Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Hearing Research 432 (2023) 108743

Contents lists available at ScienceDirect

Hearing Research
journal homepage: www.elsevier.com/locate/heares

Research Paper

Binaural pre-processing for contralateral sound field attenuation can


improve speech-in-noise intelligibility for bilateral hearing-aid users
Fernando M. San-Victoriano a,b, Almudena Eustaquio-Martín a,b,
Enrique A. Lopez-Poveda a,b,c,∗
a
Laboratorio de Audición Computacional y Psicoacústica, Instituto de Neurociencias de Castilla y León, Universidad de Salamanca, 37007 Salamanca, Spain
b
Grupo de Audiología, Instituto de Investigación Biomédica de Salamanca, Universidad de Salamanca, 37007 Salamanca, Spain
c
Departamento de Cirugía, Facultad de Medicina, Universidad de Salamanca, 37007 Salamanca, Spain

a r t i c l e i n f o a b s t r a c t

Article history: We have recently proposed a binaural sound pre-processing method to attenuate sounds contralateral to
Received 24 October 2022 each ear and shown that it can improve speech intelligibility for normal-hearing (NH) people in sim-
Revised 21 January 2023
ulated “cocktail party” listening situations (Lopez-Poveda et al., 2022, Hear Res 418:108,469). The aim
Accepted 24 March 2023
here was to evaluate if this benefit remains for hearing-impaired listeners when the method is com-
Available online 25 March 2023
bined with two independently functioning hearing aids, one per ear. Twelve volunteers participated in
Keywords: the experiments; five of them had bilateral sensorineural hearing loss and seven were NH listeners with
Noise reduction simulated bilateral conductive hearing loss. Speech reception thresholds (SRTs) for sentences in compe-
Binaural unmasking tition with a source of steady, speech-shaped noise were measured in unilateral and bilateral listening,
Binaural hearing and for (target, masker) azimuthal angles of (0°, 0°), (270°, 45°), and (270°, 90°). Stimuli were processed
Hearing device through a pair of software-based multichannel, fast-acting, wide dynamic range compressors, with and
Beamformer
without binaural pre-processing. For spatially collocated target and masker sources at 0° azimuth, the
pre-processing did not affect SRTs. For spatially separated target and masker sources, the pre-processing
improved SRTs when listening bilaterally (improvements up to 10.7 dB) or unilaterally with the acousti-
cally better ear (improvements up to 13.9 dB), while it worsened SRTs when listening unilaterally with
the acoustically worse ear (decrements of up to 17.0 dB). Results show that binaural pre-processing for
contralateral sound attenuation can improve speech-in-noise intelligibility in laboratory tests also for bi-
lateral hearing-aid users.
© 2023 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/)

1. Introduction Among these strategies are beamformers, i.e., systems that ex-
ploit the physical separation between the target and masker sound
Understanding speech presented in competition with other sources to improve the SNR. Beamformers often favor sounds com-
sounds is still one of the main challenges faced by people with ing from the front of the listener while attenuating interfering
hearing loss (Abrams and Kihm, 2015). Hearing-impaired (HI) lis- sounds from other directions (Kates, 2008). Sometimes, however,
teners require up to 9 dB higher signal-to-noise ratios (SNRs) than the beamformer’s directivity pattern is guided by sound source
people with normal hearing (NH) to achieve the same amount of identification methods, listener eye tracking, or adapts for chang-
intelligibility (e.g., Festen and Plomp, 1990; Peters et al., 1998). For ing listening scenarios (Kollmeier and Kiessling, 2018). These ap-
this reason, hearing aids (HAs) and other hearing devices often in- proaches usually require two or more microphones and a high de-
corporate sound pre-processing methods aimed at improving the mand for computational power to improve the SNR. For HA users,
SNR before the stimulus is further processed and delivered to the unilateral beamformers provide large benefits for sentence recogni-
user. tion in noise compared to omnidirectional approaches while bilat-
eral beamformers provide smaller, but statistically significant ben-
efits over unilateral beamformers (Picou and Ricketts, 2019).

