While cochlear implants (CIs) provide sufficient spectro-temporal resolution for speech recognition in quiet by deaf individuals, masked speech recognition is often difficult for CI users. Steady noise is thought to largely produce “energetic” masking; the spectro-temporal overlap between the target and masker occurs at the periphery (e.g., Brungart, 2001; Kidd et al., 2002). Competing speech is thought to produce some combination of energetic masking, “envelope” masking (target and masker envelope interference even when there is no spectral overlap; e.g., Stone and Canavan, 2016), and “informational” masking (e.g., lexical interference, target/masker similarities, etc.; Brungart, 2001; Kidd et al., 2002, 2016). Different from normal-hearing (NH) listeners, who have greater difficulty with competing noise than with competing speech, CI listeners have greater difficulty with competing speech than with competing noise (e.g., Stickney et al., 2004; Cullington and Zeng, 2008; Tao et al., 2018). The coarse spectro-temporal resolution is thought to limit CI users' segregation of target and masker speech (e.g., Friesen et al., 2001; Shannon et al., 2004; Fu and Nogaki, 2005; Luo and Fu, 2009).

Cortical measures have been used to characterize NH and CI listeners' auditory processing. Auditory event-related potentials (ERPs) reflect the brain's response to changes in an ongoing stimulus (e.g., deviant stimuli in the context of frequent stimuli in an oddball paradigm). Exogenous, pre-attentive responses (e.g., P1, N1, P2, N2 peaks) typically occur within the first 250 ms and do not reflect cognitive processing (e.g., Martin et al., 2008; Lightfoot, 2016). The latency of the endogenous P3 (or P300) response is typically between 250 and 400 ms, and is thought to reflect attention and/or arousal (e.g., Polich and Kok, 1995; Kok, 2001). Recording of P300 responses requires some sort of behavioral response to the deviant stimulus (e.g., counting the number of deviant stimuli during a test run, indicating when a deviant stimulus was heard, etc.). P300 latency has been shown to be related to the speed of information processing (e.g., Ritter et al., 1972; Kutas et al., 1977; Parasuraman and Beatty, 1980; Donchin and Coles, 1988). P300 amplitude has been shown to decrease with increasing task difficulty (e.g., Parasuraman and Beatty, 1980). Uncertainty in discrimination of sounds may be reflected in reduced P300 amplitude (e.g., Sutton et al., 1965; Hillyard et al., 1971; Squires et al., 1973; Picton, 2011).

There is great variability in CI outcomes that is largely unexplained but may be related to individual central auditory processing capacities (e.g., Dunn et al., 2005). In CI users, ERPs may be used to observe detection (exogenous components) and discrimination (endogenous components) of stimulus contrasts. Assuming there are no cognitive deficits, device-related factors (e.g., the number of implanted electrodes, frequency allocation) and patient-related factors (e.g., the electrode-neural interface, patterns of neural survival, etc.) may affect ERP responses. As such, it is important to select stimuli that are sufficiently contrastive when measuring ERPs. Some studies have used pure-tone contrasts (e.g., Groenen et al., 2001; Beynon et al., 2002; Sasaki et al., 2009; Obuchi et al., 2012; Calderaro et al., 2020; Van Yper et al., 2020; Wedekind et al., 2021) while others have used phonemic contrasts (e.g., Groenen et al., 2001; Beynon et al., 2002, 2005; Beynon and Snik, 2004; Henkin et al., 2009; Micco et al., 1995). ERPs have also been used to observe the evolution of auditory processing after cochlear implantation in longitudinal studies (e.g., Kubo et al., 2001).

P300 is closely related to cognitive processes involving auditory discrimination, selective attention, and working memory (e.g., Polich, 2007). Segregation of competing speech has been shown to involve cognitive processes (e.g., Francis, 2010). Some CI studies have compared P300 responses to standard clinical measures such as word recognition in quiet (e.g., Kileny et al., 1997; Groenen et al., 2001; Grasel et al., 2018; Abrahamse et al., 2021; Amaral et al., 2021). Others have compared P300 responses to phoneme recognition in quiet (e.g., Groenen et al., 2001; Beynon et al., 2002) or to speech recognition in steady noise (e.g., Iwaki et al., 2004). Kileny et al. (1997) found a significant correlation between P300 amplitude and sentence recognition in pediatric CI users. Groenen et al. (2001) found a significant correlation between P300 amplitude and word/phoneme recognition in quiet in adult CI users.

