Do classroom noise conditions change perception and EEG without hurting performance?

Three response types measured under the same sounds

The paper asks whether perceived, cognitive, and EEG responses to classroom noise agree with one another, and whether perception and brain activity can separate conditions that performance cannot. Sixteen university students completed the same protocol under seven conditions: a no-noise control plus air-conditioning (AC) and road-traffic (RT) noise, each played back at 55, 60, and 65 dBA. Because every participant experienced every condition, comparisons are within the same people, which reduces variability between participants.

The three response types measure different things and should not be treated as interchangeable. Acoustic appraisal used three Classroom Soundscape Assessment attributes: loudness, pleasantness, and arousal rated from calm to exciting. Perceived workload used the NASA Task Load Index, reported as raw and weighted scores plus six dimensions. Cognitive performance came from four computer-based tasks covering focused attention, divided attention, decision making, and information extraction, summarized as accuracy, response time, and a performance index equal to accuracy divided by response time. EEG was recorded during the tasks with a 32-channel headset.

The experiment ran in a teaching classroom at The Hong Kong Polytechnic University with windows closed, and stimuli were played through one loudspeaker at a fixed position near the window. The room's own window-mounted air-conditioning unit was not used to generate the AC stimulus, and the control condition added no experimental recording. These choices matter for reading the results: the study compares reproduced sounds in one room, not different ventilation modes or real air quality.

No task showed a significant performance difference

Across all four tasks, the performance index did not differ significantly between the six noise conditions and the control (Tasks 1–4: t(15) = 0.06, 1.48, 1.43, and 0.73; p = .954, .159, .173, and .478). Among the noise conditions, no test of noise type, noise level, or their interaction on the performance index was significant, the smallest p value being .063 for noise type in Task 2. Accuracy and mean response time likewise showed no significant differences in any of the eight comparisons with control or in the tests among noise conditions.

The descriptive patterns help explain why this null result needs care. Accuracy approached the ceiling in Tasks 1 and 2, was lower and more variable in Task 4, and the longest response times appeared in Task 3, yet distributions largely overlapped across the seven conditions within each task. The authors argue that the short tasks, repeated across seven conditions, may have left spare capacity for performance to be protected, and they note that the participants were university students rather than children, who are described as more susceptible to noise-induced disruption.

The paper is explicit that a non-significant difference is not evidence of equivalence. Its performance result therefore says that these four short tasks did not detect a difference under these conditions; it does not establish that the acoustic conditions were harmless or that learning would be unaffected.

Loudness, pleasantness, and arousal followed sound level

Acoustic appraisal separated the conditions far more sharply than performance did. Compared with the control, the mean of the six noise conditions was rated significantly louder (Δ = 5.77 on a 0–20 scale, t(15) = 6.51, p < .001), less pleasant (Δ = −4.03, t(15) = −3.26, p = .005), and more arousing (Δ = 4.11, t(15) = 4.89, p < .001). All three ratings also varied significantly with sound level among the noise conditions: loudness F(2, 30) = 11.72, p < .001, ηp² = .439; pleasantness F(2, 30) = 7.86, p = .004, ηp² = .344; arousal F(2, 30) = 5.41, p = .010, ηp² = .265.

Noise source mattered less. Only loudness showed a significant source effect, with RT noise rated louder than AC noise on average (F(1, 15) = 6.62, p = .021, ηp² = .306), and loudness also showed a significant source-by-level interaction (F(2, 30) = 6.89, p = .004, ηp² = .315), reflecting a steeper rise under RT noise. RT noise was rated slightly quieter than AC at 55 dBA but louder at 60 and 65 dBA. Pleasantness and arousal showed no significant source or interaction effects.

The authors interpret this as appraisal responding to both the presence and the level of noise, while the two sources differed only in loudness within the tested range. They also caution that the three retained attributes did not include the monotonous-versus-chaotic descriptors that separated sources in earlier classroom work, and that ratings were given after a block of four tasks, so they may reflect task context as well as sound.

Workload was higher under noise but not level-dependent

Perceived workload showed a weaker but still consistent pattern. Raw NASA-TLX scores were 9.59 points higher under noise than under control (t(15) = 2.47, p = .026) and weighted scores 10.17 points higher (t(15) = 2.54, p = .022), both on a 0–100 scale. Among the six dimensions, temporal demand (Δ = 13.54, p = .029) and effort (Δ = 8.59, p = .047) were significantly higher under noise, while mental demand, physical demand, performance, and frustration did not differ significantly (p = .053–.213).