We have recently proposed a binaural pre-processing method
Corresponding author at: Instituto de Neurociencias de Castilla y León, Univer-
sidad de Salamanca, Calle Pintor Fernando Gallego 1, 37007 Salamanca, Spain. that can improve speech recognition with two microphones, one
E-mail address: ealopezpoveda@usal.es (E.A. Lopez-Poveda). per ear. The method involves linear subtraction of the weighted

https://doi.org/10.1016/j.heares.2023.108743
0378-5955/© 2023 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
F.M. San-Victoriano, A. Eustaquio-Martín and E.A. Lopez-Poveda Hearing Research 432 (2023) 108743

contralateral sound spectrum from the spectrum at each ear with


weights equal to the ratio of ipsilateral to contralateral head- R ( f ) = R ( f ) − α ( f ) · L ( f ) (2)
related transfer functions (HRTF) averaged over an appropriate az-
imuth range. Instead of favoring sounds in front of the listener, our where L( f ) and R( f ) denote the complex frequency spectra of the
method favors sounds ipsilateral to each ear by attenuating con- unprocessed stimuli reaching the left and right ear, respectively;
tralateral sounds (Lopez-Poveda et al., 2022). Because in free-field, L ( f ) and R ( f ) denote the corresponding complex frequency spec-
multi-talker scenarios the target sound source is unlikely centered tra of the processed stimuli at the left and right ear, respectively;
(Grange and Culling, 2016), the method can improve the SNR in and α ( f ) and β ( f ) are the subtraction weights. The weights are
the ear closer to the target. And because listeners can switch at- also complex numbers, i.e., with magnitude and phase. Lopez-
tention to the ear with the better SNR for the momentary target Poveda et al. (2022) tested the pre-processing method for NH lis-
(e.g., Bronkhorst and Plomp, 1988), the method can improve intelli- teners using β0◦ −180◦ and α180◦ −360◦ calculated as follows:
gibility in ‘cocktail party’ scenarios. We showed that, for NH listen- 180
θ =0 Hθ L ( f )
ers tested in simulated cocktail party situations, the algorithm can β0◦ −180◦ ( f ) = 180 (3)
improve the SNR at 50% sentence recognition (termed the speech θ =0 Hθ R ( f )
reception threshold or SRT) by up to 9.6 dB for speech in compe- 360
Hθ R ( f )
tition with up to three masker sources, depending on the masker α180◦ −360◦ ( f ) = θ360
=180
(4)
type and location. θ =180 Hθ L ( f )
It remains to be shown, however, whether the benefits
where θ denotes azimuthal angle, going from 0° for a source right
found for NH listeners also apply to HA users. Because our
in front to 360° clockwise, and Hθ L ( f ) and Hθ R ( f ) are the complex
method attenuates contralateral sounds and nonlinear (compres-
HRTF for the left and right ear for a KEMAR (Gardner and Mar-
sive) HAs amplify low-level sounds more than high-level sounds
tin, 1995). Here, the subtraction weights β and α were calculated
(Edwards, 2003), in theory, HAs could partly undo the contralat-
by integrating KEMAR HRTFs over the azimuth range 60°−120° and
eral sound attenuation produced by our method, resulting in less
240°−300°, respectively, as follows:
(or no) benefit for HA users than for NH listeners.
120
The aim of the present study was to investigate the speech- θ =60 Hθ L ( f )
in-noise intelligibility benefits of our pre-processing method for
β60◦ −120◦ ( f ) = 120 (5)
θ =60 Hθ R ( f )
HI listeners when the method is used together with two inde-
300
pendently functioning HAs, one per ear. Aided SRTs for sentences Hθ R ( f )
in competition with a source of speech-shaped noise (SSN) were α240◦ −300◦ ( f ) = θ300
=240
(6)
compared for pre-processed and unprocessed stimuli in various θ =240 Hθ L ( f )
simulated spatial configurations of the target and masker sources The integration over a narrower angle than used by Lopez-
and in various listening modes, i.e., listening binaurally or listen- Poveda et al. (2022) was intended to attenuate a wider range
