Speech comprehension in noisy environments is challenging even for individuals with normal hearing and cognitive abilities (Cherry and Wiley, 1967), but it is even more difficult for those with listening difficulties (LiD; Petley et al., 2021; Parthasarathy et al., 2020). Some individuals with LiD have normal audiograms but find it extremely hard to understand speech in noisy situations. They also struggle with following complex verbal instructions and understanding rapid or degraded speech (Jerger and Musiek, 2000). Some of the children and adults who face these issues are diagnosed with auditory processing disorder (American Speech-Language-Hearing Association, 1996).

Speech comprehension relies on the integration of acoustic feature processing in the auditory pathway (bottom-up processing; Hardcastle et al., 2012) and linguistic feature processing in the speech and language pathways (top-down processing; Davis and Johnsrude, 2007). The top-down processing can be viewed as an inferential process that uses context information at different levels, such as syntax, semantics, and phonemes, to resolve ambiguities in the speech signal (Obleser, 2014). This process also involves memory and attention (Pichora-Fuller et al., 2016; Rönnberg et al., 2019), which play crucial roles. These two pathways are not mutually exclusive: bottom-up processing can influence top-down processing and vice versa (Lesicko and Llano, 2017; Morlet et al., 2019). It is plausible that many individuals with LiD who have normal audiograms have issues specifically affecting the top-down processing (Moore et al., 2010). Evaluating individual differences in neural activity during the top-down processing of speech comprehension, specifically inference processes, may be crucial for understanding the problems faced by individuals with LiD who have normal audiograms and for selecting appropriate treatment strategies.

For evaluating speech comprehension in individuals with LiD, it is common to use not only hearing tests but also behavioral auditory processing tests, such as speech-in-noise perception assessments and dichotic listening tests. However, even on following clinical guidelines, these tests can yield highly variable results (Jutras et al., 2007; Wilson and Arnott, 2013). Additionally, it is impossible to completely rule out the influence of peripheral hearing impairments, such as synaptic dysfunction between inner hair cells and type I auditory nerve fibers (Kujawa and Liberman, 2019), subclinical endolymphatic hydrops (Yoshida et al., 2023), and reduced extended high-frequency hearing (Motlagh Zadeh et al., 2019; Hunter et al., 2020), which are undetectable by audiograms and influence bottom-up processing. Neurophysiological evaluations using auditory brainstem responses (Ankmnal-Veeranna et al., 2019; Hunter et al., 2023) and event-related potentials, such as P300 (Morlet et al., 2019; Petley et al., 2024), have been reported. Although these measures provide valuable insights, they primarily examine neural activity following bottom-up processing. Therefore, similar to the behavioral auditory processing tests, these tests are limited by their inability to exclude the effects of peripheral hearing impairments that are undetectable with audiograms (Dillon and Cameron, 2021).

We propose that stimulus-preceding negativity (SPN; Brunia, 1988; Brunia and van Boxtel, 2004), an electrophysiological marker of prediction, could serve as a useful biomarker for top-down processes involved in speech comprehension. The neural dynamics underlying prediction have been studied using electroencephalogram (EEG) in an S1-S2 paradigm, where a cue stimulus (S1) is followed by a target stimulus (S2). During tasks requiring participants to respond to S2 with a motor action, brain potentials gradually shift in a negative direction between S1 and S2; this shift is known as contingent negative variation (CNV; Mento, 2013; Ruiz-Martínez et al., 2022). This phenomenon has been observed across various sensory modalities, including auditory function. Even when no motor response to S2 is required, a similar negative shift in brain potential has been reported based on mental anticipation of S2, termed SPN. SPN is characterized by a negative shift in brain potential observed prior to the arrival of sensory input, indicating that the brain is preparing for the incoming stimulus. It emerges between the cue and feedback without necessitating a direct motor response. Therefore, the SPN can be considered a predictive neural activity that is not influenced by bottom-up sensory processing or the preparation of motor responses (Brunia and Damen, 1988). Changes in the amplitude of SPN are most pronounced when individuals anticipate significant events, such as feedback on performance or decision-making outcomes (Zhang et al., 2017; Li et al., 2018; Walentowska et al., 2018).

