Cochlear-implant dichotic listening performance and effort are disrupted with functional ear asymmetry

Abstract

When speech understanding abilities differ across the ears, auditory attention and listening effort could be impacted. Twenty listeners with bilateral cochlear implants (CIs) completed this experiment. Fourteen listeners had symmetric and six listeners had asymmetric functional hearing. Listeners completed monotic and dichotic digit recall with digit string lengths of four and six digits and attention directed to each ear. Pupil size was monitored as an index of listening effort. Individual differences in working memory and inhibition abilities were measured. It was hypothesized that ear asymmetry would lead to poorer listening performance and higher listening effort, and that cognitive abilities would predict both performance and listening effort. Greater differences in performance across ears were observed with asymmetry. Lower listening effort was observed with asymmetry, regardless of which ear was attended. Poorer working memory abilities predicted higher listening effort. These results suggest that asymmetric listeners may experience reduced perception of a poorer ear, and that individuals with poorer working memory abilities are at risk to experience higher listening effort in complex listening environments with CIs. More broadly, these results suggest that the salience of sensory inputs contributes to auditory attention ability and use of cognitive resources.

I. INTRODUCTION

The auditory system has bilateral sensory inputs: the left and right ears. An advantage of having two sensory inputs is for the sensory system to make meaningful comparisons across those inputs, which leads to perceived spatial location. Small differences are encoded across inputs in the normal system (e.g., Grothe et al., 2010); thus, inputs need to be relatively symmetric for meaningful comparisons to be made. This symmetry can be disrupted with sensory loss. For example, the symmetry of the bilateral inputs from cochlear implants (CIs), auditory prostheses, can be disrupted (Gordon et al., 2015; Goupell et al., 2016; Firszt et al., 2018), making listening with CIs a model of auditory system asymmetry.

CI functional ear asymmetries, or monaural differences in speech understanding abilities across ears, are driven by various individual and device factors (Mosnier et al., 2009; Goupell et al., 2018a; Bernstein et al., 2020). Individual factors include differences in durations of deafness across ears (Goupell et al., 2018a; Bernstein et al., 2020) and hearing abilities without devices (Firszt et al., 2008). Surgical and device factors include differences in insertion depths of the implants in the inner ears (Goupell et al., 2022; Cleary et al., 2023) and the poor spectral resolution with a CI due to current spread (Nelson et al., 2008; Goehring et al., 2021).

With two CIs, there are spatial-hearing benefits, including advantages for sound localization (Grantham et al., 2007) and spatial release from masking (Loizou et al., 2009; Bakal et al., 2021). Performance improves with a second CI in these situations, but the benefits fall short of restoring spatial-hearing performance comparable to listeners with normal hearing thresholds (Hawley et al., 2004; Loizou et al., 2009). Ear asymmetries likely contribute to limited spatial-hearing benefits. The variability in binaural outcomes could be a consequence of the aforementioned variability in monaural outcomes, since simulations of mismatched place of stimulation across ears reduce binaural benefits (Yoon et al., 2011; Goupell et al., 2018b).

Given the role of attention in spatial-hearing benefits for speech understanding in noise (Carlile and Corkhill, 2015), there may be additional bilateral benefits beyond improved speech understanding that are appreciated by listeners with bilateral CIs. Reduced listening effort, or less demand on mental resources to accomplish a listening-related goal (Pichora-Fuller et al., 2016), could be a clinically relevant benefit of bilateral CIs (Hughes and Galvin, 2013) that is not captured by performance metrics alone (Winn and Teece, 2021). Pupillometry has been used as an objective measure of listening effort in listeners with CIs (Winn and Moore, 2018; Russo et al., 2020), with an ability to examine effort throughout the time course of a task. Increased listening engagement as an action is reflected in the pupil response (Peelle, 2018; Herrmann and Johnsrude, 2020), such that the task-evoked pupil response decreases when tasks are too easy as well as too difficult (i.e., giving up, Ohlenforst et al., 2017). Improved understanding of bilateral listening effort in noisy environments could inform strategies to reduce effortful listening in individuals with CIs.

Listening is particularly effortful (e.g., Koelewijn et al., 2014; Ohlenforst et al., 2017; Wendt et al., 2017) in noisy environments with competing speech (i.e., “the cocktail party,” Cherry, 1953). As a simplified proxy of listening in these noisy environments, the dichotic listening paradigm has been used to assess attentional demands during effortful speech perception. In this dichotic listening paradigm, a speech target is presented in one ear and an interferer presented in the other ear. Individuals with bilateral CIs have shown a decrement in performance attending to a target with the introduction of an interferer in dichotic listening (Goupell et al., 2016; Misurelli et al., 2020). Performance also sometimes depends on the ear presented with the target, suggesting that ear differences could play a role in dichotic listening abilities.

Acoustic CI simulations (channel vocoders; Cychosz et al., 2024) in listeners with normal hearing have been used to parametrically examine listening effort by processing auditory information in a similar way that CIs process sound. In a dichotic digit recall task, where listeners recalled digits in one ear and ignored interfering digits in the other ear, the spectral resolution of both target and interferer affected the listening effort exerted (DeRoy Milvae et al., 2021), suggesting that ear asymmetries related to spectral resolution differences could impact dichotic listening effort. A clearer interferer was more effortful to ignore (DeRoy Milvae et al., 2021). These observations in highly controlled laboratory experiments suggest that spectral resolution asymmetries have the potential to impact listening effort in more realistic listening environments, such as with competing speech.

Cognitive abilities potentially play a role in the amount of listening effort exerted, and thus should also be further examined. Young adults showed little effect of individual differences in cognitive abilities observed on listening effort (DeRoy Milvae et al., 2021), but older adults likely have a larger range of cognitive abilities that may influence both performance (Akeroyd, 2008) and effort (Koelewijn et al., 2012) on a listening task. Older adults with relatively lower cognitive abilities may be more susceptible to decreased performance and increased effort with an increase in the memory load of a task, especially with the added factor of hearing loss. The relationship between cognitive abilities and listening effort likely depends on the task (Strand et al., 2018), as findings have conflicted across studies (Koelewijn et al., 2012; Wendt et al., 2017). Several hypotheses exist for the relationship between cognitive abilities and listening effort, and it is not yet clear how task differences lead to the conflicting findings across studies (Strand et al., 2018).

In summary, in simulations of CI listening with young adult acoustic-hearing listeners, a clearer interferer is more effortful to ignore in scenarios related to cocktail party listening. Listeners' cognitive abilities do not appear to predict listening effort (DeRoy Milvae et al., 2021). However, it is not yet clear whether the observed effects on listening effort with CI simulations translate to bilateral CIs. Individuals with bilateral CIs may experience functional asymmetries in speech understanding and greater variation in cognitive abilities with their increased age relative to previous studies that tested young adults with normal symmetric hearing. This could impact performance and effort during dichotic listening tasks and, more broadly, when communicating in noisy environments.