Bimodal listening [CI in one ear, hearing aid (HA) in the other ear] provides important low-frequency temporal fine-structure cues that benefit pitch-mediated perception (e.g., music, talker identity, prosody) and segregation of target speech and maskers (e.g., Gifford et al., 2007; Cullington and Zeng, 2008; Dorman et al., 2008; Yoon et al., 2012; Crew et al., 2015; Liu et al., 2019). Previous studies have shown more robust P300 responses with bimodal than with CI-only listening. Iwaki et al. (2004) found that sentence recognition in noise was significantly better and P300 latency was significantly shorter with bimodal than with CI-only listening. Sasaki et al. (2009) also reported shorter P300 latency and better word recognition in quiet with bimodal than with CI-only listening. However, the relationship between P300 responses and segregation of competing speech with bimodal listening remains unclear.

In this study, P300 responses to pure-tone stimuli were recorded in NH listeners and bimodal CI users; speech recognition was measured in the presence of steady noise or competing speech. Given that the present participants used bimodal listening in daily life, only bimodal listening was tested. Also, previous studies have shown more robust P300 responses with bimodal than with CI-only listening (e.g., Iwaki et al., 2004; Sasaki et al., 2009). Consistent with previous studies (e.g., Beynon et al., 2005; Obuchi et al., 2012; Grasel et al., 2018), we expected greater P300 amplitudes and shorter P300 latencies in NH than in CI listeners. Given the great variability in speech performance among CI users (e.g., Stickney et al., 2004; Cullington and Zeng, 2008) and given that P300 is sensitive to auditory task difficulty (Parasuraman and Beatty, 1980; Polich, 1987; Causse et al., 2016), we expected that P300 responses would be related to masked speech recognition, especially for the more difficult segregation of competing speech by CI users.

Figure 1 shows SRTs with SSN or with a competing male or female talker for the NH and CI listeners. SRTs were much lower (better) for NH than for CI listeners. For NH listeners, mean SRTs progressively improved from SSN (−11.3 ± 1.1 dB) to the male masker (−17.0 ± 9.0 dB) and then to the female masker (−24.9 ± 7.3 dB). For CI listeners, mean SRTs were poorer with the competing male (5.5 ± 3.1 dB) or female talker (3.2 ± 3.3 dB) than with SSN (1.1 ± 5.9 dB). A mixed-design analysis of variance (ANOVA) was performed on the SRT data, with masker (SSN, male, female) as the within-subject factor and group (NH, CI) as the between-subject factor. Results showed significant effects of group [F_(1,40) = 125.1, p < 0.001] and masker [F_(2,40) = 10.7, p < 0.001]; there was a significant interaction [F_(2,40) = 16.8, p < 0.001]. Post-hoc Bonferroni pairwise comparisons showed that for the NH group, SRTs were significantly higher (poorer) with SSN than with the male (p = 0.016) or female masker (p < 0.001), and significantly higher with the male than with the female masker (p < 0.001). There were no significant differences among the maskers for the CI group. SRTs were significantly lower (better) for the NH than for the CI group for all maskers (p < 0.001 for all comparisons).

Figure 2 shows waveforms with the peak P300 response averaged across the three test runs for individual NH and CI listeners. Note that the intra-class correlation coefficient was 0.99 and 0.97 for P300 amplitude and latency, respectively, suggesting good test-retest reliability across the three runs. Because RM ANOVAs showed no significant effect of test run for NH or CI participants (p > 0.05 for all analyses), data were averaged across runs. Mean P300 amplitude was higher for the NH group (8.9 ± 3.5 μV) than for the CI group (3.2 ± 2.2 μV); mean P300 latency was shorter for the NH group (305 ± 23 ms) than for the CI group (338 ± 28 ms). T-tests showed that P300 amplitude was significantly higher for the NH than for the CI group [t(20) = 4.6, p < 0.001], and that P300 latency was significantly shorter for the NH than for the CI group [t(20) = −3.1, p = 0.006].

