Spatially separating language masker from target results in spatial and linguistic
masking release
Navin Viswanathan, Kostas Kokkinakis, and Brittany T. Williams
Citation: The Journal of the Acoustical Society of America 140, EL465 (2016);
View online: https://doi.org/10.1121/1.4968034
View Table of Contents: http://asa.scitation.org/toc/jas/140/6
Published by the Acoustical Society of America
Articles you may be interested in
 Noise control zone for a periodic ducted Helmholtz resonator system
The Journal of the Acoustical Society of America 140, EL471 (2016); 10.1121/1.4968530
 The role of stress and word size in Spanish speech segmentation
The Journal of the Acoustical Society of America 140, EL484 (2016); 10.1121/1.4971227
 Perceptually salient spectrotemporal modulations for recognition of sustained musical instruments
The Journal of the Acoustical Society of America 140, EL478 (2016); 10.1121/1.4971204
 The role of periodicity in perceiving speech in quiet and in background noise
The Journal of the Acoustical Society of America 138, 3586 (2015); 10.1121/1.4936945
 Consistent sound change between stops and affricates in Seoul Korean within and across individuals: A
diachronic investigation
The Journal of the Acoustical Society of America 140, EL491 (2016); 10.1121/1.4971203
 Flux projection beamforming for monochromatic source localization in enclosed space
The Journal of the Acoustical Society of America 141, EL1 (2017); 10.1121/1.4973193
Spatially separating language masker from target
results in spatial and linguistic masking release
Navin Viswanathan,a) Kostas Kokkinakis, and Brittany T. Williams
Speech-Language-Hearing: Sciences and Disorders, University of Kansas, 1000 Sunnyside
Avenue, Lawrence, Kansas 66045, USA
navin@ku.edu, kokkinak@ku.edu, btw@ku.edu
Abstract: Several studies demonstrate that in complex auditory scenes,
speech recognition is improved when the competing background and
target speech differ linguistically. However, such studies typically utilize
spatially co-located speech sources which may not fully capture typical
listening conditions. Furthermore, co-located presentation may overes-
timate the observed benefit of linguistic dissimilarity. The current study
examines the effect of spatial separation on linguistic release from mask-
ing. Results demonstrate that linguistic release from masking does
extend to spatially separated sources. The overall magnitude of the
observed effect, however, appears to be diminished relative to the co-
located presentation conditions.
VC 2016 Acoustical Society of America
[RS]
Date Received: September 20, 2016 Date Accepted: November 4, 2016
1. Introduction
Human listeners are routinely faced with the task of recognizing speech in the presence
of competing background speech. Speech perception under such conditions is affected
primarily by a combination of energetic masking (the reduction in intelligibility due to
the spectro-temporal overlap of the target and background masker) and informational
masking (reduction in intelligibility that occurs due to a combination of factors that
reduces intelligibility after the effect of energetic masking has been accounted for).
Specifically, in speech-in-speech recognition tasks, listeners often experience a boost in
performance when the background speech differs along specific linguistic dimensions
from the target speech (e.g., Calandruccio et al., 2010). While different terms such as
(same) language interference effect (Cooke et al., 2008; Mattys et al., 2009) and target-
masker language mismatch effect (e.g., Brouwer et al., 2012) have been previously used
to characterize this effect, we opt to use the term linguistic release from masking
(LRM; similar to Calandruccio et al., 2016) to describe the performance improvement
resulting from the target-masker linguistic mismatch. There are two primary reasons
LRM may occur. First, LRM may occur because target and background speech differ
in their acoustic-phonetic characteristics as in the case of different languages or differ-
ent accents. This is because, relative to same language maskers, both energetic masking
effects (due to lower spectro-temporal overlap between the speech sources) and infor-
mational masking effects (due to ease of perceptual segregation) are reduced. Second,
LRM may reflect decreased semantic interference due to the lack of intelligibility of
the competing background and thus a reduction of informational masking.
