Eighteen adults with HL (9 males; 20.08 $\pm$ 1.18 years) and twenty adults with NH (10 males; 19.90 $\pm$ 0.94 years) participated in this study. All the NH participants had normal pure-tone hearing thresholds at all frequencies ($\le$20 dB HL) (Table 1). In the HL group, five participants had severe (65–79 dB) HL, and 13 participants had profound ($\ge$80 dB) HL. The Levene’s test for equality of variances demonstrated that the variances of age between the two groups were homogeneous [F(1, 36) = 2.930, p = 0.096]. Under the assumption of equal variances, the results of the independent-samples t-test were as follows: t(36) = 0.134, p = 0.894, and Cohen’s d = 0.044. There was no significant difference in age between the two groups. All the HL participants reported no neurological disease and were asked to wear their HA for all following tests. The demographic information and more details on the HL participants are shown in Tables 2,3. This study was approved by the Ethics Committee of Binzhou Medical University (protocol code BMU-IRB-2021-171, approved on 5 March 2021).

All participants were tested for word recognition score (WRS) using the Mandarin speech test materials (MSTMs) [18]. The WRS was performed in monosyllables with a fixed sound intensity of 75 dB hearing level.

A three-factor ANOVA was performed. Group (NH and HL) was a between-participants variable, environment (quiet and +10 dB SNR) and syllables (/ba/, /ga/, and /bu/ for N1 and P2, and /ga/-/ba/ and /bu/-/ba/ for MMN) were inter-subject variables. The dependent variables were the peak latencies of N1, P2, and MMN, as well as the peak-to-peak amplitude of N1-P2 and the mean amplitude of MMN. In the case of violation of the sphericity assumption, a Greenhouse-Geisser correction was applied. Holm-Bonferroni correction was applied to correct for multiple comparisons. The effect size was calculated using the partial eta square. p-values 0.05 were considered statistically significant. SPSS v.22 (IBMCorp., Armonk, NY, USA) was used for all statistical analyses.

Identifiable N1-P2 indicated that speech stimuli were processed in the auditory cortex. Prolonged latencies of AERP often represent complex information processing, damage to nerves, or developmental immaturity. Beynon et al. [23] (2002) used /i/-/a/ as stimuli and found that the N1 latency was longer in CI users than in NH controls. Similar to the result of Beynon et al. [23] (2002), we also found longer N1 latencies in long-term bilateral HA users than in the NH group. As for the amplitude, lower N1-P2 amplitudes in adults were often associated with auditory deprivation and decreased cortical perception. As shown in the study of Groenen et al. [24] (2001), speech (/bu/-/da/, /ba/-/pa/, and /i/-/a/) evoked smaller N1-P2 amplitudes in CI users compared to the hearing controls. Our study obtained similar results that the N1-P2 amplitude in bilateral HA users was significantly lower than the NH group. For the results above, we suspected an association with the fact that people with HL, regardless of the auditory compensation modalities, had fewer specialized functional synapses and lower functional inter-cortical coupling to encode auditory information owing to the lack of adequate auditory inputs during brain development [25]. As shown in Table 2, all of the 18 participants in the HL group in our study had prelingual HL. The immaturity of cortical development caused inefficient early auditory encoding of speech and the missing encoding of speech feature details in the HL group [26].

