thresh-
old differences between broadband G noise and CM noise (8 kHz band-
width, 95 dB re: 1 lPa
/Hz). Across-channel thresholds represent threshold
differences between 0 and 1000 ms delays from Experiment II, and within-
channel thresholds represent threshold differences between modulation
depths of 0 and 100 from Experiment III. Total MR can be loosely approxi-
mated by combining across-channel and within-channel threshold
Dolphin Total MR Across-channel Within channel
APR 15.0 5.5 5.5
SAY 12.8 8.0 7.3
avg 13.9 6.8 6.4
J. Acoust. Soc. Am., Vol. 133, No. 3, March 2013 Branstetter et al.: Dolphin auditory masking 1817
Berg, B. G. (1996). “On the relationship between comodulation masking
release and temporal modulation transfer functions,” J. Acoust. Soc. Am.
Branstetter, B. K., and Finneran, J. J. (2008). “Comodulation masking
release in bottlenose dolphins (Tursiops truncatus),” J. Acoust. Soc. Am.
Branstetter, B. K., Mercado, E. III, and Au, W. W. L. (2007). “Representing
multiple discrimination cues in a computational model of the bottlenose
dolphin auditory system,” J. Acoust. Soc. Am. 122, 2459–2468.
Bregman, A. S. (1990). Auditory Scene Analysis: The Perceptual Organiza-
tion of Sound (The MIT Press, Cambridge, MA), pp. 261–292.
Buus, S. (1985). “Release from masking caused by envelope ﬂuctuations,”
J. Acoust. Soc. Am. 78, 1958–1965.
Carr, C. E., and Konishi, M. (1988). “Axonal delay lines for time measure-
ment in the owl’s brainstem,” Proc. Natl. Acad. Sci. U.S.A. 85, 8311–
Clark, C. W., Ellison, W. T., Southall, B. L., Hatch, L., Van Parijs, S. M.,
Frankel, A., and Ponirakis, D. (2009). “Acoustic masking in marine eco-
systems: intuitions, analysis, and implication,” Mar. Ecol.: Prog. Ser. 395,
Erbe, C. (2008). “Critical ratios of beluga whales (Delphinapterus leucas)
and masked signal duration,” J. Acoust. Soc. Am. 124, 2216–2223.
Finneran, J. J., Schlundt, C. E., Carder, D. A., and Ridgway, S. H. (2002).
“Auditory ﬁlter shapes for the bottlenose dolphin (Tursiops truncatus) and
the white whale (Delphinapterus leucas) derived with notched noise,”
J. Acoust. Soc. Am. 112, 322–328.
Fletcher, H. (1940). “Auditory patterns,” Rev. Mod. Phys. 12, 47–65.
Gellerman, L. W. (1933). “Chance orders of alternating stimuli in visual
discrimination experiments,” J. Genet. Psychol. 42, 206–208.
Greenwood, D. D. (1990). “A cochlear frequency-position function for
several species—29 years later,” J. Acoust. Soc. Am. 87, 2592–2605.
Hall, J. W., and Grose, J. H. (1988). “Comodulation masking release:
Evidence for multiple cues,” J. Acoust. Soc. Am. 84, 1669–1675.
Hall, J. W., Haggard, M. P., and Fernandes, M. A. (1984). “Detection
in noise by spectro-temporal pattern analysis,” J. Acoust. Soc. Am. 76,
Holt, M. M., and Schusterman, R. J. (2007). “Spatial release from masking
of aerial tones in pinnipeds,” J. Acoust. Soc. Am. 121, 1219–1225.
Houser, D. S., and Finneran, J. J. (2006). “Variation in the hearing sensitiv-
ity of a dolphin population obtained through the use of evoked potential
audiometry,” J. Acoust. Soc. Am. 120, 4090–4099.
Janik, V. M., Sayigh, L. S., and Wells, R. S. (2006). “Signature whistle
shape conveys identity information to bottlenose dolphins,” Proc. Natl.
Acad. Sci. U.S.A. 103, 8293–8297.
Jeffress, L. A. (1948). “A place theory of sound localization,” J. Comp.
Physiol. Psychol. 41, 35–39.
Johnson, C. S. (1968). “Masked tonal thresholds in the bottlenosed por-
poise,” J. Acoust. Soc. Am. 44, 965–967.
Klump, G. M., Kittel, M., and Wagner, E. (2001). “Comodulation masking
release in the Mongolian gerbil,” in Twenty-Fourth Midwinter Research Meet-
ing of the Association for Research in Otolaryngology (St. Petersburg, FL).
Lammers, M. O., and Au, W. W. L. (2003). “Directionality in the whistles
of Hawaiian spinner dolphins (Stenella longirostris): A signal feature to
cue direction of movement?,” Marine Mammal Sci. 19, 249–264.
Lammers, M. O., Brainard, R. E., Au, W. W. L., Mooney, T. A., and Wong,
K. B. (2008). “An ecological acoustic recorder (EAR) for long-term moni-
toring of biological and anthropogenic sounds on coral reefs and other ma-
rine habitats,” J. Acoust. Soc. Am. 123, 1720–1728.
Levitt, H. (1971). “Transformed up-down methods in psychoacoustics,” J.
Acoust. Soc. Am. 49, 467–477.
McDonald, M. A., Hildebrand, J. A., and Wiggins, S. M. (2006). “Increases
in deep ocean ambient noise in the Northeast Paciﬁc west of San Nicolas
Island, California,” J. Acoust. Soc. Am. 120, 711–718.
Moore, B. C. J. (1993). “Frequency analysis and pitch perception,” in
Human Psychophysics, edited by W. A. Yost, A. N. Popper, and R. R. Fay
(Springer-Verlag, New York), pp. 56–115.
Pressnitzer, D., Meddis, R., and Winter, I. M. (2001). “Physiological corre-
lates of comodulation masking release in the mammalian ventral cochlear
nucleus,” J. Neurosci. 21(16), 6377–6386.
Rhode, W. S., and Smith, P. H. (1986). “Encoding timing and intensity in
the ventral cochlear nucleus of the cat,” J. Neurophysiol. 56, 261–286.
Schooneveldt, G. P., and Moore, B. C. J. (1989). “Comodulation masking
release (CMR) as a function of masker bandwidth, modulator bandwidth,
and signal duration,” J. Acoust. Soc. Am. 85, 273–281.
Southall, B. L., Schusterman, R. J., and Kastak, D. (2003). “Auditory mask-
ing in three pinnipeds: Aerial critical ratios and direct critical bandwidth
measurements,” J. Acoust. Soc. Am. 114, 1660–1666.
Trickey, J. S., Branstetter, B. B., and Finneran, J. J. (2011). “Auditory mask-
ing with environmental, comodulated, and Gaussian noise in bottlenose
dolphins (Tursiops truncatus),” J. Acoust. Soc. Am. 128, 3799–3804.
Viemeister, N. F. (1979). “Temporal modulation transfer functions based on
modulation thresholds,” J. Acoust. Soc. Am. 66, 1364–1380.
1818 J. Acoust. Soc. Am., Vol. 133, No. 3, March 2013 Branstetter et al.: Dolphin auditory masking