To address these gaps in knowledge, in the current experiment, monotic and dichotic listening were examined in adults with bilateral CIs. Listeners performed a digit recall task with target ear and memory load manipulation via digit string length (DSL). It was hypothesized that performance would be worse and effort would be higher for dichotic listening compared to monotic listening as observed previously in CI simulations (DeRoy Milvae et al., 2021). It was also hypothesized that symmetry would impact dichotic listening performance and effort across ears (Goupell et al., 2016; Bakal et al., 2021; DeRoy Milvae et al., 2021). An interaction between the introduction of an interferer and increased DSL was also hypothesized. Specifically, performance was expected to decline, and effort was expected to increase to a greater extent with longer DSL and interfering digits. Finally, in this group of primarily older adult listeners, it was hypothesized that listeners with poorer cognitive abilities, specifically in working memory and inhibition, would exhibit larger decrements in performance and greater effort with increased DSL, since individual differences have been shown to impact effort in adults with CIs (Zhang et al., 2023). Measures of listening effort could dissociate from performance (Winn and Teece, 2021), giving a more holistic view of how the two ears function together to form a unified auditory percept in the broader population and especially for individuals with asymmetric hearing loss.

II. METHODS

A. Listeners

Twenty listeners with bilateral CIs, 14 symmetric and six asymmetric as measured by the difference in sentence recall scores across ears, participated in this experiment. Their demographic information can be found in Table I. Ages ranged from 23 to 82 yrs with a median of 64 yrs. All listeners reported English as a native language and scored above 23 on the Montreal Cognitive Assessment (Nasreddine et al., 2005; Carson et al., 2018). The Montreal Cognitive Assessment was completed within 1 year of testing in this experiment. The Institutional Review Board at the University of Maryland-College Park approved this research protocol and all listeners provided informed consent.

TABLE I.

CI listener demographics and sentence-in-quiet keyword recall. Listeners with sentence score differences less than 20% are labeled as symmetric (S) and with larger sentence score differences are labeled as asymmetric (A). There were 100 sentences scored per ear with the exceptions noted that indicate testing done with fewer sentences.

	Demographics						Sentence scores
Listener	Sex	Age (yr)	CI experience (yr) left/right	CI processors left/right	Duration of deafness (yr) left/right	Etiology	Percent correct left/right	Mean	Difference
S1	M	79	16/12	N6/N6	3/7	Family history	97.8/97.8	97.8	0.0
S2	F	60	19/11	N7/N7	4/12	Family history	85.8/84.2	85.0	1.6
S3	F	23	21/13	N6/N6	2/2	Unknown	98.0/95.0	96.5	3.0
S4	M	48	3/2	Q90/Q90	3/2	Unknown	95.6/91.6	93.6	4.0
S5	F	66	9/21	Q90/Q90	3/1	Ototoxicity/Trauma	79.4/75.2	77.3	4.2
S6	F	55	11/7	N6/N6	1/5	Unknown	93.8/98.2	96.0	4.4
S7	F	31	28/10	N7/N7	2/20	Family history	98.0a/91.3a	94.7	6.7
S8	F	66	8/7	N6/N6	12/13	Unknown	88.6/95.4	92.0	6.8
S9	F	71	22/22	Q70/Harmony	2/2	Ototoxicity/Trauma	45.8/53.0	49.4	7.2
S10	F	71	10/8	N6/N6	0/0	Ototoxicity/Trauma	79.4/89.4	84.4	10.0
S11	M	48	1/3	N7/N7	5/2	Meniere's disease	84.8/95.8	90.3	11.0
S12	F	66	5/7	N6/N6	5/3	Ototoxicity/Trauma	64.4/78.2	71.3	13.8
S13	M	69	5/4	N7/N7	14/23	Unknown	76.6/90.6	83.6	14.0
S14	F	61	11/10	N6/N6	5/6	Meniere's disease	63.4/81.4	72.4	18.0
A1	F	45	9/1	Q90/Q90	0/0	Unknown	36.0/12.4	24.2	23.6
A2	F	42	9/10	Q70/Q70	33/32	Ototoxicity/Trauma	3.0b/32.6	17.8	29.6
A3	F	73	18/13	N6/N6	0/4	Family history	31.2/72.2	51.7	41.0
A4	M	56	11/9	N7/N7	2/4	Unknown	48.2/5.0b	26.6	43.2
A5	M	82	17/10	N7/N7	50/1	Ototoxicity/Trauma	22.0a/81.7a	51.9	59.7
A6	M	69	1/5	Q90/Q90	5/61	Ototoxicity/Trauma	87.4/6.0b	46.7	81.4

^a60 sentences.

^b20 sentences.

B. Stimuli

Digits from one to nine, excluding seven because it is multi-syllabic, were selected from a matrix sentence corpus (Kidd et al., 2008). All digit utterances were recorded by the same male talker. The digits were low-pass filtered at 4 kHz with a third-order forward-backward Butterworth filter (–36 dB/oct attenuation) to facilitate future comparisons with acoustic hearing older adults with high-frequency hearing loss. The filtered digits were all scaled to the same root mean square (RMS) energy, and were the same unprocessed stimuli used previously with listeners with normal symmetric hearing (DeRoy Milvae et al., 2021).

Digits were presented in four- and six-digit strings, with the onset of each subsequent digit occurring 1 s after the previous digit onset. The six-digit strings were tested previously in a CI simulation with listeners with normal symmetric hearing (DeRoy Milvae et al., 2021), and four-digit strings were added to test the effects of memory load in this relatively older population with hearing loss. Stimulus presentation mode (PM) was monotic and dichotic. Digits were randomly selected and allowed to repeat within a digit string, but in dichotic conditions, the simultaneously occurring digits were always different to minimize possible binaural fusion (DeRoy Milvae et al., 2021).

To test speech understanding in quiet and assess functional interaural symmetry, sentences (IEEE sentences; Rothauser et al., 1969) were all scaled to the same RMS energy. These sentences are open-set format and consist of five keywords per sentence. All sentences were recorded by the same male talker. These scores for each ear were used as a measure of ear symmetry (Table I). Speech understanding in quiet has been previously used as a functional measure of ear asymmetry (Mosnier et al., 2009; Bakal et al., 2021).

C. Procedure

1. Processor functionality

Prior to testing, a check of the listener's CI processor functionality was performed. When possible, the researcher listened to the microphones over headphones, checking that sounds that together include the frequency information included in speech were audible and clear (Ling Six sounds; Ling, 1976). Also, when possible, a status check was performed using the processor remote. If neither of these were possible, listener report was used to verify that microphones and equipment were working well (only S9). All processor functionality checks were satisfactory.

2. Dichotic listening

Testing occurred in a dimly lit and sound-treated room, with two listeners (S7 and A5) completing the experiment with an identical system in a double-walled sound-attenuating booth (IAC, Bronx, NY). Sessions were approximately 2 h long, with the experiment consisting of two sessions per listener. Custom E-Prime (Psychology Software Tools, Sharpsburg, PA) software was used to present stimuli with a Chronos (Psychology Software Tools, Sharpsburg, PA, USA) external sound card (DeRoy Milvae et al., 2021). Circumaural open-back headphones (HD-650, Sennheiser Electronic Corporation, Wedemark, Germany) were placed over clinical processors to present stimuli (Bakal et al., 2021; Bernstein et al., 2024). Listeners were seated in front of a 24 in monitor at a distance of approximately 26 in, with their chin on a rest during eye tracking portions of the experiment. The eye tracker (EyeLink 1000 Plus, SR Research, Ottawa, Ontario, Canada) camera was in front of the monitor to track left eye gaze and pupil size. Calibration and validation of the eye tracking was performed with a nine-point grid at the start of each block.

