Acoustic hearing can interfere with single-sided deafness cochlear-implant speech perception

Abstract

Objectives.

Cochlear implants (CIs) restore some spatial advantages for speech understanding in noise to individuals with single-sided deafness (SSD). In addition to a head-shadow advantage when the CI ear has a better signal-to-noise ratio, a CI can also provide a binaural advantage in certain situations, facilitating the perceptual separation of spatially separated concurrent voices. While some bilateral-CI listeners show a similar binaural advantage, bilateral-CI listeners with relatively large asymmetries in monaural speech understanding can instead experience contralateral speech interference. Based on the interference previously observed for asymmetric bilateral-CI listeners, this study tested the hypothesis that for SSD-CI listeners in a multiple-talker situation, the acoustic ear would interfere with rather than improve CI speech understanding.

Design.

Experiment 1 measured CI-ear speech understanding in the presence of competing speech or noise for 13 SSD-CI listeners. Target speech from the closed-set coordinate response-measure corpus was presented to the CI ear along with one same-gender competing talker or stationary noise, at target-to-masker ratios (TMRs) between −8 and 20 dB. The acoustic ear was presented with silence (monaural condition) or with a copy of the competing speech or noise (bilateral condition). Experiment 2 tested a subset of 6 listeners in the reverse configuration for which SSD-CI listeners have previously shown a binaural benefit (target and competing speech presented to the acoustic ear, silence or competing speech presented to the CI ear). Experiment 3 examined the possible influence of a methodological difference between experiments 1 and 2: whether the competing talker spoke keywords that were inside or outside the response set. For each experiment, the data were analyzed using repeated-measures logistic-regression. For experiment 1, a correlation analysis compared the difference between bilateral and monaural speech-understanding scores to several listener-specific factors: speech understanding in the CI ear, pre-implantation duration of deafness, duration of CI experience, ear of deafness (left/right), acoustic-ear audiometric thresholds, and listener age.

Results.

In experiment 1, presenting a copy of the competing speech to the acoustic ear reduced CI speech-understanding scores for TMRs ≥ 4 dB. This interference effect was limited to competing-speech conditions and was not observed for a noise masker. There was dramatic inter-subject variability in the magnitude of the interference (range: 1–43 rationalized arcsine units), which was found to be significantly correlated with listener age. The interference effect contrasted sharply with the reverse configuration (experiment 2), whereby presenting a copy of the competing speech to the contralateral CI ear significantly improved performance relative to monaural acoustic-ear performance. Keyword condition (experiment 3) did not influence the observed pattern of interference.

Conclusions.

Most SSD-CI listeners experienced interference when they attended to the CI ear and competing speech was added to the acoustic ear, although there was a large amount of inter-subject variability in the magnitude of the effect, with older listeners particularly susceptible to interference. While further research is needed to investigate these effects under free-field listening conditions, these results suggest that for certain spatial configurations in a multiple-talker situation, contralateral speech interference could reduce the benefit that an SSD-CI otherwise provides.

INTRODUCTION

Individuals with single-sided deafness (SSD; one deaf ear and one ear with normal or near-normal hearing) are at a distinct disadvantage for both auditory spatial awareness and speech communication (Firszt et al. 2017). Cochlear implants (CIs) given to individuals with SSD restore some of the benefits provided by two functional ears, including improved sound localization (Erbele et al. 2015; Zeitler et al. 2015) and speech understanding in noise (Bernstein et al. 2017; Vermeire & Van de Heyning 2009). The reported speech-understanding benefits are mostly attributable to head-shadow effects, with the CI providing an advantage specifically in target-masker spatial configurations where a more favorable signal-to-noise ratio (SNR) is produced on the side of the head with the CI (Arndt et al. 2011). There is also some evidence that SSD-CI listeners may receive a small benefit from binaural summation and redundancy (Arndt et al. 2017). While head-shadow and binaural-summation advantages do not depend on binaural interactions (i.e., the computation of binaural differences in the signals arriving at the two ears), there is evidence that SSD-CI listeners can take advantage of binaural interactions to better understand speech when it is masked by competing speech that originates from a different spatial location (Bernstein et al. 2016; 2017).

Similarly to SSD-CI listeners, bilateral-CI (BI-CI) listeners also benefit from a head-shadow and binaural summation to better understand speech in noise (Laszig et al. 2004; Schleich et al. 2004). Most studies show little evidence that BI-CI listeners are able to make use of binaural interactions to better understand speech in noise (van Hoesel et al. 2008; Loizou et al. 2009; Reeder et al. 2014; Goupell et al. 2016), although Bernstein et al. (2016) found that a group of high-performing BI-CI listeners showed some ability to take advantage of binaural interactions to perceptually separate competing spatially separated talkers. However, this advantage was not universally observed. Bernstein et al. (2016) reported that for one out of the 9 BI-CI listeners tested, the second CI not only failed to improve speech understanding in a situation involving concurrent competing speech, but it also generated substantial interference relative to the monaural condition. In a follow-up study, Goupell et al. (2018) found a similar interference effect for a group of BI-CI listeners who either had substantial asymmetry in the duration of deafness before implantation in the two ears, or who were prelingually or perilingually deafened and received their CIs in adulthood. This raises the possibility that rather than providing the benefit observed for the high-performing BI-CI listeners in the Bernstein et al. study, having access to sound in both ears can sometimes be detrimental to speech understanding.

The findings of Goupell et al. (2018) raise an important potential concern for adults with SSD who receive a CI. These individuals also have a dramatic asymmetry in hearing acuity, with one normally or near-normally functioning ear, and the other ear receiving the relatively crude speech signal associated with CI processing and electrical stimulation. There is little evidence that SSD-CIs cause interference, with only anecdotal and subjective reports for a small number of individual patients within a larger cohort (Firszt et al. 2012). However, the results of Goupell et al. (2018) for asymmetric BI-CI listeners raise the possibility that the introduction of competing speech information in the acoustic ear could interfere with speech understanding in the CI ear. If such interference were to occur, it is possible that the overall spatial-hearing benefit, derived primarily from head shadow, could be reduced.

To examine how the introduction of competing speech to the acoustic ear would affect speech understanding through the CI ear, a series of 3 experiments were carried out that employed the contralateral-unmasking paradigm for SSD-CI listeners employed by Bernstein et al. (2016). In that study, target and competing talkers spoke sentences from the coordinate response-measure (CRM) corpus (Bolia et al. 2000; Brungart, 2001), consisting of a closed set of color and number keywords. In the monaural condition, a mixture of target and competing speech was presented via headphones to the acoustic ear. Then in the bilateral condition, a copy of the competing speech was presented to the CI ear via direct audio input to the sound processor. This led to an improvement in speech understanding relative to the monaural condition, termed contralateral unmasking. The improvement was attributed to a binaural-interaction advantage provided by the CI: with no target speech information presented to the CI ear, there could not have been any head-shadow benefit.

Experiment 1 tested the hypothesis that SSD-CI listeners attending to their CI ear would experience interference when competing speech was presented to their acoustic ear. This experiment employed the same contralateral-unmasking paradigm as Bernstein et al. (2016), but with one key difference. Rather than presenting the target and competing speech monaurally to the acoustic ear, the target and competing speech were instead presented monaurally to the CI ear, and the effect of adding a copy of the competing speech in the acoustic ear was evaluated.