ing unilaterally with either ear. Because the aim was to investigate of contralateral frequencies, i.e., with the present weights the di-
if contralateral sound attenuation can help HA users overall (i.e., rectivity pattern for the left ear was as shown in Fig. 2C rather
regardless of hearing loss etiology and magnitude), participants in- than that in Fig. 2A of Lopez-Poveda et al. (2022). Therefore,
cluded five people with bilateral sensorineural hearing loss (SNHL) as illustrated in supplementary Fig. S1, we expected that the
and seven people with simulated bilateral conductive hearing loss pre-processing would improve intelligibility over slightly different
(NH listeners fitted with foam earplugs). target-masker spatial configurations here than reported by Lopez-
Poveda et al. (2022).
Hearing aid. To compensate for their hearing loss, all par-
2. Methods ticipants were aided in their two ears by a software-based im-
plementation of a 12-channel, dynamic range compressor (1–
Procedures were approved by the Ethics Committee of the Uni- 80 0 0 Hz). The compressor was of the type known as “return-
versity of Salamanca. Unless otherwise stated, methods were as in to-zero” (Giannoulis et al., 2012). The HA was programmed with
Lopez-Poveda et al. (2022), where details can be found. For con- attack and release times of 5 and 200 ms, respectively. These
ciseness, only a summary is given here. time constants were chosen to produce a moderately fast-acting
Participants. Participants were volunteers and not paid for gain control (Moore, 2008). The gain in each frequency channel of
their time. All of them signed an informed consent to participate the HA was adjusted for each participant according to their air-
in the study. Seven NH adults (six women, age range: 22 to 29 yr., conduction audiometry as follows: for input sound pressure level
mean = 26 yr., SD = 2.65 yr.), and five HI adults (three women, (SPL) of 65 dB, the gain was as prescribed by the NAL-R rule
age range: 53 to 61 yr., mean = 58.2 yr., SD = 3.35 yr.) partici- (Byrne and Dillon, 1986); for input sound levels lower/higher than
pated in the experiments. All of them had normal otoscopy. HI lis- 65 dB SPL, the gain was higher/lower so that the compression ra-
teners were clinically diagnosed with bilateral, moderate-to-severe tio was 2 dB/dB. For levels lower than the compression threshold
SNHL and regularly wore bilateral HAs. NH listeners had no his- (20 dB SPL), the gain was linear; and at very high levels, no gain
tory of hearing disorders, and their audiometric thresholds were was applied (see supplementary Fig. S2).
<25 dB HL at all octave frequencies from 125 to 80 0 0 Hz. NH lis- Procedures. SRTs were measured using an adaptive procedure.
teners were bilaterally fitted with foam earplugs (3MTM E-A-RTM The target was a sentence from the Spanish matrix sentence test
classic) to simulate a conductive hearing loss. Fig. 1 shows the au- (Hochmuth et al., 2012). The masker was a single source of SSN
diograms for all individual participants and demonstrates that the with the average long-term spectrum of the matrix sentences. Dur-
use of earplugs produced mild-to-moderate hearing losses, as in- ing an SRT measurement, the speech level was fixed at 65 dB SPL
tended. and the masker level varied adaptively using a one-down, one-
Binaural pre-processing method. Level-adjusted, HRTF-filtered up rule. The SRT was thus defined as the SNR at which listeners
sounds were pre-processed using the binaural weighted subtrac- recognized 50% of full sentences (Levitt, 1971). Twenty sentences
tion method of Lopez-Poveda et al. (2022) as follows: were presented to measure an SRT. The initial SNR was set either
equal to 0 dB (for listeners with simulated conductive loss) or was
L ( f ) = L ( f ) − β ( f ) · R ( f ) (1) adjusted individually (for listeners with SNHL) as low as possible