The importance of prediction in perceptual processes has been explained within the framework of predictive coding (Friston et al., 2006; Friston, 2010). Additionally, the relationship between predictive coding theory and CNV modality, including SPN, has been further demonstrated through approaches using mathematical inference models (Bennett et al., 2015; Gómez et al., 2019). This theory posits that the brain seeks to minimize discrepancies between predictions generated by internal models and actual sensory input. This minimization is accomplished through a hierarchical system in which higher brain areas formulate predictions juxtaposed with sensory information from lower brain areas. When there is a variance between the brain’s predictions and sensory perceptions (termed “prediction error”), the brain updates its model to reduce this error. Through these prediction errors, the brain continuously learns to better anticipate future events and stimuli (Friston, 2009). This theory suggests that perception, including speech comprehension, is not merely a passive reception of sensory inputs but rather an active inferential process (Heilbron and Chait, 2018; Pezzulo et al., 2024).

Recent studies on predictive coding suggest that paying attention to stimuli enhances the gain of neurons that relay prediction errors to more advanced hierarchical inference structures (Friston, 2009; Feldman and Friston, 2010). Based on this hypothesis, auditory inputs in noisy environments could produce larger prediction errors than those in quiet settings. This is because auditory signals masked by background noise demand more attentional resources to accurately predict the original auditory information (Ono et al., 2021). When background noise reduces the clarity of the speech signal, it is expected that the level of prediction errors will also change, leading to variations in the SPN amplitude.

This study aims to clarify the relationship between SPN and performance during speech comprehension tasks in noisy environments in young adults with normal hearing. We considered the following hypotheses. (1) In noisy environments, the attentional resources required for speech comprehension increase compared to in quiet environments, which would lead to changes in SPN amplitude. (2) The change in the SPN amplitude between noisy and quiet environments would be positively correlated with performance in behavioral auditory processing tests. These hypotheses were tested using a time-estimation task with auditory feedback.

Figure 3 shows the grand average SPN at each electrode. The SPN amplitudes at the bilateral frontal and central electrodes showed a positive peak approximately 600–800 ms after the button press. Following this peak, the amplitudes gradually shifted in a negative direction until the auditory feedback was delivered. By contrast, the SPN amplitudes at the bilateral parietal electrodes showed a negative peak approximately 400–600 ms after the button press, followed by a slight shift in the positive direction until the auditory feedback was delivered. The analysis of the SPN amplitude is summarized in Table 1 and revealed significant main effects of the condition (F = 4.06 p = 0.018, f = 0.39). The results of the Bonferroni corrected post hoc comparisons are shown in Figure 4. SPN amplitude in the speech in noise condition (mean −0.32 μV, SD = 0.63) was significantly higher than in the pure tone condition (mean −0.16 μV, SD = 0.81, p = 0.027, Cohen’s d = 0.22) and the speech condition (mean −0.17 μV, SD = 0.66, p = 0.025, Cohen’s d = 0.24). The average self-reported listening effort measured using the VAS was significantly higher under the speech-in-noise condition than under the speech condition (Figure 5: p < 0.001). Participants reported greater difficulty in listening under the speech-in-noise condition than under the speech condition.

Under the speech condition, the SPN amplitude at the F3 electrode exhibited a significant negative correlation with performance in the speech-in-noise test (Figure 6A: r = −0.49, p = 0.009). Note that an increase in the SPN amplitude corresponds to a shift in the negative direction. Under the speech-in-noise condition, the SPN amplitude at the F3 electrode was not correlated with performance in the speech-in-noise test (Figure 6A: r = 0.095, p = 0.63). This indicates that the influence of prediction on performance in the speech-in-noise test decreases with the addition of multi-talker babble noise. The SPN amplitude at the F3 electrode was not correlated with performance in the dichotic listening test under both the speech (Figure 6B: r = 0.21, p = 0.32) and speech-in-noise (Figure 6B: r = −0.14, p = 0.48) conditions.