Experiment 2 was designed to replicate the study of Bernstein et al. (2016) that found a contralateral-unmasking advantage when SSD-CI listeners attended to the acoustic ear. This experiment sought to determine whether the specific listeners tested here would exhibit a similar advantage. Experiment 3 investigated whether a specific methodological difference between the first 2 experiments influenced the results. In experiment 1, the interfering talkers spoke words that were not part of the closed set of response options (out-of-set maskers) to prevent the listener from using any information from the competing speech to reduce the possible set of correct responses. In experiment 2, the competing talkers spoke words that were response options (in-set maskers), which was done to replicate the large contralateral-unmasking advantage reported by Bernstein et al. (2016). To investigate the possible influence of this procedural difference, Experiment 3 repeated experiment 1 using in-set instead of out-of-set maskers.

EXPERIMENT 1: TARGET SPEECH IN THE COCHLEAR-IMPLANT EAR

Materials and methods

Listeners

Thirteen SSD-CI listeners (3 female) participated in this experiment. Demographic data and information regarding their CI ears are provided in Table 1. The 13 listeners used MED-EL (Flex28; Innsbruck, Austria) or Cochlear Ltd.^* (CI24RE, CI422, CI512, or CI522; Sydney, Australia) CIs, and had a wide range of ages at the time of the study testing (36–74 years, mean 56.3 years), durations of deafness prior to implantation (0.25–24 years, mean 10.0 years), years of CI experience (0.7–7.0 years, mean 2.9 years), and etiologies. Audiometric thresholds and related data for listeners’ acoustic-hearing ears are provided in Table 2. Ten of the listeners had normal or near-normal hearing in the acoustic ear, with pure-tone thresholds of 25 dB HL or better at audiometric frequencies between 250 and 4000 Hz, and would be classified by Vincent et al. (2015) as having SSD. Three of the listeners (S2, S4, and S13) had mild-to-moderate hearing loss, with maximum thresholds of 30–55 dB HL in this frequency range. These 3 listeners wore a hearing aid in their acoustic ear for everyday listening, but removed their hearing aid for the tests carried out in this study. Two of these listeners (S2 and S13) would be classified by Vincent et al. (2015) as having asymmetric hearing loss (pure-tone average at 500, 1000, 2000, and 4000 Hz >30 but ≤60 dB HL), while the listener with milder hearing loss (S4) would be classified as having SSD (pure-tone average ≤30 dB HL).

Table 1.

Demographic and CI-ear data for the study participants

Listener	Sex	Age	CI	Duration	CI use	IEEE	Electrode	Etiology	Experiment
			ear	of deafness (years)	(years)	score (%)	array		1	2	3
S1	M	60	L	21	4.6	83	CI24RE	Sudden SNHL	X
S2	F	56	L	2	2.7	DNT	CI422	Sudden SNHL	X
S3	M	36	R	0.25	5.3	93	Flex28	Sudden SNHL	X	X	X
S4	M	60	R	8	1.6	22	Flex28	Sudden SNHL	X
S5	M	47	L	15	2.9	DNT	CI24RE	Endolymphatic shunt surgery	X
S6	M	46	L	20	0.7	0	CI512	Skull fracture	X	X	X
S7	M	46	R	22	4.7	DNT	Flex28	EVA; Head trauma	X
S8	M	58	L	1	0.8	88	Flex28	Labyrinthitis	X	X	X
S9	F	74	L	5	3.0	72	Flex28	Sudden SNHL	X	X	X
S10	F	63	L	24	0.9	64	Flex28	Ménière’s	X	X	X
S11	M	65	L	7	2.6	33	Flex28	Unknown	X	X	X
S12	M	69	R	0.75	7.1	90	CI512	Unknown	X
S13	M	52	L	3	0.9	82	CI522	Sudden SNHL	X

CI, cochlear implant; DNT, did not test; EVA, enlarged vestibular aqueduct; SNHL, sensorineural hearing loss

Table 2.

Acoustic-ear data for the study participants

			Audiometric threshold (dB HL)
Listener	250	500	1000	2000	3000	4000	6000	8000 Hz	IEEE score (%)	Hearing aid?	Speech level (dB SPL)
S1	10	10	0	15	15	15	20	20	100	No	65
S2	25	30	35	45	DNT	55	65	90	DNT	Yes	65
S3	10	15	15	0	10	20	20	25	100	No	60
S4	15	10	10	10	30	30	20	15	99	Yes	55
S5	15	10	5	0	10	15	25	25	DNT	No	55
S6	10	5	10	0	0	10	10	10	100	No	60
S7	5	10	10	15	25	25	15	20	DNT	No	55
S8	5	10	15	10	15	20	20	15	99	No	60
S9	15	15	10	5	10	10	15	20	99	No	60
S10	15	25	20	20	20	15	25	25	99	No	60
S11	20	20	20	10	10	20	15	40	97	No	55
S12	10	10	20	15	10	20	DNT	25	100	No	55
S13	20	20	30	40	45	35	25	45	99	Yes	55

DNT, did not test

Goupell et al. (2018) found that the magnitude of contralateral speech interference for an individual BI-CI listener was related to their monaural speech-understanding scores in quiet. Therefore, monaural speech understanding in quiet was evaluated separately for each ear for the SSD-CI listeners in the current study. Listeners were presented with Institute of Electrical and Electronic Engineers (IEEE) sentences (Rothauser et al. 1969) (e.g., “The birch canoe slid on the smooth planks” and “Open the crate but don’t break the glass”). For the acoustic-hearing ear, the speech was presented over one earpiece of a pair of HD280 headphones at a level of 65 dB SPL. For the CI ear, the stimuli were delivered directly to the sound processor via an auxiliary input cable. The level of the CI stimulus was loudness balanced to the stimulus in the acoustic ear by playing a series of pairs of sequential sentences to the two ears, with the experimenter adjusting the level of the CI stimulus upwards or downwards in 1-or 2-dB steps until the listener reported the stimuli being perceived as equally loud. Three lists were presented for each ear (6 lists total) in random order. Listeners were required to repeat back verbally the sentence they heard to the best of their ability. The experimenter then used a graphical user interface (GUI) to mark which of the 5 keywords in each sentence were correctly identified. The speech-understanding score was calculated for each ear by determining the percentage of correctly identified keywords out of the 150 (5 keywords times 30 sentences) presented to each ear. Because the IEEE sentence test was added to the study after the first several listeners were tested, only 10 of the 13 listeners participated. There was a wide range of sentence keyword-identification scores for the CI ear (0–93%, mean 62.7%, Table 1), whereas sentence keyword-identification scores for the acoustic ear were high (97– 100%, Table 2) for all 10 of the listeners who were tested.

Stimuli

The speech stimuli for the main experiment consisted of CRM sentences (Bolia et al. 2000), which have the form “Ready <call sign> go to <color> <number> now”, and were spoken by 8 different talkers (4 male, 4 female). There were 8 possible call signs (e.g., “Baron,” “Charlie,” “Arrow”). The target talker always used the call sign “Baron,” while the competing talker used other call signs. The target talker used 8 numbers (1–8) and 4 colors (“red,” ”green,” ”blue,” and “white”). The competing talker used colors (“black” or “brown”) and numbers (“nine” or “ten”) that were not part of the response set (i.e., out-of-set maskers; Iyer et al. 2010). This is in contrast to the traditional implementation of the CRM-based speech-on-speech masking task (Brungart 2001; Brungart et al. 2001), whereby on any given trial, the competing talkers produce color and number keywords that differ from the target talker, but are nevertheless selected from the available set of possible responses (i.e., in-set maskers). The reason for this choice in the current experiment was to avoid the possibility that listeners could reduce the possible set of correct responses by attending to the competing speech in the acoustic ear (Bernstein et al. 2016). The stimuli employed were the original recordings described by Iyer et al. (2010).