Testing began with loudness balancing to confirm that the stimuli were comfortably loud and matched in loudness across ears. Listeners used their clinical processors in their preferred everyday program. They were directed to ignore any clarity and pitch differences across ears and report any differences in loudness. The experimental stimuli digits were played to each ear over headphones separately, alternating by ear, and then together. Level was initially set at 65 dB SPL, the level used previously with listeners with normal symmetric hearing (DeRoy Milvae et al., 2021), and adjusted to achieve loudness balance; the balanced levels were used for the duration of the experiment. All listeners reported that the loudness was balanced across ears in the experimental settings and that the diotic digits were perceived to be centered in the head, with two exceptions. For one participant (A5), loudness balancing could not be achieved but testing proceeded with stimulus levels that were perceived as centered. Another participant (A6) was tested at 65 dB SPL since loudness balancing and centering could not be achieved.

A practice block was completed at the start of each of the two experimental sessions. Pupil dynamic range [an estimate of minimum (Min) to maximum (Max) dilation] was measured at the start of the practice block (once per testing session) in an attempt to control for individual differences, such as age-related restriction of the pupillary dynamic range (Piquado et al., 2010; DeRoy Milvae et al., 2021; Abramowitz et al., 2024). Identical to a previous study (DeRoy Milvae et al., 2021), the average pupil size during viewing of a 10 s black screen was used to estimate Max pupil dilation and the average pupil size during a 10 s white screen was used to estimate Min pupil dilation. After the dynamic range measurement, a set of practice trials was completed to familiarize listeners with the task. The practice included two trials of each experimental condition (monotic four-digit, dichotic four-digit, monotic six-digit, and dichotic six-digit) presented in a randomized order, for a total of eight trials. Participants were instructed to recall digits in the ear that would be attended throughout the session. The starting test ear was counterbalanced such that approximately half of the listeners were directed to attend to their better ear in the first session.

The testing then proceeded after the practice block. The test sessions included all four experimental conditions presented in a randomized order, with five trials per condition per block, for a total of 30 trials per condition. Breaks were provided between blocks as needed and eye tracker calibration was completed at the start of each block. At the second session, the same loudness-balanced levels were used and the same structure for the dynamic range measurement, practice trials, and experimental trials were completed with attention directed to the other ear.

3. Speech understanding in quiet

In a separate session from the dichotic listening task of approximately 0.5–1 h, functional speech understanding was assessed as a measure of asymmetry (Bakal et al., 2021). Testing was completed in a double-walled sound-attenuating booth (IAC). Stimuli were presented at loudness-balanced levels over the same type of circumaural headphones used for dichotic listening, again placed over the listener's CI processors. Responses were oral and keywords were scored after each trial by the experimenter. Breaks were taken as needed. Functional speech understanding scores are shown in Table I, with listeners ordered by level of asymmetry. Listeners with less than a 20% difference in scores across ears were categorized as symmetric (n = 14) and listeners with greater differences in speech understanding in quiet were categorized as asymmetric (n = 6).

4. Cognitive testing

Cognitive assessments lasting approximately 5 min each were also completed in an additional session. Working memory and inhibition assessments were administered on an iPad (Apple, Cupertino, CA, USA). The tests included were from the National Institutes of Health (NIH) Toolbox (Gershon et al., 2013), specifically the List Sorting Working Memory Test and the Flanker Inhibitory Control and Attention Test as measures of working memory and inhibition abilities.

D. Analysis

1. Behavioral analysis

Behavioral performance was analyzed with a generalized linear mixed-effects model (GLMM) in R 4.3.3 (R Core Team, 2024) with the buildmer 2.11 (Voeten, 2020) and lme4 1.1–35.1 (Bates et al., 2015) packages. The dependent variable was a vector of the number of correct digits and number of incorrect digits for each trial, modeled with a binomial family. The factors included were PM (monotic or dichotic), DSL (four or six digits), symmetry (symmetric or asymmetric listeners), ear (better or poorer), and their interactions. The reference levels used were monotic PM, a DSL of four, the symmetric listeners, and the better ear. Random effects included a by-participant intercept with slopes for PM, DSL, ear, and their interactions. Initially, working memory and inhibition scores were also included as fixed effects, but this led to convergence issues. Alternatively, the cognitive effects were examined by testing whether there was a correlation between each cognitive score and the DSL slope from the extracted by-subject random effects. The maximal model (Barr et al., 2013) excluding cognitive scores was used with the buildmer algorithm (default settings other than use of the “bobyqa” optimizer and increased iterations to 200 000 to give greater opportunity for the model to converge) to find the model that best fit the data and converged. The conditional and marginal R² of the final model (Nakagawa et al., 2017) were estimated using the performance package (Lüdecke et al., 2021).

2. Pupillometry analysis

a. Prepreprocessing: Dynamic range.

The pupil dynamic range was identified from the dynamic range measurement collected during each session for each participant. To exclude off-center looks, gaze in the middle third of the screen was accepted. Missing data were linearly interpolated and data were smoothed with a five-point moving window. These data were downsampled to 100 Hz. The dynamic range measurement was used to transform the experimental pupil data to units of percent of dynamic range in an effort to account for individual differences, particularly with age, on pupil size (Piquado et al., 2010). Specifically, the average pupil size during the black screen was an estimate of pupil Max and the average pupil size during the white screen was an estimate of pupil Min. Two participants had invalid pupil Min measurements (100% interpolated) at the start of one or both sessions. In these cases, the dynamic range from the other session or a repeated dynamic range measurement at the end of the session was used to estimate the Min for that session.

b. Preprocessing: Dichotic listening.

Preprocessing was completed on the experimental data as recommended for the statistical analysis (van Rij et al., 2019). The time window of preprocessing was 8 s for the four-digit data and 10 s for the six-digit data, including the 1 s baseline, 4 or 6 s of stimulus presentation, and 3 s of a retention interval before digit recall. Blinks, saccades, and gaze off-screen were removed, not interpolated (van Rij et al., 2019). This variability in gaze was allowed because in the analysis, gaze direction would be modeled to account for differences in pupil size with changes in gaze (Gagl et al., 2011). If over one-third of a trial or one-third of the baseline was removed in this preprocessing, the trial was removed from the dataset. There were 4800 total trials, with 4191 accepted for analysis (12.7% removed). The experimental data were downsampled to 50 Hz as recommended for the analysis (van Rij et al., 2019). The data were then transformed from arbitrary units to percent dynamic range (%DR) with the following equation, where Max represented the average pupil size during the black screen of the dynamic range measurement and Min represented the average pupil size during the white screen of the dynamic range measurement:

$$\%DR=\frac{\left(X-Min\right)}{Max-Min}\times 100.$$ (1)

Finally, the data were baseline-corrected by subtracting the mean pupil size during the baseline of each trial from each time point during the trial. While conditions were randomized within-session, ear was blocked across sessions, and therefore some impacts of baseline correction could be to subtract out differences in the pupil response during preparation for the stimuli across the blocked factor of ear.

c. Generalized additive mixed modeling.