Figure 3 shows SRTs with SSN or with a competing male or female talker for the NH and CI groups as a function of P300 amplitude and latency; each data point shows the mean across three test runs. When all NH and CI data were combined, and after Bonferroni correction for multiple comparisons (adjusted p = 0.016), Pearson correlation analysis showed significant relationships between P300 amplitude and SRTs with SSN (r = −0.65, p = 0.001), and with the male (r = −0.62, p = 0.002) and female maskers (r = −0.60, p = 0.003). A significant relationship was observed between P300 latency and SRTs with SSN (r = 0.60, p = 0.008), but not for SRTs with the male or female masker. For the CI group, Pearson correlation analysis showed a significant relationship only between P300 amplitude and SRTs with the male masker (r = −0.78, p = 0.005); the correlation remained significant after controlling for age at testing, duration of deafness, and CI experience (r = −0.81, p = 0.016). No significant correlations were observed between P300 amplitude and SRTs with SSN or with the female masker, or between P300 latency and SRTs with any of the maskers. For the NH group, no significant relationships were observed between P300 amplitude or latency and SRTs with any of the maskers. For the CI group, a significant correlation was observed between P300 amplitude and unaided PTA thresholds (across all frequencies; r = −0.87, p < 0.001); there were no significant correlations between P300 amplitude and aided PTA thresholds. Significant correlations were observed between P300 latency and unaided PTA thresholds (r = 0.67, p = 0.025) and aided PTA thresholds (r = 0.68, p = 0.021). Note that statistical power was >0.80 for all of the above correlations, except for P300 latency vs. unaided PTA thresholds (power = 0.63) or aided PTA thresholds (power = 0.65).

Consistent with previous studies (e.g., Kubo et al., 2001; Beynon et al., 2005; Obuchi et al., 2012; Soshi et al., 2014; Grasel et al., 2018; Han et al., 2020), P300 amplitudes were significantly larger and latencies were significantly shorter for the NH group than for the CI group. For the present bimodal CI listeners, mean P300 amplitude and/or latency values were comparable to those observed in previous studies with CI listeners (e.g., Iwaki et al., 2004; Sasaki et al., 2009; Grasel et al., 2018; Abrahamse et al., 2021; Calderaro et al., 2020; Van Yper et al., 2020). P300 responses were elicited in all CI participants, consistent with Obuchi et al. (2012).

Mean SRTs for all maskers were lower (better) for the NH group than for the CI group, and values were comparable to those in previous studies using similar methods and stimuli (Tao et al., 2018; Zhang et al., 2020). Different from previous CI studies that showed lower SRTs in SSN than in competing speech (e.g., Cullington and Zeng, 2008; Croghan and Smith, 2018; Tao et al., 2018; Liu et al., 2019), there was no significant difference in SRTs between the SSN and competing speech maskers within the CI group. Note that CI listeners were tested while wearing contralateral hearing aids, which likely aided in segregation of competing speech, thereby reducing the deficit relative to SSN.

Across all NH and CI listeners, significant correlations were observed between P300 amplitude and SRTs with the SSN, male, and female maskers; a significant correlation was also observed between P300 latency and SRTs with SSN. These correlations were largely driven by across-group differences in speech performance and P300 responses. In general, higher P300 amplitude and shorter P300 latency were associated with better masked speech recognition.

In the NH group, there were no significant correlations between P300 responses and SRTs with any of the maskers. In the CI group, a significant correlation was observed only between P300 amplitude and SRTs with the male masker, the most challenging listening condition with the greatest informational masking. The correlation between P300 amplitude and SRTs with the male masker suggests some common relation to informational masking, a central auditory process. With the female masker, informational masking was reduced, and SSN produced largely energetic masking. Given the correlations between unaided PTA thresholds and P300 amplitude and latency and between aided PTA thresholds and P300 latency, differences in P300 response across CI listeners may have represented differences in segregation of the competing male talkers with residual acoustic hearing that provided low-frequency pitch cues.