2
F.M. San-Victoriano, A. Eustaquio-Martín and E.A. Lopez-Poveda Hearing Research 432 (2023) 108743

Fig. 1. Individual pure-tone air-conduction audiometric thresholds for listeners with simulated conductive loss (circles) and sensorineural hearing loss (squares). Crosses
indicate conditions where the limit of the audiometer (Interacoustics AD229e) was reached without a response from the participant.

imuth, respectively; and S270 N90 with the speech and noise sources
at 270° and 90° azimuth, respectively. These spatial configura-
tions were the same as tested by Lopez-Poveda et al. (2022) and
were selected as representative of conditions where the benefit
of the pre-processing algorithm was expected to be zero, small-
to-moderate, and large, respectively (supplementary Fig. S1B). All
sound sources were at eye level (0° elevation). Spatial configura-
tions were simulated by convolving monophonic recordings with
HRTFs for a KEMAR manikin (Gardner and Martin, 1995). Note
that for conditions with non-collocated target and masker sources
(S270 N45 and S270 N90 ), the ear with the better acoustic SNR was
the left ear.
For listeners with simulated conductive loss, 36 SRTs in SSN
were measured per participant: 2 listening modes (bilateral and
unilateral with the left ear) × 2 processing conditions (with and
without pre-processing) × 3 spatial configurations (S0 N0 , S270 N45 ,
and S270 N90 ) × 3 SRT estimates per condition. Measurements were
organized in three blocks of 12 SRT measurements each, one block
per SRT estimate.
For listeners with SNHL, 54 SRTs in SSN were measured per par-
Fig. 2. Aided versus unaided SRTs in quiet for the five listeners with SNHL when ticipant: 3 listening modes (bilateral, unilateral with the left ear,
listening bilaterally or unilaterally with the left and the right ear. The speech source and unilateral with the right ear) × 2 processing conditions (with
was at 0° azimuth. Squares and crosses illustrate individual, and group mean scores, and without processing) × 3 spatial configurations (S0 N0 , S270 N45 ,
respectively. Error bars illustrate one standard deviation. ∗ ∗ ∗ p ≤ 0.001. and S270 N90 ) × 3 SRT estimates per condition. Measurements were
organized in three blocks of 18 SRT measurements each, one block
per SRT estimate.
while making sure that listeners could recognize the first test sen- Within each block, conditions were administered in random or-
tence. The SNR changed in 4-dB steps between sentences 1 and 5, der. Participants did not know which condition they were being
and in 2-dB steps between sentences 5 and 20. The SRT was cal- tested on (single blind approach).
culated as the mean of the final 11 SNRs (the SNR for the 21st As a control, for listeners with SNHL, SRTs were also measured
sentence was calculated and used in the SRT estimate but not pre- in quiet and for a speech source at 0° azimuth to verify that the
sented). SRT estimates whose standard deviations (SDs) exceeded experimental HA and fitting rule successfully restored audibility.
3.5 dB were discarded and repeated. Three valid SRT estimates Equipment. During the measurements, listeners were seated
were measured per condition and the mean was taken as the fi- in a double-wall sound booth. The MATLAB software environ-
nal SRT. ment (R2017b, The Mathworks, Inc.) was used to perform all sig-
Conditions. SRTs were measured in unilateral and bilateral lis- nal processing and implement all test procedures. Stimuli were
tening modes. Unilateral listening tests involved processing stim- sampled at a rate of 44,100 Hz and with 24-bit resolution. Stim-
uli through the algorithm but stimulating only one ear. SRTs were uli were controlled using custom-made software and played via
measured for three spatial configurations: S0 N0 with the speech an RME Fireface UCX soundcard and presented to the listeners
and noise source collocated in front of the listener at 0° azimuth; via Sennheiser HD-580 circumaural headphones. SPL were cali-
S270 N45 with the speech and noise sources at 270° and 45° az- brated by placing the headphones on a KEMAR head equipped