Analysis of the pupil dilation post-baseline was completed with generalized additive mixed models (GAMMs) using the mgcv package (Wood, 2011) in R. This nonlinear hierarchical regression approach has been used for pupillometry (van Rij et al., 2019; Abramowitz et al., 2024; Johns et al., 2024). For a conceptual introduction to this analysis method, see Sóskuthy (2017). This method is advantageous in that potential artifacts can be accounted for within this framework, including terms to model the impact of time series autocorrelation as well as gaze position on the data. Model testing and visualization were completed with the itsadug package (van Rij et al., 2022).

An ordered factor approach was used to examine how factors and their interactions predicted pupil dilation (Sóskuthy, 2021; Wieling, 2018). This approach is similar to dummy coding of categorical factors in linear regression, with variables for each factor and their interactions set to “true” or “false,” leaving out one category of each factor as the reference. With this approach, ordered factors are included as parametric terms, which can be interpreted as modeling height differences in the pupil response relative to the reference. Smooth terms are also included in the model, representing curve shape over time with a nonlinear fit. Coding of smooth terms is similar again to that of linear regression, with a reference smooth to represent the ordered condition not explicitly outlined with a “by” argument (similar to the intercept in a dummy-coded linear regression). The additive nature of the model is such that each difference smooth included in the model predicts the difference between the reference condition and the condition given in the “by” argument.

Random effects were included in the models with random difference smooths. These estimated by-listener random smooths for the ordered factors included in the model (Sóskuthy, 2021) allow the nonlinear curve shape over time to be modeled for each listener. The autocorrelation “rho” parameter was estimated using an initial model and confirmed to adequately reduce autocorrelation in the model (below 0.2 at and above a lag of one sample). The number of knots is a parameter of the smooth terms, related to the number of underlying basis functions. With a greater number of knots, more wiggliness (nonlinear curvature, Sóskuthy, 2017; van Rij et al., 2019) can be modeled, but at the cost of a more computationally demanding model. Knots were increased when indicated by the check.gamViz function, but this check could not be satisfied for the smooth of gaze location.

Fitted smooths were visualized using the plot_smooth function in itsadug. Horizontal lines were used in plots to indicate significant differences between conditions, as indicated by the plot_diff function in itsadug. These differences reflect differences in the combined parametric and smooth parameters.

Analysis windows were constrained by the conditions compared or collapsed in each analysis, since time windows of comparison cannot differ in duration across conditions. In analyses where data were collapsed across four- and six-digit string lengths, the early perceptual time window (where participants did not yet know if the trial would contain four or six digits) was analyzed. When four- and six-digit string lengths were compared directly, the retention period after the stimuli were compared, again to choose a time window where auditory events were similar (in this case, post-stimuli).

III. RESULTS AND DISCUSSION

A. Behavioral performance

1. Ear asymmetry and DSL impact behavioral performance on monotic and dichotic listening

Mean behavioral results for the dichotic listening task are shown in Fig. 1. It was hypothesized that performance would be significantly worse for dichotic listening compared to monotic listening, and that symmetry would interact with this effect of PM, such that the asymmetric listeners would have a larger effect of PM in their poorer ears. An interaction between PM and DSL was also hypothesized. Together, it was expected that greater demands such as attending a poorer ear, ignoring interfering digits, and holding more digits in memory would decrease performance.

FIG. 1.

Mean performance for PM (monotic and dichotic listening) and DSL (four or six digits) by ear (better or poorer). Mean data for the symmetric listeners (n = 14) are solid, and mean data for the asymmetric listeners (n = 6) are striped. Error bars are ±1 standard error.

These hypotheses were supported by the analysis shown in Table II. There was a significant interaction between PM, attended ear, and symmetry (p < 0.001). There was also a significant interaction between attended ear, DSL, and symmetry (p < 0.001). For symmetric subjects (with sentence score differences across ears less than twenty percent, solid bars in Fig. 1), performance decreased from monotic to dichotic presentation modes when attending to the better or poorer ear (p < 0.001). Performance was also poorer with greater DSL (p < 0.001).

TABLE II.

GLMM summary for behavioral data (Fig. 1) describing the effects of PM (monotic or dichotic), ear (better or poorer), DSL (four or six digits), symmetry (symmetric or asymmetric), and their interactions on performance. Significant fixed effects are shown in bold.

	Accuracy
Fixed effects	Estimate		Standard error (SE)			z		p
Intercept (monotic, better ear, four digits, symmetric)	3.94		0.34			11.45		<0.001
PM (dichotic)	−1.71		0.27			−6.36		<0.001
Ear (poorer)	−0.37		0.23			−1.64		0.10
DSL (six digits)	−2.35		0.26			−8.94		<0.001
Symmetry (asymmetric)	0.04		0.50			0.07		0.94
Ear × Symmetry	−2.09		0.42			−4.98		<0.001
PM × Ear	−0.49		0.34			−1.45		0.15
PM × DSL	1.05		0.20			5.13		<0.001
Ear × DSL	0.20		0.11			1.87		0.06
DSL × Symmetry	0.11		0.36			0.31		0.76
PM × Symmetry	0.87		0.40			2.19		0.03
Ear × DSL × Symmetry	0.96		0.21			4.55		<0.001
PM × Ear × Symmetry	−2.26		0.61			−3.69		<0.001
Random effects	Variance	SD		Correlations
By-subject: intercept	1.55	1.24
By-subject: Ear slope	0.53	0.73		−0.15
By-subject: PM slope	0.77	0.88		−0.31	0.12
By-subject: DSL slope	0.77	0.88		−0.77	−0.30		0.19
By-subject: Ear × PM slope	1.50	1.23		−0.05	0.01		−0.14		0.00
By-subject: PM × DSL slope	0.41	0.64		0.70	−0.07		−0.55		−0.70	−0.38
Conditional R2: 0.53
Marginal R2: 0.35

To examine these interactions further, the reference symmetry in the model was changed to the asymmetric listeners. All fixed effects that were significant with a reference of symmetric listeners (bold fixed effects in Table II) were also significant with a reference of asymmetric listeners (p < 0.05, see Table I in the supplementary material).

There were additional significant interactions with a reference of asymmetric listeners. First, an interaction between ear and DSL (p < 0.001). This helps to explain the ear, DSL, and symmetry interaction, in that the asymmetric listeners showed a significant effect of ear on the impact of DSL on performance. This is shown in Fig. 1, where there was little change in performance from monotic to dichotic presentation modes when attending to the better ear and a more dramatic decrease in performance when attending to the poorer ear under dichotic listening for the asymmetric listeners (striped bars).

Additionally, there was a significant interaction between PM and ear with a reference of asymmetric listeners, such that performance was significantly worse for dichotic compared to monotic listening in the poorer ear (p < 0.001). This also helps to explain the PM, ear, and symmetry interaction because the asymmetric listeners showed a greater effect of PM in the poorer ear than the symmetric listeners. Finally, the effect of ear was also significant in the asymmetric group (p < 0.001), reflecting the greater ear effects in the asymmetric listeners.

These results are consistent with our hypothesis that performance would be worse for dichotic listening compared to monotic listening, with an effect of symmetry on dichotic listening performance across attended ear. They are also consistent with the hypothesized worsening accuracy with the introduction of an interferer or increased DSL, such that performance declined with greater DSL and interfering digits. However, the positive interaction effect suggests that the impact of PM on performance was weaker for higher digit string lengths, potentially driven by relatively worse performance in the more cognitively demanding six-digit condition, even for monotic listening.