Among the noise conditions, neither raw nor weighted workload varied significantly with noise type, noise level, or their interaction (all p ≥ .316). At the dimension level, mental demand was higher under RT than under AC noise (F(1, 15) = 5.30, p = .036, ηp² = .261), and physical demand showed a significant source-by-level interaction (F(2, 30) = 5.09, p = .018, ηp² = .253).

The authors connect the higher effort and temporal-demand ratings with the idea that performance was maintained through additional effort. That reading is an interpretation, not a direct test: workload was rated once per condition block after all four tasks, so each rating may have averaged over demands that differed between tasks.

Task-period EEG differed by source or level in some regions

EEG power during the tasks also separated conditions, but only in specific task and scalp-region combinations. The analysis used region-of-interest linear mixed-effects models on the normalized change in band power, comparing noise conditions with one another rather than with the control. Across the omnibus tests, the paper reports six main effects of frequency band, four main effects of noise type, and six main effects of noise level, with further effects in particular tasks and regions.

Two examples show how localized these differences were. In Task 1, the AC–RT difference in posterior-attention power reversed direction across levels; in Task 4, the midfrontal response was larger under RT than under AC. The authors describe the EEG results as showing that task-period power was sensitive to source, and less robustly to level, in some task and region combinations. They explicitly do not claim that either source has a consistent neural signature or that one task or region was more affected than another.

Two measurement details limit how far these EEG findings travel. Region-of-interest labels describe scalp locations rather than neural sources, and the normalized change used in the mixed models is a different quantity from the relative band power used in the correlation analyses, so the two sets of EEG results do not describe the same effect.

EEG covaried mainly with loudness and pleasantness

Within participants, the clearest cross-domain link was between EEG and acoustic appraisal. Of all the correlations between questionnaire measures and relative band power in four regions of interest, more than a quarter were significant, compared with about four expected by chance at α = .05. The significant associations involved loudness (11) and pleasantness (9) across all four regions and arousal (2) in the temporal region, while none involved weighted workload.

The direction was consistent: higher loudness ratings went with higher relative theta and alpha power and lower relative gamma power, whereas higher pleasantness ratings went with lower relative theta power and higher relative gamma power. The strongest associations were between loudness and gamma power in the central-motor region (rrm = −.547, 95% CI [−.70, −.35], p < .001) and in the posterior-attention and temporal regions (both rrm = −.535, p < .001), and between pleasantness and gamma power in the temporal region (rrm = .477, 95% CI [.26, .65], p < .001). Each correlation was based on 15 participants and 77 paired observations.

Associations with performance were rare: only two of 80 correlations between task performance index and same-task band power were significant, fewer than the four expected by chance. The authors note that the scarcity of associations for workload, arousal, and performance may partly reflect how little those measures varied across conditions, since a response that changes little offers little within-participant variance to relate to EEG. They also state that the two sets of associations were not compared directly, so this contrast does not show that EEG is more closely coupled to perception than to performance.

Partial consistency changes how noise should be assessed

The three response types were only partly consistent. Appraisal changed with the acoustic conditions while performance did not, and EEG tracked loudness and pleasantness but rarely arousal or performance and never workload. The authors' practical conclusion follows from that split: assessments of air-conditioning and road-traffic noise in classrooms should include perceived responses alongside cognitive performance, because conditions that produced no detectable difference in short cognitive tasks were clearly distinguished by how they were perceived.

They also place a boundary on EEG. Because its sensitivity depends on analytical choices, they treat it as a research tool for studying how noise affects task processing rather than as a criterion for judging acoustic quality. And because appraisal depended more on sound level than on source within the tested range, they suggest as a hypothesis that the level of sound reaching a classroom may matter more for perceived quality than whether it comes from air-conditioning or from traffic through open windows.

That hypothesis is not tested here. The experiment reproduced recordings through a loudspeaker with windows closed and did not change ventilation mode, air quality, or thermal conditions, and the sample was 16 university students in one classroom with brief computer-based tasks. The paper's own priority is narrower and more useful than a general call for multimodal measurement: when short task scores show no difference between classroom noise conditions, loudness, pleasantness, and the theta–gamma pattern that follows them can still show that students experienced those conditions differently.

Sources used in this answer