Different from Soshi et al. (2014), we observed a significant correlation between P300 amplitude and SRTs with the male masker, but not between P300 amplitude and SRTs with SSN. Differences in cortical measure stimuli (1 vs. 2 kHz contrasts; consonant contrast), speech tests, methods, and CI patients (bimodal vs. CI-only listening) may have contributed to differences in results across studies. The 1 and 2 kHz stimuli used for ERP recording were presented at 20–30 dB above the aided thresholds, meaning that the aided acoustic hearing should have contributed to the response.

In the present study, ERPs and speech performance were measured only with bimodal listening. Some studies have shown greater P300 response and speech performance with bimodal than with CI-only listening (e.g., Iwaki et al., 2004; Sasaki et al., 2009). Interestingly, Wedekind et al. (2021) found no significant difference in P300 response between the NH ear and the CI ear in unilaterally deaf CI recipients; speech recognition in noise was better with the CI on than off. While it was not directly measured in Wedekind et al. (2021), speech performance would be expected to be much poorer with the CI ear alone than with the NH ear alone (e.g., Galvin et al., 2019). It is unclear why the P300 response would be similar across ears when speech performance would be different. As shown in Figure 3, significant relationships were observed between P300 amplitude and masked SRTs, presumably due to the underlying spectro-temporal resolution that was much better for NH than for CI listeners. However, some caution is warranted regarding the correlational analyses, given the limited number of participants and test runs. ERPs and speech performance were not measured with the acoustic-hearing ear alone or the CI ear alone in this study. It is possible that strong P300 responses may have been elicited within the acoustic-hearing ear alone, despite the expectedly poor speech performance. In future studies, it would be worthwhile to collect ERPs and speech performance with each ear alone and both ears together to better understand how the peripheral representations might affect the relationship between ERPs and speech performance.

The present results show some evidence that ERPs may be a useful objective measure to predict complex perception such as segregation of competing speech. However, eliciting P300 also requires a behavioral component in the oddball presentation, and the magnitude of the response may depend on the strength of the stimulus contrast. Obuchi et al. (2012) showed increasing P300 amplitude in CI listeners as the stimulus frequency contrast was increased from 1.5 to 4 kHz. Depending on the acoustic-to-electric frequency allocation and the electrode-neural interface (electrode position relative to healthy neurons), small contrasts (e.g., 1 vs. 1.5 kHz) may be perceived differently among CI listeners. The 1 vs. 2 kHz contrast in this study appeared to be sufficiently large to be discriminated by the present MED-EL CI users, most likely resulting in stimulation of electrodes 6 and 8, given the default frequency allocation. Note that there may have been some contribution from residual acoustic hearing for discrimination of the stimuli contrast.

Auditory ERPs and speech recognition in steady noise or competing speech were measured in NH and bimodal CI listeners. P300 amplitude was larger and latency was shorter in the NH group than in the CI group. Similarly, speech performance was better for the NH group than for the CI group. Significant correlations were observed across all participants between P300 amplitude and SRTs with steady noise and the male and female maskers. Within the CI group, P300 amplitude was significantly correlated with SRTs with the male masker, suggesting some relation between cortical response and informational masking.

The raw data from the study are included in the supplementary material; further inquiries can be directed to the corresponding author.

The studies involving human participants were reviewed and approved by the Ethical Committee from the First Affiliated Hospital of Soochow University (Approval number 2021122). All participants provided written informed consent before participating in the study; approval was obtained for pediatric CI and NH participants from their parents or adult next of kin.

D-DT: study design, data analysis, and writing of manuscript. Y-MZ and DZ: data analysis and writing of manuscript. JG: data analysis, data visualization, and writing of manuscript. HL, WZ, and MX: study supervision. J-SL: study design. All authors contributed to the article and approved the submitted version.

This work was supported by the National Natural Science Foundation of China (81970877), the Jiangsu Provincial Natural Science Foundation—Outstanding Youth Foundation (BK20200054), the Jiangsu Provincial Key Research and Development Program Special Funds (BE2019670), and the Suzhou Science and Technology Project (SYS2019049).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.