The performance of these CI listeners can be compared to a previous dichotic listening experiment with CI simulations. Performance significantly decreased from monotic to dichotic listening in this experiment, as has been seen with vocoded digits (DeRoy Milvae et al., 2021). Further, the results are similar to results with simulated asymmetry, where performance was significantly higher when attending a better ear and ignoring a poorer ear (Table III of DeRoy Milvae et al., 2021) than in other combinations of better and poorer spectral resolutions.

Importantly, CI performance was generally poorer than predicted based on simulations of CI listening involving degradation of the auditory input. In particular, performance was near the performance floor for dichotic listening with attention directed to the poorer ear in asymmetric listeners, which was never the case in simulations; asymmetric auditory signals led to small differences (less than 5%) in performance from symmetric conditions (DeRoy Milvae et al., 2021). These performance data suggest that asymmetry does not only reflect a difference in the quality of the auditory input itself, but there are likely other factors impacting performance to result in such a large effect for dichotic listening in this study. Individual differences were important to the model of performance data as well. Unexplained variance across individuals, as captured by the random effects of the model, suggests that individual difference factors not examined in this study have the potential to account for a large amount of the additional variance in performance. The model had a conditional R² of 0.53, related to the explained variance by both fixed and random effects, which was much larger than the marginal R² of 0.35, related to the explained variance by just the fixed effects.

Other studies point to possible relevant factors for poorer performance and greater variability than in simulations. A previous study examining dichotic listening with CI listeners showed that two of the 11 listeners also had very poor performance on dichotic listening with directed attention to the poorer ear (Goupell et al., 2016). In that experiment, they proposed that a possible explanation was that long duration of deafness led to changes in neural pathways, impacting the ability to direct attention to the poorer ear when there is also input to the better ear. This explanation was consistent with the hearing history of one listener but not the other. In the present study, again the asymmetric listeners more often had longer durations of deafness, associated with potential for greater neural degeneration (Webster and Webster, 1981), but this explanation does not hold for all of them (see Table I). Another possible factor is age, since the CI listeners were on average older than the young adult listeners in DeRoy Milvae et al. (2021) and age can impact dichotic listening performance (Humes et al., 2006).

2. Cognitive factors did not predict individual differences in performance

The relationship between cognitive scores and individual differences in performance were examined with correlations between the working memory and inhibition z-scores (calculated from the sample distribution of standard scores, not age corrected, in this dataset) and by-subject random effects from the model in Table II. The by-subject DSL slope was chosen for this analysis. Working memory scores and inhibition scores were not significantly related to individual differences in performance (p > 0.05). These results are inconsistent with the hypothesis that listeners with poorer cognitive abilities would exhibit larger decrements in performance with increased DSL. This result is in contrast to that found in younger listeners with normal symmetric hearing listening to CI simulations, where higher working memory scores were associated with higher performance as well (DeRoy Milvae et al., 2021). One possible explanation is that other factors, such as age and duration of deafness, played larger roles on individual differences in this population.

B. Pupillometry measures of effort

1. Dichotic listening is more effortful than monotic listening with CIs

As performance was expected to decrease with the introduction of interfering digits, pupil dilation was expected to also increase with interfering digits in the dichotic PM. Data from the symmetric group were modeled to test this hypothesis of an effect of PM. Asymmetric listeners were excluded because this hypothesis was unrelated to symmetry effects. The data were collapsed across attended ear and DSL by examining the initial perceptual time window of 4 s post-baseline. Because DSL was randomized, listeners did not know until 4 s post-baseline if they were listening for four or six digits, which made collapsing the data in this way possible. Model smooth fits are shown in Fig. 2, with vertical lines marking the onset of each digit or digit pair. The peak following each vertical line is in response to each digit or digit pair (Piquado et al., 2010). The horizontal black line denotes the time window of significant differences across condition. For most of the time window, pupil dilation was significantly higher for dichotic compared to monotic listening. This effect is shown in the Table III summary of this model, with a significantly higher parametric coefficient for dichotic listening, associated with curve height (p < 0.001) and a significant difference in the fitted smooth as well, associated with curve shape (p < 0.001).

FIG. 2.

Estimated pupil dilation for the symmetric listeners (n = 14) collapsed across ear and DSL. Estimated pupil dilation is plotted in %DR. Time is post-baseline period, with the onset of the first digit occurring at 0 s. Vertical gray lines represent the onsets of each digit or digit pair. The black horizontal line indicates the time window of significant differences from the GAMMs analysis (Table III).

TABLE III.

GAMM summary for pupil data describing the effects of PM on pupil size for symmetric listener data (n = 14) collapsed across DSL and ear for the initial perception period (Fig. 2). Significant parametric and smooth effects are shown in bold.

Parametric coefficients	Estimate	SE	t	p
Intercept (monotic)	3.52	0.99	3.57	<0.001
PM (dichotic)	3.62	0.79	4.56	<0.001
Smooth terms	Estimated degrees of freedom (edf)	Reference degrees of freedom (Ref.df)	F	p
Time	28.60	32.82	4.02	<0.001
Time by PM (dichotic)	23.35	29.02	4.32	<0.001
Pupil position (X, Y)	57.85	58.94	246.55	<0.001
Random smooth terms
Time × subject	130.76	140.00	72.78	<0.001
Time × subject by PM	121.73	140.00	11.23	<0.001

These results show the effort underlying the performance. It was possible that the degraded signal experienced by CI users would require high listening effort, even in the monotic condition. If this were the case, this population could have shown similar effort for monotic and dichotic listening. However, the condition with poorer performance (dichotic listening) had higher pupil dilation than monotic listening during auditory perception, interpreted as higher listening effort. This pattern is consistent with previous data with young adult listeners with normal symmetric hearing (DeRoy Milvae et al., 2021), showing that these perceptual effects extend to listeners primarily older in age and listening with CIs. Notably, these data suggest that listeners with CIs do not reach a listening effort ceiling during monotic listening. It is not the case that they must dedicate similar cognitive resources across presentation modes resulting in less successful accuracy during dichotic listening. Instead, for the condition where performance was poorer, effort increased.

2. The time course of pupil dilation for dichotic listening differs across symmetry groups, but not attended ear

It was hypothesized that symmetry could also affect dichotic listening effort, which would depend on attended ear. Lower effort for dichotic listening in the poorer ear was hypothesized for asymmetric listeners, who might give up when attending a more degraded signal in the presence of a clearer signal given poor performance (Fig. 1). The pupil response gives insight into auditory perception with ear asymmetry that could help to explain performance differences observed.

To examine the effects of symmetry group and attended ear, the dichotic condition data from all 20 listeners were collapsed for analysis to the first 4 s of the time window post-baseline. Collapsing the data allowed for the direct comparison of the effects of attended ear (better or poorer) and symmetry (symmetric or asymmetric) on the pupil response over time. There was no significant effect of attended ear or interaction between ear and symmetry on dichotic pupil size when ear was included in the parametric coefficients and smooth terms, so backwards elimination was used to simplify the model and remove these terms. Figure 3 shows the effect of symmetry on the pupil response over time.