3
F.M. San-Victoriano, A. Eustaquio-Martín and E.A. Lopez-Poveda Hearing Research 432 (2023) 108743

with a Zwislocki DB-100 coupler connected to a sound level me- significant [F(2,22)=297.90, p<0.001]. Post-hoc pairwise compar-
ter (Brüel&Kjaer, mod. 2238). Calibration was performed at 1 kHz isons with Bonferroni correction showed that the pre-processing
and the obtained sensitivity was used at all other frequencies. improved SRTs significantly only in the S270 N90 (p<0.001), but not
in the S0 N0 (p = 0.331) or the S270 N45 (p = 0.067) configurations.
3. Results
3.4. Listening unilaterally with the acoustically worse ear
3.1. SRTs in quiet
To assess the effect of pre-processing when listening unilat-
For the five participants with SNHL and for a speech source at erally with the acoustically worse ear (i.e., with the ear con-
0° azimuth, the use of the HA (without pre-processing) improved tralateral to the target), SRTs were also measured for the five
mean SRTs in quiet from 63.6 to 33.7 dB SPL [two-tail paired Stu- participants with SNHL listening unilaterally with the right ear
dent’s t-test, t(4)=6.435, p = 0.003] in bilateral listening, from 71.6 (Fig. 3E,F). A two-way RMANOVA revealed a significant main effect
to 37.3 dB SPL [t(4)=17.949, p<0.001] in unilateral listening with of processing [F(1,4)=160.42, p<0.001]. Mean SRTs (across loca-
the left ear, and from 72.5 to 42.1 dB SPL in unilateral listening tions and participants) were significantly worse for pre-processed
with the right ear [t(4)=11.733, p<0.001] (Fig. 2). These improve- than for unprocessed stimuli (13.5 vs 3.0 dB SNR, respectively).
ments are comparable or greater than HA improvements reported It also revealed a significant main effect of spatial configuration
elsewhere (e.g., Duquesnoy and Plomp, 1983; Moore et al., 1985; [F(2,8)=728.98, p<0.001]. Post-hoc pairwise tests with Bonferroni
Peters et al., 1998), which shows that our software-based HA and correction showed that SRTs were significantly better (lower) in
fitting rule were reasonable in restoring audibility and intelligibil- the collocated condition than in any of the spatially separated con-
ity in quiet. ditions [S0 N0 vs S270 N45 , p<0.001; S0 N0 vs S270 N90, p<0.001], and
they were not significantly different in the S270 N90 and S270 N45
conditions (p = 1.0 0 0). The interaction between processing and
3.2. Listening unilaterally with the acoustically better ear spatial configuration was also significant [F(2,8)=231.75, p<0.001].
Post-hoc pairwise comparisons with Bonferroni correction showed
Fig. 3A shows aided SRTs with and without pre-processing in that the pre-processing worsened SRTs significantly in all spa-
unilateral listening with the left ear (i.e., the acoustically bet- tial configurations: p<0.001 in S270 N45 , p = 0.001 in S270 N90 , and
ter when there was one) and for three spatial configurations p = 0.032 in S0 N0 .
of the speech and noise sources. Fig. 3B illustrates correspond-
ing SRT improvements produced by the pre-processing method 3.5. Comparisons across participant groups
in each case. A two-way repeated measures analysis of the vari-
ance (RMANOVA) was conducted to test for the effect of process- Because the aim was to investigate the effect of the algorithm
ing (with and without pre-processing) and spatial configuration for HA users overall, the analyses reported in the preceding sec-
(S0 N0 , S270 N45 , S270 N90 ) on the SRT. The test revealed a significant tions were done by combining data for all participants. For com-
main effect of pre-processing [F(1,11)=338.49, p<0.001]. Mean SRTs pleteness, however, this section reports complementary analyses
(across locations and participants) were significantly better for pre- aimed at comparing the performance of participants with (simu-
processed than for unprocessed stimuli (−13.3 vs −8.3 dB SNR, lated) conductive loss and SNHL as well as the benefit of the pre-
respectively). It also revealed a significant main effect of spatial processing method for the two participant groups.
configuration [F(2,22)=1269.52, p<0.001]. Post-hoc pairwise tests A RMANOVA with type of hearing loss as a between-subject
with Bonferroni correction showed that SRTs were significantly factor and processing and spatial configuration as within-subject
worse (higher) for collocated than for spatially separated speech factors revealed that SRTs were overall better for participants
and noise sources [S0 N0 vs S270 N45 , p<0.001; S0 N0 vs S270 N90 , with (simulated) conductive hearing loss than for participants
p<0.001], and they were significantly better in the S270 N90 than in with SNHL, both in bilateral listening [−13.2 vs. −9.6 dB SNR;
the S270 N45 condition (p<0.001). The interaction between process- F(1,10)=8.341, p = 0.016] and in unilateral listening with the left
ing and spatial configuration was also significant [F(2,22)=377.77, ear [−12.0 vs. −8.9 dB SNR; F(1,10)=10.577, p = 0.009]. Note that
p<0.001]. Post-hoc pairwise comparisons with Bonferroni correc- the comparison could not be extended to unilateral listening with
tion showed that pre-processing improved SRTs significantly in the the right ear because participants with conductive hearing loss
S270 N45 (p = 0.002) and S270 N90 (p<0.001) configurations, but not were not tested in that condition.
in the S0 N0 condition (p = 0.153). A RMANOVA with type of hearing loss as a between-subject
factor and spatial configuration as within-subject factor revealed
3.3. Listening bilaterally that in unilateral listening, the algorithm provided a significantly
larger benefit for participants with SNHL than for those with
Fig. 3C shows individual and group mean SRTs obtained in bi- conductive loss [5.7 vs. 4.5 dB; F(1,10)=8.011, p = 0.018]. Post
lateral listening with and without pre-processing for the three hoc pairwise comparisons revealed, however, that the benefit was
target-masker spatial configurations. Fig. 3D depicts individual, and different only in the S270 N90 configuration [difference=2.6 dB,
group mean SRT improvements in each case. A two-way RMANOVA p = 0.014], but not in the other spatial configurations [S0 N0 :
revealed a significant main effect of processing [F(1,11)=113.16, −0.63 dB, p = 0.922; S270 N45 : 1.1 dB, p = 0.149]. In bilateral
p<0.001]. Mean SRTs (across locations and participants) were sig- listening, the benefit provided by the algorithm was 3.4 dB for
nificantly better for pre-processed than for unprocessed stimuli participants with conductive hearing loss compared to 4.3 dB for
(−13.6 vs −9.8 dB SNR, respectively). It also revealed a signifi- participants with SNHL, and the difference was not significant
cant main effect of spatial configuration [F(2,22)=564.33, p<0.001]. [F(1,10)=2.031, p = 0.185].
Post-hoc pairwise tests with Bonferroni correction showed that
SRTs were significantly worse (higher) in the collocated condition 4. Discussion
than in any of the spatially separated conditions [S0 N0 vs S270 N45 ,
p<0.0 01; S0 N0 vs S270 N90 , p<0.0 01], and they were significantly The aim of the present study was to investigate if our binau-
better in the S270 N90 than in the S270 N45 condition (p<0.001). The ral pre-processing method for contralateral sound attenuation im-
interaction between processing and spatial configuration was also proves speech-in-noise intelligibility for HA users as it does for NH