A number of previous studies have described a robust LRM benefit across dif-
ferent target-masker language pairs (e.g., Brouwer et al., 2012), although the detection
of this effect may depend on specific task demands placed on the listener (Mattys
et al., 2009). Furthermore, other recent studies, have shown that the similarity in the
linguistic properties of the target and background languages is an important factor
toward determining the prevalence of LRM benefits (Calandruccio et al., 2010;
Brouwer et al., 2012). Interestingly, even in cases, where the background masker con-
sists of accented versions of the target language, listeners show a reliable LRM benefit
which is modulated by the intelligibility of the background accented speech
(Calandruccio et al., 2010). The aforementioned studies collectively indicate that LRM
is a robust phenomenon that is, in part, determined by the target-masker linguistic sim-
ilarity. Despite this strong evidence, it is unclear whether under more ecologically typi-
cal scenarios, listeners would actually experience LRM. Most experimental studies
assessing linguistic release have used spatially co-located (i.e., originating from the
a)Author to whom correspondence should be addressed.
J. Acoust. Soc. Am. 140 (6), December 2016 VC 2016 Acoustical Society of America EL465
Viswanathan et al.: JASA Express Letters [http://dx.doi.org/10.1121/1.4968034] Published Online 1 December 2016
same spatial location) target-masker conditions [cf. Freyman et al. (2001) used
simulated spatial separation] that are ecologically atypical and may overestimate the
overall LRM benefit because they do not permit source segregation based on
spatial information and forces the listener to rely exclusively on acoustic-phonetic
differences.
The effects of spatial-separation of target and masker on speech recognition is
well-documented. For instance, it is clear that actual physical differences, or even per-
ceived differences, between the individual spatial locations of a target and a masker
result in a substantial increase of speech intelligibility compared to situations where the
target and masking speech are co-located. This benefit, known as spatial release from
masking (SRM), describes the reduction in masking that occurs in the presence of an
active masker due to the availability of spatial acoustic cues. This type of masking
release is fairly robust in normal-hearing listeners (e.g., Freyman et al., 2001) and can
produce substantial differences in speech perception.
As noted earlier, whether LRM occurs when the target and the masker are
spatially separated remains to be established. In fact, the findings of Freyman et al.
(2001) offer a reason to doubt whether LRM persists under spatial separation. In
experiment 3, the authors investigated whether SRM depended on the intelligibility of
background language. They tested monolingual English listeners using speech back-
grounds produced by a bilingual Dutch-English speaker that were either produced in
Dutch or Dutch-accented English. Critically, these backgrounds were presented either
with or without simulated spatial separation. Their results indicated clear SRM effects
in both language conditions. However, because they were not focused on LRM, they
did not evaluate whether Dutch maskers produced better performance than English or
the effect of spatial separation on the resulting LRM. A visual inspection of their
results (see Freyman et al., 2001, Fig. 9, p. 2120) suggests a clear LRM for higher sig-
nal-to-noise ratios (SNRs) when the speech sources were co-located. However, there
appears to be no LRM for the spatially displaced condition.
Motivated by these findings and by the general observation that in typical lis-
tening situations multiple masking sources often originate from different locations
around a listener, we argue that a specific examination of whether LRM effects extend
to spatially displaced speech sources is warranted. In the present study, similar to
Freyman et al. (2001), we used Dutch language backgrounds to examine whether lis-
teners can benefit from LRM under three spatial configurations. Specifically, we maxi-
mized the spatial separation by using spatially displaced conditions to account for any
asymmetry in addition to a spatially co-located condition. Our rationale is that, by
using relatively similar languages (thereby minimizing LRM by minimizing acoustic-
phonetic differences; see Calandruccio et al., 2013) under conditions of maximal spatial
separation (thereby maximizing SRM), we provide the strongest test of whether LRM
effects persist under spatially separated listening conditions.
2. Methods
2.1 Participants
Forty-two undergraduate students from the University of Kansas with American
English as their first language and no exposure to Dutch participated in this study for
course credit. All subjects reported no history of speech or hearing disorders. Median
age of subjects was 19 years (ages from 18 to 21 years).
2.2 Materials
Three female, native English speakers produced the English stimuli. The target stimuli
consisted of English sentences taken from the revised Bamford-Kowal-Bench (BKB-R)
Standard Sentence Test (Bench et al., 1979). Each sentence contained three or four
keywords (e.g., The CLOWN had a FUNNY FACE). Following Brouwer et al. (2012),
the English background stimuli consisted of syntactically well-formed, semantically
simple sentences that were produced by the two speakers who did not record the tar-
get sentences. Translated versions were used to elicit the Dutch background stimuli
that were recorded by a pair of female, Dutch, native speakers. All stimuli were
recorded in a sound-attenuating booth at a 44.1 kHz sampling rate and a 16-bit
resolution.