The target speech was always presented to the CI ear. In the monaural condition, the masker was also presented to the CI ear. In the bilateral condition, the masker was presented to both the CI and acoustic ears. The benefit (or interference) provided by the acoustic ear was characterized by subtracting monaural from bilateral performance. Two masker types were tested. The competing-speech masker was produced by a different talker of the same gender as the target talker. The target and competing speech started together and were closely time aligned to the extent that each recorded CRM sentence was similar in length. The stationary-noise masker was produced by selecting a speech masker as in the competing-speech condition, taking its fast-Fourier transform (FFT), randomizing the phases, and performing an inverse FFT. This produced a speech-shaped noise whose spectrum varied from trial to trial depending on the characteristics of the speech token from which it was derived.

The acoustic stimuli were delivered via one channel of a Hammerfall Multiface II (RME Audio, Haimhausen, Germany) sound card to one earpiece of a pair of HD280 closed circumaural headphones (Sennheiser, Wedemark, Germany). The CI stimuli were delivered via the other sound-card channel to the direct audio input to the sound processor. Listeners used their regular everyday clinical sound-processor map loaded onto a Nucleus 6 (Cochlear Ltd., Sydney, Australia) or Opus 2 (MED-EL, Innsbruck, Austria) sound processor kept in the laboratory for research purposes. The Cochlear Ltd. listeners all used the Advanced Combination Encoder (ACE) strategy, while the MED-EL listeners used a variety of signal-processing strategies. The Cochlear Ltd. listeners used a variety of front-end processing features, with 4 of the listeners (S2, S5, S6, and S13) using all available features [automatic scene classifier system (SCAN), adaptive dynamic range optimization (ADRO), signal-to-noise ratio noise reduction, and wind noise reduction], one listener (S12) using only ADRO, and one listener (S1) using none of these features. The MED-EL system does not have front-end features available for selection in the sound processor. While it is likely that in a free-field configuration, the activation of the front-end processing features might influence spatial cues through the activation of beam-forming technology, the possible influence of beam-formers was limited in the current study where the signals were presented directly to the sound processor using a direct audio-input cable. Although the delay of the sound processor was not measured directly, Wess et al. (2017) reported that the CI signal is expected to be delayed relative the acoustic-ear signal by 10.5–12.5 ms for Cochlear Ltd. devices (Van Dijk, private communication) or 0.5–1.6 ms for MED-EL devices (Zirn et al. 2015). It should be noted that vocoder simulations show little effect of interaural delay on contralateral unmasking of speech for the same paradigm employed in the current study until the delays exceed 24 ms (Wess et al. 2017). The immunity of the unmasking effect to interaural delay was interpreted as reflective of the echo threshold for speech stimuli, where an echo is perceived as part of the same auditory object for delays as long as 30–40 ms (Litovsky et al. 1999).

Listeners were told that the target speech would always be in their CI ear and that they should ignore any speech presented to their acoustic ear. They were also told that they should ignore the competing speech that included call signs other than “Baron,” and color and number choices that were not in the response set (“black,” “brown,” “nine,” and “ten”).

Procedure

On each trial, listeners were presented with a monaural or bilateral mixture of target speech and masker. They were instructed to listen for the talker speaking the call sign “Baron,” and report back the color and number spoken by this talker. The response was made using a GUI consisting of 32 virtual buttons, with the 8 buttons in each row colored to represent the response color, and the 4 buttons in each column containing a numeral representing the response number. Following each response, feedback was provided with the response button corresponding to the correct answer blinking 3 times.

Listeners were tested at 8 different target-to-masker ratios (TMRs; −8 to +20 dB in 4-dB steps) for all 4 combinations of ear configuration (monaural or bilateral) and masker type (speech or noise). To achieve the desired TMR, the level of the target speech was held fixed and the masker level was adjusted. The level for the target speech presented to the CI ear was set to a fixed level for each listener using a sequential interaural loudness-balancing procedure. The sound-processor volume control was set to the maximum level, and the voltage level of the input signal was adjusted to match the loudness of a 60 dB SPL speech signal presented to the acoustic ear. This was done rather than adjusting the level using the sound-processor volume control to ensure the maximum electrical dynamic range of the speech signals presented to the CI ear. Two CRM sentences were selected at random from the target speech set and presented sequentially, first to the CI ear and then to the acoustic ear. The level of the CI stimulus was adjusted upward and downward in 1-to 3-dB steps until the listener reported that the stimuli were equal in loudness. Once the target-speech level was established, the next step was to check that the masker stimulus delivered to the CI was audible at the lowest masker level (i.e., −20 dB for the highest TMR tested) and not uncomfortably loud at the highest masker level (i.e., +8 dB for the lowest TMR) that would be tested in the experiment. The level of this stimulus was adjusted over this range and the listener was asked to subjectively report whether the stimulus was audible and not uncomfortably loud across this range. If necessary, the process was repeated with the acoustic stimulus held fixed at a higher (65 dB SPL) or lower (55 dB SPL) stimulus level. The final target speech levels for the acoustic ear for each listener (to which the target speech level in the CI ear was loudness matched) are shown in Table 2.

The stimuli were presented in blocks of 64 trials, with the ear configuration and masker type held fixed within each block, but with the TMR varying randomly from trial to trial. Depending on how much time each listener had available to complete the experiment, listeners completed at least 2 blocks, and in most cases 3 blocks, for each combination of ear configuration and masker type, for a total of 16 or 24 trials per condition, and 512 or 768 trials for the experiment. Blocks were presented in pseudorandom order, with one block completed for all 4 combinations of ear configuration and masker before the next block was initiated for any combination.

Before the main experiment, listeners completed a set of 4 training blocks (one block for each combination of ear configuration and masker type). Each training block consisted of 24 trials: 8 trials for each of 3 TMRs (0, 8, and 16 dB). As in the main experiment, feedback was provided. Including the training blocks, the experiment took 1.5–2 hours for each listener to complete.

Analysis

Repeated-measures logistic-regression analysis (RM-LRA) was employed to investigate the effects of ear configuration (monaural or bilateral), masker type (noise or speech), TMR, and their associated interactions. An RM-LRA is similar in concept to a repeated-measures analysis of variance (RM-ANOVA) in that the analysis provides an assessment as to which factors significantly influenced the results and which combinations of factors interacted. The main difference between the RM-LRA and a RM-ANOVA is that, like a binomial test, the RM-LRA accounts for the total number of trials that were completed for each listener as well as the dependence of the variance on the performance score. For example, because the variance for a 95% score is much smaller than for a 60% score, the analysis would be more likely to identify a change from 95 to 100% as a significant effect than a change from 60 to 65% correct. Where significant interactions were present, post-hoc comparisons (χ² test for the difference between 2 proportions) were carried out to identify the specific subset of conditions for which ear configuration (bilateral vs. monaural presentation) had a significant influence on performance. Bonferroni corrections for multiple comparisons were applied by dividing the standard criterion p-value (p<0.05) by the number of TMR conditions tested (i.e., p<0.00625=0.05/8).