4
F.M. San-Victoriano, A. Eustaquio-Martín and E.A. Lopez-Poveda Hearing Research 432 (2023) 108743

Fig. 3. The effect of front-end pre-processing (FE) on aided SRTs. Left panels illustrate individual and group mean aided SRTs with (FE-HA) and without (HA) pre-processing,
while right panels depict SRT improvements produced by the pre-processing in each case. A, B. Listening with the left ear alone, i.e., the acoustically better ear when there
was one. C, D. Listening bilaterally. E, F. Listening with the right ear alone, i.e., the acoustically worse ear when there was one (only tested for listeners with SNHL). In all
panels, circles and squares illustrate scores for individuals with simulated conductive loss and SNHL, respectively; crosses represent group mean scores (N = 12 in panels A-D,
and N = 5 in E-F). Error bars illustrate one standard deviation. Bottom, middle and top lines in each box plot denote the 25, 50 (median), and 75 percentiles, respectively.
n.s.: not significant, ∗ p ≤ 0.05, ∗ ∗ p ≤ 0.01, ∗ ∗ ∗ p ≤ 0.001.

listeners (Lopez-Poveda et al., 2022). We have shown that with Contrary to expectations, for matching listening modes, SRTs
this method, mean aided SRTs for sentences in competition with improvements tended to be greater for HA users (present study)
SSN improve by up to 10.7 dB when listening bilaterally and up to than for NH listeners (reported by Lopez-Poveda et al., 2022). For
13.9 dB when listening unilaterally with the acoustically better ear example, when listening bilaterally the largest improvement for HA
(Fig. 3). users was 10.7 dB in the S270 N90 spatial configuration compared

5
F.M. San-Victoriano, A. Eustaquio-Martín and E.A. Lopez-Poveda Hearing Research 432 (2023) 108743

Fig. 4. Correlation between the SRT improvement produced by the pre-processing method and the pure tone average (PTA) threshold (mean threshold at 0.5, 1.0, 2.0 and
4 kHz) in unilateral and bilateral listening for different azimuthal locations of the speech (S) and noise (N) sources, as indicated by the inset.