FIG. 3.

Estimated pupil dilation in %DR for dichotic listening by symmetry (symmetric or asymmetric listeners). Time is post-baseline period, with the onset of the first digit pair occurring at 0 s. Vertical gray lines represent the onsets of each digit pair. The black horizontal lines indicate the time windows of significant differences from the GAMMs analysis (Table IV).

The GAMMs analysis final model (Table IV) showed that the asymmetric listeners had a lower dichotic pupil response for curve height (p = 0.03) and a significant difference in curve shape (p < 0.001) from the symmetric group. The times where these significant differences occurred across symmetry group are shown by the horizontal black line in Fig. 3, with higher estimated pupil dilation with symmetry. Pupil size increased over the time course for the symmetric listeners and was steadier over time for the asymmetric listeners. This difference in curve shape suggests that symmetric listeners dedicate additional cognitive resources to perception and retention of digits as more digits are presented, while asymmetric listeners hold these resources more constant over the time course of auditory perception.

TABLE IV.

GAMM summary for pupil data describing the effects of symmetry group (symmetric or asymmetric) on dichotic pupil size for data collapsed across number of digits and ear for the first 4 s of the time window post-baseline (Fig. 3). Significant parametric and smooth effects are shown in bold.

Parametric coefficients	Estimate	SE	t	p
Intercept (symmetric)	6.88	0.91	7.52	<0.001
Symmetry (asymmetric)	−3.56	1.66	−2.14	0.03
Smooth terms	edf	Ref.df	F	p
Time	8.58	8.62	56.11	<0.001
Time by symmetry (asymmetric)	2.02	2.34	8.92	<0.001
Pupil position (X, Y)	57.98	58.96	198.26	<0.001
Random smooth terms
Time × subject	168.28	179.00	81.47	<0.001

These results show that listeners with greater functional ear symmetry exhibited higher pupil dilation for dichotic listening. Interestingly, this pattern did not differ across ear, despite the large differences in performance across ear with asymmetry. Asymmetric listeners performed well when attending their better ear, but effort was not significantly different (per the initial model including ear terms) from attending to the poorer ear where performance was notably poor.

One observation is that the trajectory of listening effort over time in the asymmetric listeners is similar to that of the symmetric listeners during monotic listening (see Fig. 2). Therefore, this difference may reflect a difference in auditory perception for the asymmetric listeners, where processing of the interfering digits is reduced when attending the better ear. This could be due to central changes related to the functional asymmetry, where auditory processing prioritizes the historically more reliable ear, as can be seen in pediatric asymmetric hearing loss (e.g., Gordon et al., 2015). Performance for dichotic listening with attention directed to the better ear led to similar performance in asymmetric and symmetric listeners, but the trajectory of effort exerted was different with asymmetry. In the case of attending to the poorer ear, higher listening effort for symmetric listeners likely, at least in part, reflects the performance difference, where asymmetric listeners are performing at a level where they are likely to “give up” and disengage effort that is not able to help them overcome their poor performance (Ohlenforst et al., 2017).

3. Interaction between memory load and PM suggests that memory load dominates the pupil response at high demands

It was hypothesized that the effect of PM and DSL would interact such that greater digit string lengths would result in a larger effect of PM. To examine these effects, the retention interval of 3 s following the final digit and prior to the response was examined, with this analysis limited to the symmetric group data because this hypothesis was unrelated to symmetry effects. This time window was chosen because it is the time period where the listeners are holding the target digits in memory before responding, it has the same duration across digit string lengths, and the retention window is an analysis window used previously in this population (Winn and Teece, 2022).

Figure 4 shows the fitted data, where there was a significant interaction between DSL and PM (p = 0.01) for the parametric coefficient (Table V). This interaction is driven by the significantly greater pupil dilation for dichotic listening compared to monotic listening for four digits, but no significant effect of PM with six digits. Pupil dilation was always greater for six-digit retention compared to four-digit retention, with a significant difference across the entire retention time interval for both monotic and dichotic presentation modes (first row of Fig. 4).

FIG. 4.

Estimated pupil dilation for the symmetric listeners (n = 14) during the 3 s retention interval before digit recall for PM (monotic and dichotic) and DSL (four or six digits). Estimated pupil dilation is plotted in %DR. The black horizontal lines indicate the time windows of significant differences from the GAMMs analysis (Table V). D-6, dichotic, six-digit; D-4,  dichotic, four-digit; M-6, monotic, six-digit; M-4, monotic, four-digit.

TABLE V.

GAMM summary for pupil data describing the effects of PM and DSL on pupil size for data collapsed across ear for the retention interval (Fig. 4). Significant parametric and smooth effects are shown in bold.

Parametric coefficients	Estimate	SE	t	p
Intercept (monotic, four digits)	0.94	1.22	0.77	0.44
PM (dichotic)	4.54	0.82	5.54	<0.001
DSL (six digits)	10.24	1.51	6.76	<0.001
PM × DSL	−3.02	1.19	−2.53	0.01
Smooth terms	edf	Ref.df	F	p
Time	7.62	7.85	7.97	<0.001
Time by PM (dichotic)	5.15	5.57	1.97	0.08
Time by DSL (six digits)	4.70	5.15	6.16	<0.001
Time by PM × DSL	1.01	1.01	2.72	0.10
Pupil position (X, Y)	57.74	58.86	266.82	<0.001
Random smooth terms
Time × subject	120.59	139.00	11.47	<0.001
Time × subject by PM	111.40	139.00	5.49	<0.001
Time × subject by DSL	111.61	139.00	5.30	<0.001
Time × subject by PM × DSL	108.87	140.00	3.66	<0.001

This analysis shows that listeners with CIs have increased effort with retention of more items, as seen previously with normal hearing (e.g., Piquado et al., 2010). Additionally, there is greater engagement of cognitive resources during this retention period when listening involved ignoring interfering digits, a more demanding PM during auditory perception (see Fig. 2), but in a high memory load situation of remembering six digits, the difference between monotic and dichotic pupil size during the retention period is not significant. This shows that when memory load is high, the effort involved in retaining the digits dominates the response over PM effects. This is consistent with previous studies that have found that high memory load can obscure expected acoustic effects on effort (Zekveld et al., 2019; Zhang et al., 2021).

4. Working memory scores predict how listening effort is deployed over time

Working memory and inhibition cognitive scores were added, one at a time, as interaction terms with time to the model in Table V to examine whether these cognitive variables predicted pupil size, as hypothesized in this group of primarily older adults with CIs. Working memory scores had a significant effect on pupil size over time, but inhibition scores did not. To simplify the working memory model, ordered factors that did not interact with working memory were removed with backwards elimination. The simplified model summary is shown in Table VI, with a significant interaction between time and working memory scores on curve shape (p < 0.001). Figure 5 shows individual differences in working memory abilities, visualized with predicted pupil size over time at each quartile of working memory score. Individuals with the lowest and highest working memory scores visualized exerted similar changes in effort over time, with greater pupil size with lower working memory score during the analyzed retention period. Individuals with median working memory scores showed a different trajectory of deployment of effort over time, with low effort at the start of the retention period and increasing effort over the time course. These visualizations demonstrate the significant interaction between time and working memory scores; pupil size changes over time depended on these scores.