Results

Figure 1 shows the group-mean results for experiment 1 (target speech presented to the CI ear). The vertical axis represents the mean proportion of keywords correctly identified (out of 2 for each sentence, i.e., color and number). The results clearly show that for the competing-talker condition (Fig. 1A), listeners experienced considerable contralateral speech interference (i.e., a decrease in speech understanding in the bilateral relative to the monaural condition) when a copy of the competing speech was presented to the acoustic ear. In contrast, there was no effect of adding a copy of the stationary-noise masker to the acoustic ear (Fig. 1B). The results of the RM-LRA supported these observations. There were significant main effects of TMR [χ²(7)=576, p<0.0001], masker type [χ²(1)=19.1, p<0.0001], and ear configuration [χ²(1)=21.8, p<0.0001]. There was a significant two-way interaction between TMR and masker type [χ²(7)=135, p<0.0001], consistent with the observation that the slope of the performance function was steeper for stationary noise than for the competing talker. There were also significant two-way interactions between TMR and ear configuration [χ²(7)=19.9, p=0.006] and between masker type and ear configuration [χ²(1)=23.1, p<0.0001], as well as a significant interaction between the 3 factors [χ²(7)=40.6, p<0.0001]. These interactions suggest that the difference between monaural and bilateral performance varied with masker type and TMR. In particular, post-hoc tests (Bonferroni-corrected for 8 comparisons) showed that significant interference occurred (i.e., performance was significantly poorer in the bilateral case, asterisks in Fig. 1) with a competing talker for TMRs of 4 dB (p=0.006), 8 dB (p<0.0001), 12 dB (p<0.0001), 16 dB (p<0.0001), and 20 dB (p=0.0003). In contrast, no significant differences between monaural and bilateral presentation occurred at any TMR for the noise masker (p>0.00625).

Fig. 1.

Mean data from experiment 1 showing performance for masked speech identification in the CI ear (monaural condition) and the effect of presenting a copy of the masker to the acoustic ear (bilateral conditions) with (a) one same-gender competing talker or (b) a stationary-noise masker. Asterisks indicate conditions where performance was significantly different between the monaural and bilateral conditions (Bonferroni-corrected criterion p<0.00625). Error bars indicate ± 1 standard error of the mean.

To examine whether the different types of sound processors employed by the listeners in the study could have influenced the results, the data for the Cochlear Ltd. and MED-EL listeners were also analyzed with separate RM-LRA analyses. Both analyses (not shown) yielded the same pattern of results as the main analysis. All main effects and interactions involving TMR, ear configuration, and masker type were found to be significant, and Bonferroni (p<0.00625) post-hoc analyses identified significantly decreased performance with bilateral presentation for the interfering-talker condition at high TMRs. The only difference was that the Cochlear listeners did not show a significant effect at a TMR of +20 dB (p=0.037), while the MED-EL listeners showed a significant effect at 0 dB (p=0.003). Thus, processor type had no bearing on the overall pattern of results.

While Figure 1 shows that, on average, the SSD-CI listeners experienced interference, there was considerable intersubject variability in the magnitude of the interference. Figure 2 shows the results for the competing-talker condition for each individual listener in the study, with the individual listeners ordered in panels A-M from the least to most interference. Figure 3 shows that the inter-subject variability in the magnitude of the interference mainly reflected variability in the bilateral condition and not in the monaural condition. The left column (Figs. 3A and 3C) plots the fitted psychometric functions [error function (erf) in Matlab with 3 free parameters describing the slope, intercept, and maximum plateau value] for all 13 SSD-CI listeners for the competing-talker conditions shown in Figure 2. The right column (Figs. 3B and 3D) shows the functions for the stationary-noise masker conditions (individual data not shown in Fig. 2). There was some degree of inter-subject variability in performance in the monaural competing-talker conditions (Fig. 3A) and in both the monaural (Fig. 3B) and bilateral stationary-noise conditions (Fig. 3D), as would be expected for a group of CI listeners (e.g., Gifford et al. 2008). The variability, however, was most dramatic in the bilateral competing-talker condition (Fig. 3C). If the one listener who showed much poorer monaural speech understanding than the other 12 listeners in the monaural conditions (S6; Fig. 2F and dashed curves in Fig. 3) is excluded from consideration, maximum performance at the highest TMR tested ranged from 88–100% for the monaural conditions (Figs. 3A and 3B) and for the bilateral stationary-noise condition (Fig. 3D). In contrast, there was a much larger range in performance for the bilateral competing-talker condition (Fig. 3C), with maximum performance ranging from 53–100%. The fact that there was relatively little performance variability in the monaural CI-alone conditions suggests that the differences in the magnitude of interference are not likely to be attributable solely to differences in monaural CI speech understanding.

Fig. 2.

Individual data for the competing-talker condition in experiment 1. The individual listeners are ordered based on the magnitude of interference experienced. Error bars indicate ± 1 standard error of the mean.

Fig. 3.

Fitted psychometric functions to the data for each individual listener for (a) the monaural and (c) the bilateral competing-talker conditions, and for (b) the monaural and (d) the bilateral noise-masker conditions. The dashed curves represent the one outlier with overall poor performance (listener S6; Fig. 2F).

The observation of large individual differences in the magnitude of the interference suggested that it would be worthwhile to conduct a correlation analysis to identify factors that might relate to the interference. Based on the group-mean results showing a statistically significant interference effect at the highest TMRs tested (4–20 dB), the magnitude of the interference was defined for each individual listener as the average interference across these 5 TMRs. The performance scores were first converted to rationalized arcsine units (rau; Studebaker 1985) to equalize the variance and offset ceiling effects for scores greater than 80% correct. Then, the difference between monaural and bilateral performance was calculated. Note that the rau transform was not applied for the RM-LRA conducted on the group data (Fig. 1) because this binomial analysis already took these effects into account. This rau transformation was only applied to calculate the magnitude of the interference for each individual listener for the purposes of the correlation analysis, and was done specifically to address the fact that a given percentage-correct change in performance at a high performance level (e.g., from 90 to 98%) is more meaningful than the same change at a lower performance level (e.g., from 70 to 78%). The magnitude of the interference for each individual listener (reported in each panel of Fig. 2) ranged from 1–43 rau.

The magnitude of the interference effect was first compared to 3 factors to specifically test the hypothesis, based on the results of Goupell et al. (2018), that the magnitude of any observed interference would be inversely correlated with the acuity of the CI ear. Specifically, it was expected that listeners with poorer CI-ear speech understanding or longer duration of deafness would be more susceptible to interference. Monaural word recognition in the CI ear for IEEE sentences in quiet and for CRM sentences in stationary noise at a 20-dB TMR collected in the main experiment were examined, along with the duration of deafness (years between severe-to-profound hearing loss and CI activation). Because these 3 relationships were specifically hypothesized based on the previous results for BI-CI listeners, no Bonferroni corrections were applied (Keppel & Wickens 2004). Four additional factors were examined that were not specifically hypothesized to relate to performance, with Bonferroni corrections applied for 4 comparisons (criterion p<0.0125 = 0.05/4): (1) the degree of hearing loss in the acoustic-hearing ear as defined by the pure-tone average threshold (500, 1000, 2000, and 4000 Hz); (2) the effect of ear of implantation (right or left); (3) the duration of CI experience (time between CI activation and the experiment); and (4) listener age at the time of testing. For all but one of the factors, a simple Pearson correlation was computed between each factor and the magnitude of the effect of binaural presentation. For the ear of implantation (a binary variable), an unpaired t-test was conducted between the left-and right-ear subgroups.

Of the 7 factors examined, only listener age showed a significant correlation with the magnitude of the interference [R² = 0.46, p=0.011]. The p-values for the other 6 factors ranged from 0.41–0.99. The correlation between listener age and the magnitude of the interference remained significant (p=0.011) after partialling out any possible differences attributable to processor type.