to 9.6 dB in the S270 N45 configuration for NH listeners. The im- 1994). In theory, the greater the compression ratio and the faster
provement could be larger for HA users because SRTs in the cor- the compressor, the less the advantages of the pre-processing
responding reference condition (without pre-processing) tended to should be because the method attenuates contralateral sounds
be worse (higher) for HA users than for NH listeners (−14.1 vs and nonlinear compressors amplify weaker sounds more and more
−17.9, respectively), hence HA users had greater room for im- quickly with greater compression ratios and faster time constants.
provement. Indeed, SRT improvements tended to increase with in- Further research is necessary to generalize the found benefits to
creasing hearing loss, particularly for spatially separated target and other HA settings or to cochlear implants, which apply instanta-
masker sources (Fig. 4). The greater SRT improvements for HA neous compression to map the wider range of acoustic hearing into
users could also be due to the use of different subtraction weights, the narrower range of electric hearing (e.g., Lopez-Poveda et al.,
i.e., β 60°−120° and α 240°−300° used here for HA users compared to 2006).
β 0°−180° and α 180°−360° (also referred to as β diff and α diff ) used by The present findings, together with the ease of implementation
Lopez-Poveda et al. (2022) for NH listeners. The difference in sub- already discussed in Lopez-Poveda et al. (2022), reaffirm that bin-
traction weights likely explains why the largest SRT improvement aural weighted subtraction is potentially useful for implementation
occurred for different spatial configurations for HA users and NH as a binaural sound pre-processing strategy in bilateral HAs. Since
listeners (S270 N90 vs S270 N45 ) (supplementary Fig. S1; see also Fig. contralateral sound attenuation can improve speech-in-noise intel-
12 from Lopez-Poveda et al., 2022). ligibility for listeners with normal hearing (Lopez-Poveda et al.,
Previous reports have shown that the intelligibility improve- 2022), the method could also be implemented in consumer elec-
ment that results from spatially separating the target and masker tronic hearing devices (e.g., Chong-White et al., 2022) to help peo-
sources (known as spatial release from masking) can be less ple who suffer from speech-in-noise difficulties but no hearing
for HA users than for NH listeners presumably partly because loss, a condition sometimes referred to as King-Kopetzky syndrome
HA compression reduces inter-aural level differences (Festen and (Zhao et al., 2007).
Plomp, 1986; Marrone et al., 2008). Here, the pre-processing im- In the present tests, stimuli were presented over headphones
proved SRTs more for spatially separated than for collocated tar- and spatial configurations were simulated by filtering monophonic
get and masker sources both when listening bilaterally (Fig. 3D) recordings through free-field KEMAR HRTFs for a fixed distance of
and unilaterally with the acoustically better ear (Fig. 3B). In other 1.4 m. Test conditions were limited to a single masker source (SSN)
words, the pre-processing increased spatial release from masking and to static heads and sound sources. Furthermore, both the HAs
relative to unprocessed, aided conditions. This shows that besides and the pre-processing algorithm were implemented in software.
improving speech-in-noise intelligibility, contralateral sound field These conditions are far from realistic. In realistic listening situa-
attenuation can also increase spatial release from masking for HA tions, HA microphones may be located behind the ear, the type of
users. target and the type and number of interfering sources may vary
Seven of the present participants had simulated rather than ac- from moment to moment, the different sources may be at dif-
tual conductive hearing losses and their audiometric thresholds ferent distances from the listener, potentially moving, and/or not
were overall better than those of the five listeners with SNHL so spatially separated, heads may rotate, and the room may have
(Fig. 1). Nonetheless, the simulated losses were representative of reverberation. In addition, in realistic hardware implementations,
real conductive losses and were large enough for HA treatment the pre-processing algorithm would almost certainly operate over
(Hoppe and Hesse, 2017). Because the aim of the study was to a narrower frequency range (e.g., as determined by the HA mi-
investigate if contralateral sound attenuation can help HA users crophones), with non-matched HRTFs (Lopez-Poveda et al., 2022),
overall (i.e., regardless of the type of hearing loss), heterogeneity and would involve delays in bilateral signal exchange. These factors
in thresholds and hearing loss etiology was desirable. were disregarded in the present tests and may affect the benefit of
The present evaluations were limited to HAs with a compres- the algorithm. Furthermore, given the weak relationship between
sion ratio of 2 dB/dB, and with attack and release times of 5 and beamformer benefits in laboratory tests with self-perceived field
200 ms, respectively. These settings were chosen because they are benefits (e.g., Cord et al., 2004; Gnewikow et al., 2009), the algo-
within the range of values preferred by HA users (Neuman et al., rithm may turn out to be less beneficial in realistic settings and