TABLE VI.

GAMM summary for pupil data describing the effects of working memory z-scores (WMZ) on pupil size for data collapsed across ear for the retention interval in the symmetric group. Significant parametric and smooth effects are shown in bold.

Parametric coefficients	Estimate	SE	t	p
Intercept	7.20	1.29	5.59	<0.001
Smooth terms	edf	Ref.df	F	p
a Time	5.62	5.68	9.29	<0.001
aWMZ	1.77	1.79	0.66	0.45
a Time × WMZ	5.20	5.30	4.88	<0.001
Pupil position (X, Y)	57.49	58.82	243.41	<0.001
Random smooth terms
Time × subject	238.60	278.00	10.62	<0.001

^a Included as (ti) terms rather than (s) terms to examine the interactions with working memory scores.

FIG. 5.

Estimated pupil dilation for the symmetric subjects (n = 14) during the 3 s retention interval before digit recall is plotted in %DR for each quartile of working memory z-scores.

There are several hypotheses about the relationship between cognitive resources and effort. The effort hypothesis is that individuals with greater cognitive capacity invest greater effort, regardless of task difficulty, to improve performance (Ahern and Beatty, 1979; van der Meer et al., 2010). The resource hypothesis is that individuals with greater cognitive capacity invest greater effort only on difficult tasks (Ahern and Beatty, 1979; van der Meer et al., 2010). The cognitive efficiency hypothesis is that individuals with greater cognitive capacity invest lower effort, regardless of task difficulty, due to more efficient processing (Ahern and Beatty, 1979; van der Meer et al., 2010). Finally, the ease of language understanding (ELU) hypothesis is that individuals with greater cognitive capacity invest lower effort only on difficult tasks (e.g., Rönnberg, 2003).

These results suggest that working memory impacts pupil size during this task, with individuals with poorer abilities using greater cognitive resources during digit retention across condition. This is most consistent with a cognitive efficiency hypothesis, that higher cognitive abilities lead to more efficient processing across conditions varying in difficulty (Ahern and Beatty, 1979; van der Meer et al., 2010). Additionally, the trajectory of pupil size changes suggests that higher working memory ability is associated with greater early investment of cognitive resources such that fewer resources are necessary later in the retention period, which suggests that general cognitive abilities relate to how effort is deployed over time.

IV. GENERAL DISCUSSION

This experiment investigated how sensory symmetry across inputs can impact directed auditory attention and perception. Cochlear implantation leads to varying functional asymmetry across ears, making this population a model of interest for studying asymmetry. CI listeners performed a simplified version of attending to a target talker amidst competing speech, where they attended to digits in one ear and ignored digits in the other ear. Target ear, PM (monotic or dichotic), and DSL (four or six digits) were manipulated to examine effects on performance and listening effort. Additionally, cognitive measures were used to examine whether these abilities predicted performance or listening effort. It was hypothesized that performance would decrease and effort would increase with dichotic listening compared to monotic listening. Ear symmetry was hypothesized to be associated with differences in dichotic listening performance and effort across ear. PM and DSL were hypothesized to interact such that greater DSL would lead to a larger effect of dichotic listening, with poorer performance and higher effort with greater DSL for dichotic listening. Finally, individual differences in cognitive abilities were hypothesized to be associated with poorer performance and higher effort with increased DSL.

As hypothesized, behavioral performance was worse (Fig. 1 and Table II) and effort was higher (Fig. 2 and Table III) for dichotic listening compared to monotic listening. Ear symmetry impacted dichotic listening performance (Fig. 1 and Table II) and effort (Fig. 3 and Table IV). An interaction between the introduction of an interferer and increased DSL was observed (Tables II and V). Contrary to the hypothesis, performance declined and effort increased to a lesser extent with greater DSL and interfering digits, suggesting that memory load demands of monotic listening increased to a greater extent than dichotic listening demands with greater DSL. Finally, given most of these listeners were middle aged or older, the impact of individual differences in cognitive abilities were of interest. Consistent with the hypothesis, better working memory abilities were associated with less listening effort later in the retention period (Fig. 5 and Table VI). Contrary to the hypothesis, working memory abilities did not predict performance. Inhibition scores were not significant predictors of performance or listening effort.

A. Listening effort revealed information about perception beyond performance measures

Pupil size was measured as an index of listening effort during monotic and dichotic listening in this study. The perceptual time window (during the auditory perception of the digits) and retention time windows (before recall) were examined to look at how functional asymmetries and memory load demands impact listening effort in CI listeners. Of particular interest was how the effort measures revealed aspects of perception not revealed by performance.

As observed previously with sentences (Koelewijn et al., 2014) and CI simulations (DeRoy Milvae et al., 2021), effort was significantly higher for dichotic compared to monotic listening during the perceptual time window, reflecting the cognitive effort involved in auditory attention toward the target and its processing, as well as the effort to ignore interfering speech. Functional performance symmetry impacted this dichotic listening effort, with greater effort for more symmetric listeners. There was no effect of attended ear on dichotic listening effort even though there were differences in performance. This differs from previous results with CI simulations, where attending to the poorer ear and ignoring the better ear led to greater pupil dilation than the reverse (DeRoy Milvae et al., 2021). Based on that finding, one would expect attending to the poorer ear in the asymmetric group to be the most effortful scenario, but this was not the case.

The results of the present study suggest that there is lower listening effort for asymmetric listeners overall, reflecting the degree of cognitive load when one auditory stream is less reliable. With functional asymmetry, it could be that the auditory system prioritizes the better ear and that there are obligatory processes to attend the better ear over the poorer ear. This is suggested by the time course of listening effort in this study for asymmetric listeners that resembles monotic listening effort in symmetric listeners. This phenomenon could not be observed in simulations of CI listening, since listeners with normal symmetric hearing have two ears that consistently provide high-quality information. When both ears are reliable, selective attention must be a factor in choosing which stream to attend; it could be that rather than applying an attenuated filter for selective attention, where unattended stimuli are processed to some degree (Treisman, 1960), earlier filtering occurs based on relative functionality of the two ears. This phenomenon may be an extension of perceptual load theory, which posits that the complexity of distractors drives attentional resources (Lavie, 1995). For asymmetric listeners, the reduced complexity of the input to the poor ear, with dissimilarity to the better ear in some way, such as spectral resolution differences, appears to lead to reduced processing load regardless of the attended ear. Future studies are needed to explore further this interplay between auditory processing and attention for listening to competing speech streams (Price and Moncrieff, 2021).

The retention time window was examined to test effects of DSL and PM. There was a significant interaction such that greater DSL increased effort for monotic and dichotic listening, and PM led to higher effort for dichotic listening with four digits but there was no effect of PM in the six-digit condition. This occurred despite the significant performance difference between monotic and dichotic listening in the six-digit condition. Again, this measure gave additional information about perception than the behavioral scores alone, demonstrating that high memory load leads to high cognitive demands that dominate the response. The strong significant effect of interfering digits no longer impacted listening effort when memory load was high, as seen previously when acoustic demands were not reflected in listening effort measures with high memory load (Zekveld et al., 2019; Zhang et al., 2021). This finding is also consistent with reaching a capacity limit on listening effort due to the high memory load (Kahneman, 1973; Pichora-Fuller et al., 2016), such that other factors that increase effort in less demanding situations no longer differentiate the response.