Figure 7 shows the relationship between the SPN amplitude change, i.e., the difference in SPN amplitude between the speech-in-noise and speech conditions, and performance in the speech-in-noise test. Multiple linear regression analysis was conducted to further explore the relationship between the SPN amplitude change and performance in the speech-in-noise test while adjusting for performance in the dichotic listening test. The SPN amplitude change emerged as a significant predictor of performance in the speech-in-noise test (adjusted R-square = 0.12, β = 0.017, p = 0.038). This predictive relationship remained significant even after adjusting for the potential influences of the performance in the dichotic listening test.

We hypothesized that in noisy environments, the attentional resources required for speech comprehension increase, leading to changes in the SPN amplitude. Additionally, we hypothesized that these changes in the SPN amplitude would be related to performance in behavioral auditory processing tests. These hypotheses were partially confirmed by the following three findings. The SPN amplitude shifted more negatively under the speech-in-noise condition than under the speech and pure-tone conditions. There was a negative correlation between the SPN amplitude under the speech condition and performance in the speech-in-noise test; however, no correlation was observed between the SPN amplitude under the speech-in-noise condition and performance in the speech-in-noise test. Finally, the change in SPN amplitude induced by background noise was related to performance in the speech-in-noise test.

Recent studies on predictive coding have reported that attention increases the gain of neurons, leading to prediction errors (Friston, 2009; Feldman and Friston, 2010). Based on this idea, the increase in the SPN amplitude under the speech-in-noise condition can be interpreted as follows: the addition of background noise decreased the clarity of the input from the bottom-up processing, thereby distorting the auditory feedback loop. To create accurate predictions of speech likely degraded by background noise (note that “accurate” here is used differently from “uncertainty” of the feedback stimulus), more attentional resources are required than those used in the speech condition. This is also suggested by the increased self-reported listening effort in the speech-in-noise condition compared to in the speech condition. Thus, the increase in attentional gain due to the added background noise may be reflected in the SPN amplitude. Similarly, understanding speech in a noisy environment is much more difficult than detecting pure tones in a quiet environment (Rudner and Lunner, 2014). Thus, in the speech-in-noise condition, there might have been an increase in attentional gain and SPN amplitude compared to in the pure-tone condition.

The second finding suggests that prediction is related to performance in speech comprehension tasks in noisy environments. In this experiment, the SPN amplitude under the speech condition was negatively correlated with performance in the speech-in-noise test. However, the SPN amplitude under the speech-in-noise condition was not correlated with performance in the speech-in-noise test. This result partially deviates from our hypothesis. Our findings did not show an association between SPN amplitude and performance in the dichotic listening test. This suggests that SPN amplitude does not reflect the role of selective auditory attention (Kerlin et al., 2010; Lesenfants and Francart, 2020) in challenging speech comprehension tasks with low signal-to-noise ratios, which involve spatially separating competing and target stimuli to focus appropriately on the target stimulus. In other words, the observed correlation between SPN amplitude and performance in the speech-in-noise test may be due to factors other than individual differences in selective auditory attention.

Schubert et al. (2023) reported that individual differences in predictive tendencies observed in the left anterior temporal region are related to performance in speech comprehension tasks in quiet environments. An interesting difference between their findings and ours is that the SPN as an indicator of predictive processes in a quiet environment correlated with performance in speech comprehension tasks in noisy environments. Individuals who exhibit increased SPN amplitude under the speech condition, which is a quiet environment without background noise, may rely more heavily on prediction than on sensory input from bottom-up processing in speech comprehension. Schubert describes this as a “strong predictive tendency.” So far, a strong reliance on prediction has been associated with maladaptive auditory experiences, such as hallucinations and tinnitus (Sedley et al., 2016; Schilling et al., 2023). However, a moderate reliance on prediction may be more robust to distractors, such as background noise, than a strong reliance on sensory input from bottom-up processing. Individuals who moderately rely on prediction in speech comprehension are expected to allocate more attentional resources to generating predictions than those who heavily rely on sensory input from bottom-up processing. This difference in attentional gain may have increased the SPN amplitude. However, it is important to note that the effect size of SPN amplitude under the speech-in-noise condition was relatively small.