Consistent with expectations, the two groups showed inconsistent AERP characteristics for different syllables in the early speech perception stage. Although the N1 latency showed no difference either in quiet or noisy conditions, P2 latency evoked by /ba/ was shorter than /bu/ and /ga/. Firstly, this might be related to the fact that N1 and P2 originated from different anatomic sources, and reflected different functional properties of the cortical auditory processing mechanisms. N1, a front-centrally component generated bilaterally mainly in the auditory cortices, corresponds to the first stage of detection and recognition of acoustic features of speech [27]. P2 appears more susceptible to being modulated by cognitive factors and reflects further stimulus evaluation and classification [28]. For our findings, we assumed that the acoustic properties of speech were crudely encoded in the primary auditory cortex during the early perception stage and then were further elaborated during the late perception stage to emphasize the differences in the speech. On the other hand, the reduction of neural responses in the brain when repeatedly exposed to the same stimulus, referred to as adaptation, is usually presented in ERP study as decreased amplitudes and earlier latencies of N1-P2 [29]. In our study both deviant stimuli /ga/ and /bu/ evoked higher N1-P2 than the standard stimulus /ba/, which might be due to the adaptation effect. However, it was only in the P2 latency that we found /ba/ to be shorter than those evoked by the other two stimuli, while no such effect was found in the N1 latency, which might be explained by the fact that P2 seemed to be more sensitive to the adaptation effects than N1 [30]. A most recent study found that children with higher cognitive abilities showed adaptations in P2 amplitude to the syllable /ta/ [28]. Thus, in addition to being related to the progressive processing of speech, the earliest P2 latency evoked by /ba/ might also be attributed to the adaptations shown by the cortex to frequent standard stimuli.

Although HL and different syllables resulted in significant differences in the latencies of N1 and P2, our study did not show much longer latencies of N1 or P2 in the noisy condition. Inconsistent with our results, Small et al. [31] (2018) used /t/ and /m/ as stimuli and found that N1-P2 latencies in NH adults were longer under the same 10 dB and even higher SNR conditions (15 and 20 dB SNR). It is worth noting that our study used CV syllables in babble noise as stimuli, while Small et al. [31] (2018) used separate consonants in white noise as stimuli, which were weaker in energy and more susceptible to poor listening environments. Thus, differences in stimulus signal and noise type might account for the discrepancy between our findings and those of Small et al. [31] (2018). As for the HL group, Korczak et al. [11] (2005) suggested that maintaining individual parameters of HA for HL patients in speech discrimination experiments might provide more meaningful and typical responses to aided speech signals. Therefore, participants in the HL group perhaps benefited from being given a speech signal optimized by the noise reduction procedure, resulting in the N1-P2 latency not suffering from the interference of noise.

Noticeably, the effect of noise attenuating speech encoding in the auditory cortex was consistent across hearing levels. We found that both the NH and the HL groups showed lower N1-P2 amplitude in the noisy condition than in the quiet condition, and it was only in the quiet condition that higher N1-P2 amplitude was found in the NH group than in the HL group, while the differences in speech central encoding between the two groups seemed to be obscured in the noisy condition. These suggested that even individuals with NH were not immune to noise interference and showed a diminished central response to speech. Furthermore, as an objective marker of speech processing in the primary auditory cortex, the integral change of N1-P2 amplitude seems more stable than that of a single component [32]. As shown in Fig. 2, we found that the N1-P2 amplitudes evoked by all three syllables consistently attenuated in the noisy condition, which might be attributed to noise-induced cognitive resource streaming [33]. That is, the auditory cortex is unable to reconcile the simultaneous encoding of noise and target speech signals, consequently impairing the processing of target auditory information. In conclusion, noise is an essential factor affecting the sensitivity of the primary auditory cortex to speech perception.

The MMN is a valuable tool in assessing the integrity of the speech discrimination process in the auditory cortex of people with HL. Long-term auditory deprivation made it difficult for people with HL to perceive small changes in sound signals and required a longer central response time to discriminate sound details [34]. Similar to the study by Boo et al. [34] (2022), we found that the MMN latencies were longer in the HL group than in the NH group, indicating that individuals with prelingual HL, even if they received bilateral HA under the chronological age of 6 years, still have difficulty in recognizing speech sounds after long-term usage and that the effects of prelingual HL can persist throughout adulthood.