Figure 4 shows scatterplots for 4 of the factors that were examined. Qualitatively, these data suggest that the lack of significant correlations between the magnitude of the interference and monaural speech-understanding scores for the CI ear (Figs. 4A and 4B) or duration of deafness (Fig. 4C) were not simply because of the small sample size and a lack of statistical power. Individuals with both high and low speech scores showed both large and small amounts of interference (Fig. 4B). Similarly, individuals with both short (<5 years) and long (≥20 years) durations of deafness showed both large and small amounts of interference (Fig. 4C). This was even the case when the one listener with very low monaural CI speech scores was ignored. In contrast, Figure 4D suggests the presence of an aging effect, whereby the younger listeners in the study (<50 years) all showed relatively little interference, while the older listeners in the study (>60 years) tended to show much greater interference.

Fig. 4.

Relationship between interference magnitude for individual listeners and their monaural speech-identification performance for (a) CRM sentences in noise for a 20-dB TMR, (b) IEEE sentences in quiet, (c) duration of deafness before CI activation, and (d) age.

In summary, experiment 1 showed that SSD-CI listeners attending to their CI ear experienced interference when a copy of the competing speech was added to the acoustic-hearing ear, but not when a copy of the stationary-noise masker was added to the acoustic-hearing ear. There was considerable inter-subject variability in the interference effect (1–43 rau), which was found to correlate with listener age.

EXPERIMENT 2: TARGET SPEECH IN THE ACOUSTIC EAR

Materials and methods

Rationale

Experiment 1 found that SSD-CI listeners experienced interference when the target speech was presented to the CI ear. Experiment 2 examined the reverse configuration, evaluating whether the particular SSD-CI listeners in the current study experienced the same contralateral-unmasking benefit reported by Bernstein et al. (2016) when the target speech was presented to the acoustic ear. Because Goupell et al. (2018) found that the presence of a benefit or interference using this paradigm depended greatly on the particular characteristics of the individual BI-CI listeners, it was important to test whether the findings of Bernstein et al. for SSD-CI listeners would be replicated for these individuals.

Listeners

Because this experiment was added to the study after it became apparent from the initial cohort of listeners that experiment 1 would show a substantial interference effect, only 6 of the 13 listeners from the first experiment were tested in experiment 2 (X’s in Table 1).

Procedure

The methods generally followed Experiment 1 except that the target speech was presented to the acoustic ear instead of to the CI ear, and the competing speech was presented either just to the acoustic ear (monaural condition) or to both ears (bilateral condition). Only speech maskers were tested. There were two methodological differences in experiment 2. First, there were two competing talkers of the same gender as the target talker instead of the single competing talker employed in experiment 1. This was done because with the target presented to the acoustic ear instead of the CI ear, performance would have been close to ceiling over a wide range of TMRs with a single competing talker. Second, the competing talkers used color and number keywords that formed part of the response set (i.e., in-set maskers) instead of color and number keywords that were not part of the response set (i.e., out-of-set maskers) employed in Experiment 1. The sentence selection was constrained on each trial such that all 3 concurrent talkers used different call signs and color and number keywords. This was done because Bernstein et al. (2016) found that this yielded the greatest amount of benefit from the CI ear. The key reason for using out-of-set maskers in experiment 1 was that in the bilateral condition, listeners had access to the highly intelligible, unmasked competing speech in the acoustic ear. With in-set maskers, they potentially could have used this information to reduce the set of possible response options that could have been spoken by the target talker. This was less of a concern in experiment 2 because in the bilateral condition, listeners only had access to the much less intelligible concurrent interferers in the CI ear. Note that the effect of in-set versus out-of-set maskers was explicitly examined in experiment 3 (see below).

Listeners were presented with blocks of 40 trials each. In each block, the ear configuration was held fixed (monaural or bilateral presentation), but the TMR varied randomly from trial to trial (range: −8 to +8 dB in 4-dB steps). The TMR was defined as the level of the target speech relative to the level of each competing talker. Thus, the SNR (target level relative to the level of the combined competing speech) was 3 dB lower than the TMR. There were a total of 6 blocks (3 blocks for each ear-configuration condition), for a total of 24 trials per condition and 240 trials overall. Blocks were presented in pseudorandom order, with one block completed for both ear configurations before the next block was initiated for either ear configuration. As in experiment 1, feedback was provided after each response. No additional training was provided before this experiment. The experiment took approximately 30 minutes for each listener to complete.

Analysis

RM-LRA was employed to examine the effects of ear configuration and TMR and their interaction. Post-hoc comparisons, Bonferroni corrected for the 5 TMRs tested (i.e., p<0.01=0.05/5), examined the effect of ear configuration at each TMR.

Results

Figure 5A shows the mean results for the 6 SSD-CI listeners who participated in experiment 2. These results show that when these listeners were presented with target speech in their acoustic ear, the copy of the competing speech presented to the CI ear generated a clear advantage, thereby replicating the findings of Bernstein et al. (2016) for this particular set of listeners. This observation was supported by a RM-LRA. Although the main effect of ear condition was not significant (p=0.059), there was a significant main effect of TMR [χ²(4)=409, p<0.0001] and a significant interaction between ear condition and TMR [χ²(4)=55.9, p<0.0001]. Bonferroni-corrected post-hoc tests showed a significant effect (p<0.01) of ear configuration for TMRs of −4 and 0 dB (asterisks), confirming the observation of an improvement in performance in the bilateral condition.

Fig. 5.

(a) Mean data from experiment 2 showing performance for speech identification in the acoustic ear with 2 same-gender competing talkers (monaural condition) and the effect of presenting a copy of the maskers to the CI ear (bilateral condition). (b) Mean data from experiment 1 (target speech presented to the CI ear) for the same 6 listeners who participated in experiment 2. Asterisks in (a) indicate conditions where performance was significantly different between the monaural and bilateral conditions (Bonferroni-corrected criterion p<0.01). Post-hoc comparisons were not carried out in (b) because there was no significant interaction between TMR and ear configuration. Error bars indicate ± 1 standard error of the mean.

The competing-talker data from experiment 1 are replotted in Figure 5B for the subset of 6 SSD-CI listeners who completed experiment 2. These data show that these 6 listeners, on average, experienced interference when the target speech was presented to the CI ear, thereby showing the same result that was observed for all 13 listeners in experiment 1 (Fig. 1). A RM-LRA conducted on the competing-talker data from experiment 1 for just these 6 listeners confirmed that this subgroup of listeners experienced interference in the main experiment. There were significant main effects of ear configuration [χ²(1)=12.3, p<0.0001] and TMR [χ²(6)=3.82×10¹⁴, p<0.0001]. Because there was no significant interaction between the two factors (p=0.12), post-hoc tests were not conducted for individual TMRs.

In summary, the results from experiment 2 showed that on average for the subset of 6 SSD-CI listeners tested, there was a binaural advantage from adding a contralateral copy of the competing speech when the target speech was presented to the acoustic ear, replicating the findings of Bernstein et al. (2016). These same listeners demonstrated contralateral speech interference in experiment 1 when the target speech was presented to the CI ear.