The correlation between the SPN amplitude under the speech-in-noise condition and performance in the speech-in-noise test was weaker than that between the SPN amplitude under the speech condition and performance in the speech-in-noise test. This result aligns with those of previous studies showing that the influence of predictive tendencies in the left frontal region changes as the number of distracting speakers increases, making speech comprehension more difficult (Schubert et al., 2023). Together with the third finding, these results indicate individual differences in the degree to which the SPN amplitude increases during speech comprehension tasks in noisy environments. One possible explanation is that there are individual differences in the allocation of the level of attention needed to suppress background noise. The performance of participants who exhibited a greater increase in the SPN amplitude under the speech-in-noise condition than under the speech condition tended to be worse in the speech-in-noise test. The human auditory cortex rapidly and selectively suppresses background noise, enhancing the neural encoding and perception of target stimuli (Leonard et al., 2016; Khalighinejad et al., 2019). Participants who adapted well to background noise likely allocated attention more efficiently, resulting in no significant increase in SPN amplitude. Conversely, participants who struggled to adapt to background noise required more attention, leading to an increase in SPN amplitude. However, this is speculative, as the experimental design did not allow for the differentiation of neural activity related to attention directed at background noise from that related to attention directed at the target stimulus. Additionally, prediction and attention are not mutually exclusive but interactively influence sensory processing (Marzecová et al., 2018). Therefore, further validation studies are needed to confirm these relationships.

Understanding the relationship between SPN and speech comprehension may be beneficial for developing assessment methods for LiD in individuals with normal audiograms. SPN is a predictive neural activity that is not influenced by sensory input from bottom-up processing or the preparation of motor responses. Therefore, unlike traditional behavioral auditory processing tests used to evaluate speech comprehension in LiD, SPN is not affected by peripheral hearing loss undetectable by audiograms. This implies that it may serve as a tool for evaluating the influence of top-down processing in speech comprehension. Additionally, this insight could aid in developing new auditory rehabilitation strategies for individuals struggling with speech perception in challenging auditory environments, offering significant clinical benefits. For example, neuromodulation therapies utilizing SPN (Kukke et al., 2017; Wang and Chen, 2019; Mandalà et al., 2021) could enhance the predictive processes occurring just before speech comprehension, potentially improving the speech comprehension performance.

This study has some limitations. First, the effect size of the SPN amplitude under the speech-in-noise condition was relatively small. This suggests that the practical applications of this result may be limited, and caution is needed when interpreting these findings (Primbs et al., 2023). One possible reason for this could be the impact of the online analog filter. In this study, we applied an online analog filter with a range of 0.5–70 Hz for noise removal. However, the commonly used value for high-pass filters in CNV research, including SPN studies, is 0.05 Hz (Linssen et al., 2011). This difference in filter settings may have affected some or all of the slow negative components of the EEG signals. Second, SPN amplitude is influenced by multiple factors, complicating its interpretation (Kotani et al., 2015; Kotani et al., 2017). For instance, aspects such as arousal state, general emotion, and motivation may affect the modulation of SPN amplitude. Particularly, in the design of this experiment, it is impossible to separate the aspects of motivation related to reward and predictive coding within the SPN components, suggesting that further investigative studies are necessary. Additionally, top-down processing in speech comprehension involve factors, such as language, attention, memory, and working memory (Rudner and Signoret, 2016; Schiller et al., 2022). This study did not investigate which of these factors influenced SPN amplitude in speech comprehension. In conjunction with the first limitation, future research should aim to address these experimental shortcomings and develop protocols to investigate the factors that remain unidentified. Third, the controlled environment necessary for maintaining scientific accuracy may not reflect the complexity and unpredictability of real-world auditory scenarios. While ideal for isolating variables, such a laboratory setting may not consistently represent the varying levels and types of noise encountered in real-life situations. This discrepancy between experimental and real-world conditions could raise concerns about the ecological validity of the conclusions (Protzak and Gramann, 2018).

In conclusion, our results suggest that predictive brain activity, as reflected by SPN amplitude, may serve as a potential indicator of speech comprehension performance in noisy environments. Importantly, SPN represents predictive neural activity that is not influenced by sensory input from bottom-up processing. The application of predictive attentional resources several seconds before a speech perception event may be a factor influencing performance in speech comprehension under noisy conditions. Further validation studies are necessary to confirm the practical applications of these findings.