Two-talker babble maskers were created by combining waveforms of each pair
of speakers for each language. Following Brouwer et al. (2012), we reduced unequal
amounts of energetic masking between the two language conditions by minimizing dif-
ferences in the long-term average speech spectrum (LTASS) of the two background
speech tracks. Pilot testing ensured that the overall amount of spectral manipulation
Viswanathan et al.: JASA Express Letters [http://dx.doi.org/10.1121/1.4968034] Published Online 1 December 2016
EL466 J. Acoust. Soc. Am. 140 (6), December 2016 Viswanathan et al.
was minimal and that the original and normalized files were not easily distinguishable.
Random portions of babbles were excised from this babble to serve as background
stimuli for each trial. The background stimuli always preceded the target onset by 0.5 s
and persisted 0.5 s after the target offset. The root-mean-square (RMS) levels of the
target sentences were all equalized to 66 dB sound pressure level (SPL) and that of the
background maskers were equalized to 71 dB SPL and 76 dB SPL to produce SNRs of
5 dB and 10 dB, respectively. An SNR equal to 5 dB results in a fairly challenging
listening task and produces conditions with strong informational masking (e.g., see
Calandruccio et al., 2010). The 10 dB SNR condition was chosen to increase task dif-
ficulty while simultaneously ensuring that there were no floor effects in task perfor-
mance. Finally, for each target-masker configuration, three different spatial listening
conditions were created. In all conditions, the target was placed in front of the listener
(0), similar to a typical conversational scenario, and the masker originated from either
the front of the listener (0), the right of the listener (þ90) or the left of the listener
(90). To place the target speech in front of the listener and the masker in different
virtual locations in the horizontal plane, we used publicly available pre-recorded head-
related transfer functions (HRTFs). These HRTFs were recorded inside an anechoic
chamber at the University of Oldenburg (Kayser et al., 2009) using behind-the-ear
hearing aid dummy shells placed on the pinnae of an artificial head-and-torso simula-
tor. The left and right ear (binaural) stimuli containing the combined auditory target
and masker streams were generated for each spatial location in MATLAB by convolving
each target and masker stimulus with each pair of the direction-dependent HRTFs
(90, 0, þ90). Subjects in pilot testing were successful in externalizing the auditory
image thus confirming that most perceptually significant acoustic cues were retained in
the HRTFs.
2.3 Procedure
Listeners were tested individually, seated in front of a computer in a sound-isolated
room. Stimuli were presented binaurally through Sennheisser HD-558 headphones
(Sennheiser, Hanover, Germany). Prior to testing, listeners were asked to transcribe
ten sentences presented in quiet conditions to train them to identify the target speaker.
During each experimental trial, one target sentence was presented in conjunction with
the masker. The subjects were instructed to listen to sentences spoken by a female
native English speaker in the presence of two-talker background speech. In addition,
they were informed that the background speech would be perceived from various azi-
muth locations. They were asked to type what they heard from the target speaker and
report individual words if they were unable to identify the whole sentence. They were
only allowed to listen to the sentence once. Each participant listened to 20 sentences
from each of the three spatial locations in two background languages resulting in a
total of 120 sentences. The presentation order of these sentences were completely ran-
domized for each subject. The SNR of the target-background combinations, which
determines the difficulty of listening conditions, was manipulated between subjects. All
other factors were manipulated within subjects.
3. Results
Data from all subjects (N¼ 20 in SNR¼5 dB and N¼ 22 in SNR¼10 dB condi-
tions) were included in the final analyses. Figure 1 depicts intelligibility scores (percent
key words, per the BKB database, correctly reported) by location and masker
language. The left and right panels depict performance in the 5 and 10 dB SNR
conditions, respectively. As evident from Fig. 1, the general pattern of results was
qualitatively similar across both SNRs. As expected, overall performance was lower in
the more difficult, 10 dB SNR condition. Regardless of masker language, perfor-
mance was better when the masker was spatially separated from the target rather than
when co-located suggesting a clear SRM effect. Across spatial conditions, it also
appears that listener performance was better when presented with the Dutch masker
language than with English. This suggests that listeners reliably benefitted from LRM.