EXPERIMENT 3: IN-SET VERSUS OUT-OF-SET MASKERS

Materials and methods

Rationale

The first two experiments clearly showed opposite effects: adding a copy of the interfering speech to the acoustic ear reduced performance relative to monaural CI-ear performance (experiment 1), whereas the addition of the CI ear improved performance relative to monaural acoustic-ear performance (experiment 2). There was, however, one important methodological difference between the first two experiments that could have contributed to the different results: whereas experiment 1 used out-of-set maskers, experiment 2 used in-set maskers. Bernstein et al. (2016) found that while SSD-CI listeners still received a benefit from adding a copy of the interferers to the CI ear with out-of-set maskers, the benefit was smaller than with in-set maskers. Experiment 3 examined the effect of masker keyword selection by repeating experiment 1 (target speech to the CI ear) using in-set maskers. The same 6 listeners from experiment 2 participated in this experiment (Table 1).

Procedure

The methods followed experiment 1, with the following exceptions. First, the competing talkers used in-set keywords, as in experiment 2. Second, a smaller set of TMRs was tested (0, 8, and 16 dB) to limit testing time. Third, only the competing-talker masker type was tested. Listeners were presented with blocks of 24 trials each. In each block, the ear configuration was held fixed (monaural or bilateral presentation), but the TMR varied randomly from trial to trial. There were a total of 6 blocks (3 blocks for each ear-configuration condition), for a total of 24 trials per condition and 144 trials overall. Blocks were presented in pseudorandom order, with one block completed for both ear configurations before the next block was initiated for either ear configuration. As in the other experiments, feedback was provided after each response. No additional training was provided before this experiment. The experiment took about 30 minutes for each listener to complete.

Analysis

A RM-LRA was conducted on the combined data for the 6 listeners who participated in both experiments 1 and 3. Only the competing-talker condition and the 3 TMRs tested in both experiments (0, 8, and 16 dB) were included in the analysis. The analysis included 3 factors and their interactions: TMR, ear configuration, and masker set (in-set or out-of-set). Post-hoc tests Bonferroni corrected for multiple comparisons (criterion p<0.0167=0.05/3) examined the effect of ear configuration at each TMR tested.

Results

The symbols in Figure 6 show the results of control experiment 3 conducted using in-set maskers. The data from the experiment 1 (out-of-set maskers) for these 6 listeners are replotted as dashed lines in Figure 6 for direct comparison with the in-set masker results. There were significant main effects of ear configuration [χ²(1)=10.4, p=0.0012] and TMR [χ²(2)=26.6, p<0.0001], and a significant interaction between ear configuration and TMR [χ²(2)=12.0, p=0.0025]. There was neither a significant main effect of masker set (p=0.19) nor a significant two-way interaction between masker set and ear configuration (p=0.82). There was a significant two-way interaction between masker set and TMR [χ²(2)=11.0, p=0.0041], and a significant three-way interaction between masker set, TMR, and ear configuration [χ²(2)=31.1, p<0.0001]. Despite these interactions, Bonferroni post-hoc tests showed a similar pattern of interference as a function of TMR for both the out-of-set (experiment 1) and in-set (experiment 3) maskers, with significant interference observed in both experiments at TMRs of 8 and 16 dB (p<0.0167), but not at a TMR of 0 dB. In summary, the results of experiment 3 show that the contralateral speech interference observed in experiment 1 occurred regardless of whether the competing talker spoke keywords that were in-set or out-of-set.

Fig. 6.

Mean results from experiment 3 where the competing talker produced keywords that were part of the possible response set (in-set maskers). Mean data from experiment 1 (out-of-set maskers) for the same 6 listeners are replotted as dotted and dashed lines for comparison. Asterisks indicate TMRs where mean performance was significantly different between the monaural and bilateral conditions (p<0.0167). Error bars indicate ± 1 standard error of the mean.

The fact that experiments 2 and 3 used in-set maskers allowed further inspection of the data to determine the types of errors that listeners produced (this analysis was not possible in experiment 1, where the competing talker used out-of-set keywords, such that there was no way for the listener to respond with the words spoken by the competing talker). For each error, there were two possibilities: listeners responded either with a keyword spoken by a competing talker, or with another random keyword that was not spoken by any talker during that trial. Figure 7 plots the proportion of incorrect responses for which listeners reported the keyword spoken by a competing talker. Only incorrect responses were considered in the analysis, and no inferential statistics were performed because each condition involved a dramatically different number of trials depending on the percentage of correctly identified target words. Figure 7A shows the results for experiment 2, where the target speech was presented to the acoustic ear. Figure 7B shows the results for experiment 3, where the target speech was presented to the CI ear. The horizontal dashed line in each panel represents the expected proportion of competing-keyword responses if the listener was responding randomly. This value was higher in experiment 2 than in experiment 3 because there were two competing talkers instead of just one, and therefore fewer keyword choices available that were not spoken by one of the competing talkers.

Fig. 7.

Analysis of incorrect responses for (a) experiment 2 and (b) experiment 3. Each panel plots the mean proportion of incorrect responses where the selected keyword was spoken by the interfering talker(s). Error bars indicate ± 1 standard error of the mean across listeners. Means and standard errors were calculated by weighting the data for each listener by the proportion of incorrect responses. Dashed lines represent chance performance.

For experiment 2, where bilateral presentation produced contralateral unmasking (i.e., an advantage) rather than interference, Figure 7A shows that the incorrect responses nearly always reflected the competing speech, even in the monaural condition (except at the highest TMRs where the competing speech would have been less intelligible). In contrast, Figure 7B shows that this was not the case in experiment 3. Here, the proportion of competing-keyword responses was somewhat greater for bilateral than for monaural presentation at the two highest TMRs where Figure 6 shows that the interference effect occurred. However, despite this slight increase, listeners only reported the competing keywords on less than half of the incorrect trials.

DISCUSSION

When attending to masked speech in their acoustic ear, SSD-CI listeners experience an improvement from adding a copy of the competing speech to the CI ear (Bernstein et al. 2016), demonstrating that the CI can provide binaural benefits that enable the listener to better organize a complex multiple-talker auditory scene. The goal of this study was to test the hypothesis that SSD-CI listeners attending to masked speech in their CI ear would experience the opposite effect – interference instead of a benefit – when a copy of the masker was presented to the acoustic ear. This prediction was based on results from BI-CI listeners who showed large interaural asymmetries in speech understanding (Goupell et al. 2018). In contrast to relatively symmetric BI-CI listeners who showed a benefit, much like SSD-CI listeners attending to their acoustic ear (Bernstein et al. 2016), the asymmetric BI-CI listeners from Goupell et al. (2018) showed interference, especially when attending to their poorer ear.

The current study had 5 principal findings. First, the results supported the hypothesis of a contralateral speech-interference effect for SSD-CI listeners attending to their CI ear. Experiment 1 showed that the presentation of a copy of the competing speech to the acoustic-hearing ear interfered with the ability to understand speech masked by a competing talker in the CI ear (Fig. 1). Second, there was substantial inter-subject variability in the magnitude of the interference effect, which ranged from 1 to 43 rau (Fig. 2). Third, the magnitude of the interference was found to correlate with listener age: older listeners showed more interference than younger listeners (Fig. 4D). Fourth, the interference effect was only observed for a speech masker; a noise masker did not produce interference for any of the listeners in the study (Figs. 1B, 4B, and 4D). Fifth, the interference effect observed in experiment 1 (Figs. 1A and 6B) was opposite to the effect observed in experiment 2, where the presentation of a copy of the competing speech to the CI ear provided a modest improvement in speech understanding for the acoustic ear (Fig. 5A and Bernstein et al., 2016).