As for MMN amplitude, Turgeon et al. [35] (2014) found that the MMN evoked by syllable /ba/-/ga/ showed much lower amplitude in CI users with poorer speech recognition compared to NH controls. In our study, as shown in Figs. 3,4, the HL group showed lower MMN amplitude than the NH group, and the MMN evoked by the vowel contrast in the HL group showed lower prefrontal negative energy compared to the NH group. In addition, it was only in the NH group that we observed that the vowel contrast evoked higher MMN than that of the consonant contrast, whereas in the HL group the vowel contrast and the consonant contrast evoked approximate MMN amplitudes. These results suggested that long-term bilateral HA users with prelingual severe to profound HL, in a way similar to CI users, varied from those with NH in automatic speech discrimination at the pre-cognitive stage, even though the hearing level has been compensated or constructed adequately in the HL group. This deficit in central encoding of speech differences might further affect the recognition and intelligibility of speech in the HL people.

Discerning differences in phonemes is a prerequisite for accurate speech recognition. A previous study used /bi/-/bu/, /ba/-/da/, and /da/-/ta/ as stimulus signals and found that MMNs induced by vowel contrasts all had greater amplitudes as well as shorter latencies than consonant contrasts [36]. Similarly, our study also showed longer MMN latencies evoked by the consonant contrasts than vowel contrasts in both groups and listening conditions. Moreover, MMN amplitudes induced by the vowel contrast were higher, and the negative prefrontal component of the scalp topography was more pronounced than the consonant contrast. The findings above suggested that the auditory cortex automatically evaluated and classified consonants and vowels at the early pre-attentive phase. When pronouncing consonants, the airflow was obstructed to varying degrees in various organs of phonation, whereas when pronouncing vowels, the airflow was not obstructed by any of the organs. Differences in pronunciation patterns made consonants show rapidly shifting resonance peaks and faint loudness, while vowels showed a relatively steady-state spectrum and were much louder [37]. Furthermore, as shown in Fig. 1B and Table 4, the F0, F1, and F2, as well as the envelope, were more similar between /ba/ and /ga/ than between /ba/ and /bu/. It has been reported that the variability influenced the MMN in stimulus characteristics. The greater the variability between deviant and standard stimuli, the earlier and greater the MMN appeared [38]. Thus, it was easy to explain why vowel contrasts, compared to subtle consonant contrasts, were more easily processed by the cortex in our study. Moreover, probably owing to more challenging discrimination of consonant contrasts, it was only the vowel contrast that evoked higher MMN in the NH group than in the HL group in our study.

It is noteworthy that, when processing speech in noisy environments, we observed a significant decrease in the amplitude but no significant delay in the latency of the N1 and P2 components during the original speech perception stage (Fig. 2), whereas during the discrimination stage in which speech subtle differences were processed, a significant delay in the latency of the MMN occurred in both groups but only a slight and insignificant decrease in the amplitude (Figs. 3,4). These results might indicate that noise affects processes of speech and speech differences differently, with the latency of the endogenous component MMN and the amplitude of the exogenous component N1-P2 being more sensitive to noise [39]. Therefore, N1-P2 and MMN should be combined as an objective electrophysiological index for evaluating speech auditory performance in noisy environments.

This study examined the cortical electrophysiological characteristics of noise on speech recognition by bilateral HA users with prelingual severe to profound HL. In real life, SNR are varied. In the future, we plan to increase the gradient of the SNR to explore central processing at different noise levels further.

Secondly, we selected only three CV syllables for the stimuli: /ba/, /ga/, and /bu/. To explore the response characteristics of the auditory cortex to syllables composed of different unitary and consonant sounds in Mandarin Chinese, we need to add more syllable types. Re-choose a suitable experimental paradigm, such as a multi-feature paradigm, might be necessary.

Finally, although the N1-P2 and MMN have been proposed as electrophysiological metrics for assessing speech perception and behavioral outcomes, speech recognition can provide a more direct, simple, and valid reflection of participants’ speech discrimination abilities. We consider the addition of an active speech recognition paradigm in future studies that have been conducted with more ERP components as well as intuitive speech perception results.