6
F.M. San-Victoriano, A. Eustaquio-Martín and E.A. Lopez-Poveda Hearing Research 432 (2023) 108743

implementations than reported here. Further research is necessary Duquesnoy, A.J., Plomp, R., 1983. The effect of a hearing aid on the speech-reception
to generalize the found benefits to more realistic implementations threshold of hearing-impaired listeners in quiet and in noise. J. Acoust. Soc. Am.
73, 2166–2173.
and test conditions. Nevertheless, the pre-processing method pro- Edwards, B, 2003. Hearing aids and hearing impairment. In: Greenberg, S.,
posed by Lopez-Poveda et al. (2022) holds promise to improve the Ainsworth, W.A., Popper, A.N., Fay, R.R. (Eds.), Speech Processing in the Audi-
intelligibility of speech in noise and enhance spatial release from tory System. Springer, New York, NY, pp. 339–421 vol. 18.
Festen, J.M., Plomp, R., 1986. Speech-reception threshold in noise with one and two
masking also for HA users. hearing aids. J. Acoust. Soc. Am. 79, 465–471.
Festen, J.M., Plomp, R., 1990. Effects of fluctuating noise and interfering speech on
CRediT authorship contribution statement the speech-reception threshold for impaired and normal hearing. J. Acoust. Soc.
Am. 88, 1725–1736.
Gardner, W.G., Martin, K.D., 1995. HRTF measurements of a KEMAR. J. Acoust. Soc.
Fernando M. San-Victoriano: Investigation, Formal analysis, Am. 97, 3907–3908.
Data curation, Visualization, Writing – original draft, Writing – re- Giannoulis, D., Massberg, M., Reiss, J.D., 2012. Digital dynamic range compressor de-
sign - a tutorial and analysis. J. Audio Eng. Soc. 60, 399–408.
view & editing. Almudena Eustaquio-Martín: Software, Investiga-
Gnewikow, D., Ricketts, T., Bratt, G.W., Mutchler, L.C., 2009. Real-world benefit from
tion, Formal analysis, Data curation, Visualization, Writing – re- directional microphone hearing aids. J. Rehabil. Res. Dev. 46, 603–618.
view & editing. Enrique A. Lopez-Poveda: Conceptualization, Su- Grange, J.A., Culling, J.F., 2016. The benefit of head orientation to speech intelligibil-
pervision, Funding acquisition, Project administration, Methodol- ity in noise. J. Acoust. Soc. Am. 139, 703–712.
Hochmuth, S., Brand, T., Zokoll, M.A., Castro, F.Z., Wardenga, N., Kollmeier, B., 2012. A
ogy, Writing – review & editing. Spanish matrix sentence test for assessing speech reception thresholds in noise.
Int. J. Audiol. 51, 536–544.
Acknowledgements Hoppe, U., Hesse, G., 2017. Hearing aids: indications, technology, adaptation, and
quality control. GMS Curr Top Otorhinolaryngol Head Neck Surg 16. doi:10.3205/
cto0 0 0147.
We thank Milagros J. Fumero and Gabriela Araujo Chavéron for Kates, J.M., 2008. Digital Hearing Aids. Plural publishing.
help with data collection. Work supported by the Spanish Ministry Kollmeier, B., Kiessling, J., 2018. Functionality of hearing aids: state-of-the-art and
future model-based solutions. Int. J. Audiol. 57 (sup3), S3–S28.
of Science and Innovation (grant PID2019–108985GB-I00) and by Levitt, H., 1971. Transformed up-down methods in psychoacoustics. J. Acoust. Soc.
MED-EL GmbH (Innsbruck, Austria). Am. 49, 467–477.
Lopez-Poveda, E.A., Eustaquio-Martín, A., Stohl, J.S., Wolford, R.D., Schatzer, R., Wil-
son, B.S., 2006. A binaural cochlear implant sound coding strategy inspired by
Supplementary materials
the contralateral medial olivocochlear reflex. Ear Hear. 37, e138–e148.
Lopez-Poveda, E.A., Eustaquio-Martín, A., San Victoriano, F.M, 2022. Binau-
Supplementary material associated with this article can be ral pre-processing for contralateral sound field attenuation and improved
speech-in-noise recognition. Hear. Res. 418, 108469.
found, in the online version, at doi:10.1016/j.heares.2023.108743.
Marrone, N., Mason, C.R., Kidd Jr, G., 2008. Evaluating the benefit of hearing aids in
solving the cocktail party problem. Trends Amplif. 12, 300–315.
References Moore, B.C.J., 2008. The choice of compression speed in hearing aids: theoretical
and practical considerations and the role of individual differences. Trends Am-
Abrams, H.B., Kihm, J., 2015. An introduction to MarkeTrak IX: a new baseline for plif. 12, 103–112.
the hearing aid market. Hearing Rev. 22 (6), 16. Moore, B.C.J., Laurence, R.F., Wright, D., 1985. Improvements in speech intelligibility
Byrne, D., Dillon, H., 1986. The National Acoustic Laboratories’(NAL) new procedure in quiet and in noise produced by two-channel compression hearing aids. Br. J.
for selecting the gain and frequency response of a hearing aid. Ear Hear. 7, Audiol. 19, 175–187.
257–265. Neuman, A.C., Bakke, M.H., Hellman, S., Levitt, H., 1994. Effect of compression ra-
Bronkhorst, A.W., Plomp, R., 1988. The effect of head-induced interaural time tio in a slow-acting compression hearing aid: paired-comparison judgments of
and level differences on speech intelligibility in noise. J. Acoust. Soc. Am. 83, quality. J. Acoust. Soc. Am. 96, 1471–1478.
1508–1516. Peters, R.W., Moore, B.C.J., Baer, T., 1998. Speech reception thresholds in noise with
Chong-White, N., Mejia, J., Valderrama-Valenzuela, J.T., Edwards, B., 2022. Evaluation and without spectral and temporal dips for hearing-impaired and normally
of Apple Airpods Pro with Conversation Boost and Ambient Noise Reduction for hearing people. J. Acoust. Soc. Am. 103, 577–587.
People With Hearing Loss in Noisy Environments. Hearing Review. Picou, E.M., Ricketts, T.A., 2019. An evaluation of hearing aid beamforming micro-
Cord, M.T., Surr, R.K., BE, Walden, Dyrlund, O., 2004. Relationship between labora- phone arrays in a noisy laboratory setting. J. Amn. Acad. Audiol. 30, 131–144.
tory measures of directional advantage and everyday success with directional Zhao, F., Stephens, D., 2007. A critical review of King-Kopetzky syndrome: hearing
microphone hearing aids. J. Am. Acad. Audiol. 15, 353–364. difficulties, but normal hearing? Audiol. Med. 5, 119–124.

You might also like