The results of experiment 3 (Fig. 6) showed that the use of in-set maskers (experiment 2) versus out-of-set maskers (experiment 1) did not contribute to the opposite effects in the two experiments. The only other major methodological difference between experiments 1 and 2 was that the experiments involved different numbers of competing talkers (one in experiment 1; two in experiment 2). This was done because two competing talkers would have made the task too difficult in the target CI ear in experiment 1, while one competing talker would have made the task too easy in the target acoustic ear in experiment 2. However, this methodological difference is unlikely to explain the different results observed in experiments 1 and 2 (see Fig. 2) since Bernstein et al. (2016) showed that SSD-CI listeners received a contralateral unmasking advantage with one or two competing talkers when the target speech was presented to the acoustic ear.

What caused the interference effect?

One possible interpretation of the interference effect is that it might be exceedingly difficult for some individuals to ignore a salient speech signal in one ear when trying to focus on much less salient CI-processed speech in the other ear. Completing the speech-identification task required a listener to carry out two processes simultaneously: (1) selectively attending to the target speech in the CI ear while ignoring the competing speech in the contralateral ear, and (2) discerning the speech information in the target ear. The process of discerning speech in the CI ear requires substantial attentional resources because of the distortion associated with electrical stimulation. In the context of the Framework for Understanding Effortful Listening (Pichora-Fuller et al. 2016), this leaves the listener with fewer available resources to ignore the high-fidelity speech in the contralateral ear. In contrast, when SSD-CI listeners were attending to their acoustic ear (experiment 2; Fig. 5A), the task was made difficult not from CI processing, but from the presence of competing speech in the same ear as the target speech at a relatively poor TMR. While this primary task still likely required substantial attentional resources, it is likely that the degree of selective attention required to ignore salient speech in the acoustic ear (experiment 1; Figs. 1A and 6B) would have been greater than the degree required to ignore the distorted CI speech in the contralateral ear (experiment 2; Fig. 5A), yielding the divergent results.

One caveat to this interpretation of the results in terms of a failure of selective attention is that the competing speech presented to the acoustic ear was not, in fact, an additional voice in the mixture, but was instead a copy of the competing speech already present in the CI ear. If listeners perceived the two copies of the competing speech as a single fused voice, then they should not have experienced interference, because they were already tasked with ignoring the competing voice in the monaural case. It remains unknown whether listeners were able to perceptually fuse the two copies of the competing speech, and whether they were able to hear a spatial difference between the target and masker speech, because we did not explicitly ask them these questions. However, Bernstein et al. (2016) showed that SSD-CI listeners obtained an advantage from bilateral presentation of the interfering speech in the case where the target speech was presented to the acoustic ear (replicated here in experiment 2). This suggests that SSD-CI listeners were able to achieve at least partial fusion of the bilateral competing speech that enabled the perceptual separation from the monaural target speech. Bernstein et al. (2016) interpreted the contralateral-unmasking advantage observed when the target speech was presented to the acoustic ear in terms of two possible perceptual bases. First, the presentation of the interfering speech to both ears could have given listeners a spatial cue to hear the target and interfering speech as originating from different spatial locations, thereby allowing the perceptual separation (Freyman et al. 2001). Second, the interfering speech could have been perceived as having a different sound quality (i.e., a combined electric and acoustic timbre) than the target speech (i.e., a purely acoustic timbre), thereby providing a cue for the perceptual separation of the competing voices.

One major difference between experiments 1 and 2 is that the largest interference effects in experiment 1 were observed at the highest TMRs tested (i.e., positive TMRs of 8–20 dB, Fig. 1A), in contrast to experiment 2 where the contralateral-unmasking advantage was observed at much lower TMRs (−4 and 0 dB, Fig. 5A). At high TMRs, the competing speech in the CI ear would have been much lower in level than the target speech, and therefore of poor intelligibility. Thus, it is possible that the introduction of a copy of the interfering speech would have effectively represented a new speech signal to ignore. We did not examine a purely dichotic condition (one voice in each ear) to assess this possibility. A future study would be required to determine whether the interference effect would occur with purely dichotic presentation.

An alternative possible explanation for the results is that the presentation of the competing speech to the target ear simply increased the likelihood that listeners would erroneously respond to the competing speech instead of the target. This possibility was examined by analyzing the kinds of errors that listeners made when they did not correctly identify the target keywords (Fig. 7). In experiment 3, there was a slight increase in the proportion of competing-keyword responses in the bilateral condition at TMRs of 8 and 16 dB. This is not surprising given that listeners now had undistorted access to the competing speech in the masker ear. Importantly, they still only reported the competing-talker keyword on fewer than 50% of the incorrect trials. In other words, when they failed to correctly identify the target keyword, they were slightly less likely to report competing-talker keywords than they were to report keywords that were not spoken by the competing talker. (At a TMR of 0 dB, the proportion of competing-keyword responses decreased with bilateral presentation, perhaps because the availability of the undistorted competing-speech signal in the acoustic ear provided a cue as to which of the two equally loud talkers in the CI ear not to report.) The error pattern in experiment 3 (Fig. 7B) was in sharp contrast to the pattern of error in experiment 2 (Fig. 7A), where listeners reported competing-talker keywords in the vast majority of cases where they did not correctly identify the target speech. This suggests when the target speech was presented to the acoustic ear, errors were due to target-masker confusion, and the addition of a copy of the competing speech to the CI ear provided a cue to perceptually separate the signals. Overall, this analysis suggests that the observed interference effect in experiment 3 was not because listeners were simply erroneously reporting the keywords spoken by the competing talker.

Although it is clear from the results that the interference effect was caused by competing speech and not by noise, it is not known whether the interference effect was dependent on the strict time-alignment and target-masker similarity associated with the CRM sentences used here. In the context of monaural speech understanding, this particular paradigm is known to generate a great deal of target-masker confusion, especially in the case where the target and masker speech are produced by same-gender talkers (Brungart, 2001; Brungart et al., 2001). Further research would be needed to determine whether the observed interference effect relies on target-masker similarity in a similar way, or whether any salient speech masker would produce a similar effect.

Inter-subject variability

The results showed substantial inter-subject variability in the interference effect, with some listeners exhibiting little or no interference and others showing interference of more than 40 rau. The causes of this variability could be attributable to poor monaural speech understanding or to a central processing issue. For example, listeners with poorer peripheral encoding of speech information in the CI ear could be more susceptible to interference effects. Alternatively, listeners with central processing limitations might be more likely to experience a breakdown in the selective attention required to ignore salient speech in the acoustic ear.

The idea that poor monaural speech understanding in the CI ear leads to interference is supported by the divergent results of experiments 1 and 2. The fact that SSD-CI listeners showed contralateral speech interference when attending to the CI ear (experiment 1) but not when attending to the acoustic ear (experiment 2 and Bernstein et al. 2016) suggests that the presence of interference depended on the magnitude and direction of the asymmetry in hearing acuity. In other words, it would have been relatively easy to ignore a relatively low-salience CI-processed speech signal when trying to focus on a task involving undistorted speech in an acoustic-hearing ear. Data from Goupell et al. (2018) for asymmetric or prelingually or perilingually deafened adult BI-CI listeners generally support this idea. In that study, a CI ear with good speech understanding was more likely to be immune to the interference from the opposite ear, but also more likely to cause interference in the opposite ear. However, there was no evidence for the SSD-CI listeners in the current study of a correlation between the magnitude of the interference and speech-understanding scores in the CI ear (Fig. 4). It could be that regardless of their individual CI speech-understanding scores, the salience of speech in the CI ear was so much poorer than the acoustic ear for all of the SSD-CI listeners that individual variability in CI-alone performance had relative little influence on the magnitude of the interference.