To evaluate these trends statistically, inferential analyses were performed. The
percent correct scores were transformed into respective logit values in order to normal-
ize the error variance. In four instances, proportion scores of 1 were replaced by 0.99
to avoid singularities in the transformed data. Data were submitted to a 2 (masker lan-
guage)  3 (location) factorial repeated measures analysis of variance (ANOVA) for
each SNR (which was manipulated between subjects). Table 1 summarizes the results
of this analysis. In both SNRs, the main effects of masker language (p< 0.001) and
location (p< 0.001) were significantly different indicating that listeners experienced a
Viswanathan et al.: JASA Express Letters [http://dx.doi.org/10.1121/1.4968034] Published Online 1 December 2016
J. Acoust. Soc. Am. 140 (6), December 2016 Viswanathan et al. EL467
clear advantage due to both LRM and SRM. There was also a significant interaction
between masker language and location (p< 0.01) suggesting that the amount of LRM
depended on spatial location in both SNRs.
To further tease apart these effects, we explored two specific questions. First,
does LRM persist when the speech sources are spatially separated? To answer this
question, we performed a planned analysis of simple main effects for both SNRs.
Again, in each SNR, the LRM remained reliable for both spatially separated condi-
tions (Bonferroni corrected p< 0.001) confirming the persistence of LRM effects under
spatial separation. Second, we asked how the relationship between LRM and spatial
separation changes between the two SNR conditions. To answer this, for each subject,
and for each masker language, we calculated an average intelligibility score in the spa-
tially separated conditions by averaging the intelligibility scores in the 90 and þ90
location. Following this, we calculated LRM scores by subtracting the intelligibility in
English from the Dutch conditions in both spatial conditions for each subject. We sub-
mitted these LRM scores to a 2 (spatial location: co-located and separated)  2 (SNR:
5 dB and 10 dB) omnibus mixed ANOVA. Figure 2 depicts the average LRM
scores observed for the co-located and the separated spatial locations for 5 dB SNR
(left panel) and 10 dB SNR (right panel). We found a reliable effect for location
[F (1, 39)¼ 89.88, p< 0.001, g2p¼ 0.70] confirming that the LRM was greater in the co-
located condition. Finally, the significant interaction between location and SNR [F (1,
39)¼ 14.92, p< 0.001, g2p¼ 0.28] confirms that in more difficult listening conditions
(Fig. 2, right panel), LRM benefits increase especially under conditions of spatial sepa-
ration. The effect for SNR [F (1, 39)¼ 4.25, p¼ 0.046, g2p¼ 0.01] was significant indi-
cating that in the harder SNR there is slightly more LRM benefit.
4. Discussion
Previous studies that have reported a benefit for speech-in-speech recognition tasks
with language backgrounds have used target-masker pairs that were spatially co-
located. In typical listening situations, listeners often derive a substantial benefit due to
spatial separation in addition to linguistic differences between the sources. In this
study, we investigated whether LRM effects persist even under conditions in which the
target and masker speech are spatially separated. We did so by manipulating both the
Fig. 1. Sentence recognition performance (percent correct) for 42 listeners in the presence of two-talker English
(white) and two-talker Dutch (gray) maskers in the 5 dB (left) and 10 dB (right) SNR conditions. Each open
circle represents the score of a single participant. Boxplots in the two masking conditions (English and Dutch)
are plotted for each spatial location condition (left, center, right). The boxes depict the values between the 25th
and 75th percentiles and the whiskers denote the 61.5 interquartile range. Medians are shown as horizontal
lines.
Table 1. Summary of analyses of variance results for both SNR conditions. All compared effects were signifi-
cant at the p< 0.001 level.
Source SNR¼5 dB SNR¼10 dB
Language F (1, 19)¼ 77.38, p< 0.001, g2p¼ 0.80 F (1, 21)¼ 203.27, p< 0.001, g2p ¼ 0.91
Location F (2, 38)¼ 274.50, p< 0.001, g2p ¼ 0.94 F (2, 42)¼ 369.40, p< 0.001, g2p ¼ 0.95
LanguageLocation F (2, 38)¼ 5.29, p¼ 0.009, g2p¼ 0.22 F (2,42)¼ 36.04, p< 0.001, g2p ¼ 0.63
Viswanathan et al.: JASA Express Letters [http://dx.doi.org/10.1121/1.4968034] Published Online 1 December 2016
EL468 J. Acoust. Soc. Am. 140 (6), December 2016 Viswanathan et al.