B. Cognitive abilities predicted differences in listening effort but not performance

It was expected that in this group of relatively older adults with more variability in cognitive abilities than younger adults (e.g., Schmiedek et al., 2009), that cognition would play a more substantial role in individual differences in performance and listening effort than in previous studies with younger adults. For behavioral performance, cognitive abilities of working memory and inhibition did not predict individual differences. It could be that the variability in performance with CIs due to other factors (Holden et al., 2013) made it unlikely to observe a significant effect of cognition on performance, especially with a limited sample size. However, working memory abilities did predict individual differences in listening effort. The results of this study were consistent with a cognitive efficiency hypothesis, since lower working memory scores were associated with greater listening effort across conditions of varying difficulty (for review of competing cognitive hypotheses, see Strand et al., 2018). This effect of working memory on listening effort was not observed previously with younger adults on the same dichotic listening task (DeRoy Milvae et al., 2021). This was likely due to the difference in populations, with less variability in cognition in a younger group with a smaller age range.

C. Dichotic listening models

Several models have been described to explain the factors that impact selective attention performance on dichotic listening. The shared resource model predicts that the difficulty of the task in the attended ear is related to how successful one is at ignoring the unattended ear (Brungart and Simpson, 2002). In other words, resources are dedicated to segregating the target in the attended ear and fewer resources are left to reduce interference from the unattended ear when the target ear segregation is more difficult. A second model is the integrated strategy model (Brungart and Simpson, 2007). This model differs from the first in that it does not assume that target ear segregation is always the preferred strategy, given that the interference does not always depend on the task difficulty in the attended ear (Kidd et al., 2003). Instead, the best strategy for segregation is employed, whether that is within-ear or across-ear. A third model is the relative salience model (Goupell et al., 2021), combining aspects of the two earlier models. This model assumes limited resources for segregation, but that multiple strategies can be employed with reduced interference when the segregation strategies are less demanding.

The data from the current experiment differ from the data from those other experiments for which these models were developed, such that asymmetric sensory inputs were involved in this sample and these models were based on performance data alone. With inclusion of pupillometry, there is an objective measure of cognitive resources at work to obtain the performance observed, giving greater insight into resource allocation described in these models. These results suggest expansion of the relative salience model. That model considered the salience of the inputs for the ease of dichotic listening, but these data suggest specifically that acoustic inputs do not always inform relative salience. The interplay between the peripheral signals and central processing should be considered in success for dichotic listening. For example, individuals with asymmetric hearing in this experiment had very poor ability to attend to their poorer ears when an interferer was present in their better ears, given their poor performance in that condition. The level of effort exerted was similar to when attending to their better ears, with the poorer ear input causing less interference. Effort was also lower for asymmetric listeners than symmetric listeners, even when performance level was similar, suggesting that the asymmetric listeners needed fewer cognitive resources to perform the task. This may occur because less cognitive effort is needed to segregate when there is diminished perception of the poorer ear. As previously suggested, the neural salience is an important component of the relative salience model (DeRoy Milvae et al., 2021).

D. Limitations and future directions

One limitation of this work is the relatively small sample size in the asymmetric listener group; in this clinical population, it is uncommon to find individuals with a >20% functional speech perception difference between ears. This limits the generalizability of the findings. However, analysis of trial-level data and time series data were used to strengthen the statistics, despite the small sample size. The findings suggest that functional asymmetry in CI listeners can have a strong impact on the ability to attend to a poorer ear and ignore a better ear, to a greater extent than is suggested by CI simulations (DeRoy Milvae et al., 2021). Future work should explore whether the findings in this study hold for a larger sample size, and whether there are other impacts of asymmetry; for example, are listeners with mid-range asymmetry more similar to symmetric or asymmetric listeners in performance and effort?

Additional exploration could also be conducted to best choose a cutoff between symmetric and asymmetric listeners. For this study, the cutoff of a 20% functional speech understanding difference in quiet was chosen based on a previous study (Mosnier et al., 2009). However, such a cutoff does not consider performance range; a fixed percentage difference does not take into account changing variability across the speech perception percentage range (Studebaker, 1985). This warrants future exploration, with a larger sample size to refine this criterion for functional asymmetry and to explore other methods outside of speech understanding in quiet scores. This additional work could also help to disentangle effects of asymmetry and overall ability.

Another future direction of this work is to investigate age as a possible explanation for functional asymmetry. Since previous CI simulations were completed with only young adults with normal symmetric hearing to examine the impacts of asymmetry, the possible contribution of age has not yet been examined. CI simulations in older adults could be a useful approach, since they control some factors that contribute to variability in performance, such as surgical and device-related factors, and allow for easy parametric manipulation of the stimuli (Cychosz et al., 2024).

The impact of level is another area that could be explored in this population, since relative salience with level changes across ear could impact dichotic listening effort. In this study, loudness balancing was used in an attempt to match loudness across ears, but this could not always be achieved and was a subjective process that assumed that clarity and other subjective perceptual differences across ears did not impact judgements. Effects of level are known to impact the pupil response, with increased pupil dilation to higher level stimuli (Liao et al., 2016; Zekveld et al., 2018). In most cases, adjustments in presentation level were small for loudness balancing and unlikely to have impacted the results. For both symmetric and asymmetric listeners, our analysis did not show significant differences with ear on listening effort, even for asymmetric listeners, again suggesting that these adjustments did not impact the pupillometry results.

Finally, this study examined listening effort for a simplified version of attending to target speech and ignoring interfering speech where target and interferers were presented to different ears. When listening with CIs to sound sources at different spatial locations in real life (e.g., multiple speakers surrounding a listener at a cocktail party; Cherry, 1953), this is never the case. Sound from the target and interferer(s) will reach both ears and the listener has the potential to improve their speech perception through the benefit from spatial cues. Therefore, examination of more realistic spatial configurations could provide insight into how effort is modulated by spatial cues and spatial-hearing benefits.

V. CONCLUSIONS

• •Functional asymmetry disrupts selective attention performance and effort, with greater differences in performance across attended ear and lower effort observed for more asymmetric auditory systems.
• •Functional ear symmetry, attended ear, memory load, and PM impacted performance on a listening task with directed attention in CI listeners.
• •Symmetric CI listeners had higher effort for dichotic listening (with the introduction of interfering digits) compared to monotic listening.
• •Symmetric CI listeners had higher dichotic listening effort than asymmetric CI listeners, possibly indicating reduced perception of the poorer ear in asymmetric listeners under dichotic listening situations.
• •Memory load and PM interacted to predict listening effort for symmetric CI listeners, with no effect of PM (monotic vs. dichotic) with higher memory load (six digits).
• •CI listeners with poorer working memory scores experienced higher listening effort over time.

SUPPLEMENTARY MATERIAL

See the supplementary material for Supplemental Table I.

AUTHOR DECLARATIONS

Conflict of Interest

The authors have no conflicts to disclose.

Ethics Approval

Study procedures were reviewed and approved by the Institutional Review Board at the University of Maryland-College Park. All participants provided informed consent.

DATA AVAILABILITY

Deidentified data are available from the corresponding author upon request.