The fact that the interference effect was, on average, greatest at the highest TMRs tested (Figs. 1 and 2) is surprising in that the level of the competing speech decreased with increasing TMR. In other words, those conditions where the interference effect was the greatest were also the conditions where the competing speech was quietest. It is important to point out that performance in the bilateral condition tended to flatten out with increasing TMR, especially for some individual listeners (e.g., S1, S2, and S9 in Fig. 2). This suggests that any audible speech in the acoustic ear was enough to generate interference. The reference level of the speech presented to the acoustic ear was 55–65 dB SPL, which means that at a TMR of +20 dB the level of the competing speech was 35–45 dB SPL, which is quiet, but should have been audible to most listeners. At some point, the level of the competing speech would have become so low that the speech was no longer sufficiently intelligible, which might explain why performance appeared to increase again at the highest TMR tested for some listeners (e.g., S4, S11, and S12 in Fig, 2). The observation that even very quiet competing speech can cause contralateral speech interference is consistent with the results of Goupell et al. (2016). For the bilateral-CI listeners in that study who showed contralateral interference effects, the interference effect persisted until very high TMRs of 20–30 dB when the level of the competing speech likely fell below the threshold of the sound processor.

Arguing in favor of a central origin of the interference effect, the only variable that was significantly correlated with the magnitude of the interference was listener age (Fig. 4D). While further study with a larger sample size is needed to confirm this observation, this possible aging effect is consistent with the idea that older adults have difficulty in perceptually suppressing irrelevant competing speech in a speech-understanding task. The literature is mixed with regard to this question. Several studies have found that younger and older listeners experience the same degree of informational masking – i.e., reduced speech understanding in cases of competing speech that cannot be explained in terms of peripheral masking effects (Li et al. 2004; Humes et al. 2006; Helfer & Freyman 2008). Other studies, however, do show evidence that older listeners experience difficulty suppressing irrelevant information in a speech-on-speech masking task. For example, Janse (2012) demonstrated that for older adults (age 65–83 years), performance in a speech-on-speech masking task was correlated to performance in a visual test of the ability to ignore competing information. Pressaco et al. (2016) found an enhanced cortical representation of the envelope associated with competing speech for older listeners. It could be that a reduced ability for older listeners to inhibit the competing speech is exacerbated when SSD-CI listeners are faced with the attentionally demanding task of ignoring salient, acoustic speech when attending to distorted CI speech, leading to large interference effects observed here.

Central interference might be specifically caused by cortical reorganization related to SSD, such as occurs for children during development (Breier et al. 1998; Gordon et al. 2013; 2015), although evidence for similar reorganization for adult-onset SSD is more equivocal, with some electrophysiological and neuroimaging studies identifying asymmetric reorganization in cortical responses favoring the unaffected ear (Firszt et al. 2006; Hanss et al. 2009), and others showing no such evidence (Hine et al. 2008). Alternatively, central interference might reflect a more general selective-attention deficit that simply becomes more problematic with asymmetric auditory input. This view is supported by the fact that there were no observed relationships between the magnitude of interference and variables related to duration of deafness or CI use. Only age (which, as discussed above, has sometimes been linked to deficits in selective attention for nonSSD listeners) was related to the magnitude of the interference. This result raises the possibility that any older listener with a general selective-attention deficit would have difficulty attending to distorted speech in one ear while ignoring salient speech presented to the other ear, and that the effect is not specifically related to deafness and CI use. More research will be needed to determine if the observed effects are specific to CI use.

Implications for everyday listening

Regardless of the cause of the interference effect and inter-subject variability observed here, the results suggest that many listeners will be at risk of interference in exactly those conditions where SSD-CIs have been shown to provide benefits for speech understanding – when the SNR is more favorable at the CI ear than at the acoustic ear. In this kind of configuration, having access to sound via a CI in the deaf ear can provide anywhere from 2–5 dB of head-shadow benefit when the masker is noise (Vermeire & Van de Heyning 2009; Bernstein et al. 2017). The current results suggest that in a case where the masker is a competing talker, the interference effect might substantially reduce any head-shadow benefit that the CI might otherwise provide. For example, consider the case where an SSD-CI listener is seated next to two talkers, one on each side, and is listening to the person on the CI side. In this case, the CI ear will have a better SNR than the acoustic ear. Therefore, the CI should provide a head-shadow benefit, and better speech understanding than if the CI were not available, as long as monaural speech understanding in the CI ear is sufficient to provide this benefit. However, the current results suggest that the competing talker on the acoustic-hearing side might cause interference, potentially diminishing the head-shadow benefit. One practical suggestion for SSD-CI listeners is that in multiple-talker situations, they might be better off if they are able to position themselves such that the target talker is on the side of their better-hearing, acoustic ear. While this recommendation applies in general, the current results suggest that it might be even more important in multiple-talker situations.

While the current results indicate that SSD-CI listeners may be susceptible to interference effects, it is not known how this interference and head-shadow benefit would interact in a real-world listening condition. To clearly determine whether the interference would reduce the head-shadow advantages that a CI can provide, further studies would be required to evaluate how the CI and acoustic ears contribute to speech understanding in a free-field listening condition comparable to the configuration employed in the current study – that is, when the CI ear has a better SNR than the acoustic ear. One important difference between the current study and this hypothetical free-field listening scenario is that in the free field, the target speech will not be completely absent from the acoustic-hearing ear, which might mitigate some of the interference effect. While it remains unknown whether interference would reduce head-shadow benefit for SSD-CI listeners, Goupell et al. (2018) found that BI-CI listeners who experienced a similar interference effect as the SSD-CI listeners in the current study also showed less spatial release from masking for a speech masker than would be expected with a noise masker, suggesting that the head-shadow benefit was reduced by the interference.

Finally, it is important to point out that the interference effect observed here in no way suggests that CIs might cause interference for individuals with SSD. The interference was strictly limited to the case where listeners were attending to the CI ear and additional speech was introduced to the acoustic ear. There was no evidence that introducing speech information to the CI ear caused interference relative to a monaural condition involving the acoustic ear only. In fact, experiment 2 (Fig. 5A) showed that adding the CI produced a benefit, and not a deficit relative to monaural performance in the acoustic ear, consistent with Bernstein et al. (2016). Thus, the results of this study should not be taken as a contraindication to cochlear implantation in the case of SSD.

Summary and Conclusions

Previous results showed that for SSD-CI listeners, the CI ear can facilitate the perceptual separation of concurrent voices in the acoustic ear. In the current study, it was found that presenting competing speech to the acoustic ear interfered with speech understanding in the CI ear. These results corroborate similar findings in a group of highly asymmetric BI-CI listeners (Goupell et al. 2018) and suggest that CI listeners with asymmetric hearing are particular susceptible to contralateral speech interference. There was a great deal of inter-subject variability in the magnitude of the interference effect, ranging from 1–43 rau. There was a significant correlation between the magnitude of the interference and listener age. The interference effect is consistent with the idea that older SSD-CI listeners experience a deficit in selective attention that reduces their ability to ignore salient acoustic speech when trying to attend to less salient CI-processed speech. The practical implication of this finding is that in a multiple-talker situation, contralateral speech interference could potentially reduce the benefit that SSD-CI listeners might otherwise receive in spatial configurations where they are attending to a target talker in their CI ear. Further research is required to determine whether the observed interference would occur under real-world free-field test conditions where the target speech is not completely isolated to one ear.