language and the spatial location of the background masker relative to the target in
two SNR conditions. Our results, across both SNR conditions, indicated that listeners
demonstrated a reliable increase in performance when the locations of the target and
background speech differed—a clear indication of SRM. Similarly, the listeners’ per-
formance improved when the language of the target and the background was mis-
matched—a clear indication of LRM. Critically, the benefit due to LRM was reliable
across all different spatial locations tested, confirming that this benefit will occur under
typical listening conditions. Finally, follow-up analyses indicated that the size of this
effect is significantly reduced when the sources are spatially displaced. We offer a pos-
sible explanation for this outcome. Co-located presentation produces more challenging
listening conditions, under which the segregability of the language masker is especially
beneficial. This advantage is less useful under spatially separated conditions that
already provide listeners with additional information for segregation. A corollary sug-
gestion is that, in general, spatially displacing the competitors results in a substantially
improved intelligibility of the target in the English masker conditions (see Fig. 2). This
benefit, due to SRM, limits the overall advantage that listeners can experience due to
LRM in less challenging SNRs. Taken together the claim that LRM is modulated by
the overall difficulty of the listening conditions is supported by our finding of more
LRM under spatial separation in the harder SNR (10 dB) condition (also see Van
Engen and Bradlow, 2007). To conclude, our results clearly demonstrate that listeners
are better able to disregard background speech in a foreign language even when it is
spatially separated from the target.
Acknowledgment
This research was supported by NSF grant BCS-1431105 to N.V.
References and links
Bench, J., Kowal, A˚., and Bamford, J. (1979). “The BKB (Bamford-Kowal-Bench) sentence lists for parti-
ally–hearing children,” Br. J. Audiol. 13(3), 108–112.
Fig. 2. The individual benefit due to linguistic release from masking plotted in percentage points for the 42 lis-
teners tested in different spatial locations (co-located vs separated) in the 5 dB (left) and 10 dB (right) SNR
conditions. Each open circle represents the benefit estimated for every individual. The boxes depict the values
between the 25th and 75th percentiles and the whiskers denote the 61.5 interquartile range. Medians are shown
as horizontal lines.
Viswanathan et al.: JASA Express Letters [http://dx.doi.org/10.1121/1.4968034] Published Online 1 December 2016
J. Acoust. Soc. Am. 140 (6), December 2016 Viswanathan et al. EL469
Brouwer, S., Van Engen, K. J., Calandruccio, L., and Bradlow, A. R. (2012). “Linguistic contributions to
speech-on-speech masking for native and non-native listeners: Language familiarity and semantic con-
tent,” J. Acoust. Soc. Am. 131(2), 1449–1464.
Calandruccio, L., Brouwer, S., Van Engen, K. J., Dhar, S., and Bradlow, A. R. (2013). “Masking release
due to linguistic and phonetic dissimilarity between the target and masker speech,” Am. J. Audiol. 22(1),
157–164.
Calandruccio, L., Dhar, S., and Bradlow, A. R. (2010). “Speech-on-speech masking with variable access to
the linguistic content of the masker speech,” J. Acoust. Soc. Am. 128(2), 860–869.
Calandruccio, L., Leibold, L. J., and Buss, E. (2016). “Linguistic masking release in school–age children
and adults,” Am. J. Audiol. 25(1), 34–40.
Cooke, M., Lecumberri, M. G., and Barker, J. (2008). “The foreign language cocktail party problem:
Energetic and informational masking effects in non-native speech perception,” J. Acoust. Soc. Am.
123(1), 414–427.
Freyman, R. L., Balakrishnan, U., and Helfer, K. S. (2001). “Spatial release from informational masking
in speech recognition,” J. Acoust. Soc. Am. 109(5), 2112–2122.
Kayser, H., Ewert, S. D., Anem€uller, J., Rohdenburg, T., Hohmann, V., and Kollmeier, B. (2009).
“Database of multichannel in-ear and behind-the-ear head-related and binaural room impulse
responses,” EURASIP J. Adv. Signal Processing 2009(1), 298605.
Mattys, S. L., Brooks, J., and Cooke, M. (2009). “Recognizing speech under a processing load:
Dissociating energetic from informational factors,” Cognit. Psychol. 59(3), 203–243.
Van Engen, K. J., and Bradlow, A. R. (2007). “Sentence recognition in native- and foreign-language multi-
talker background noise),” J. Acoust. Soc. Am. 121(1), 519–526.
Viswanathan et al.: JASA Express Letters [http://dx.doi.org/10.1121/1.4968034] Published Online 1 December 2016
EL470 J. Acoust. Soc. Am. 140 (6), December 2016 Viswanathan et al.