The theoretical backdrop is what some researchers call the sensory-first account of autism. According to this view, deviations in sensory processing appear early in development, sometimes in infancy, and cascade upward into the higher perceptual, cognitive, motor and social-communication differences that define the condition. Multisensory integration, the process by which the brain combines information from different senses into a single coherent percept, sits near the base of this cascade. Compared with unimodal processing, integration yields faster and more accurate perception, localization and discrimination, and it underpins language processing in particular, where prosody, background noise, lip movements and facial expressions must all be reconciled. Prior work has shown that autistic individuals gain less from seeing a speaker’s face when listening to speech in noise than typically developing individuals do, and that this difference is not explained by impaired auditory cortical processing, pointing instead to integration itself.

A central concept in this literature is the temporal binding window, the range of time lags within which stimuli from different senses are highly likely to be fused into one percept. The window is not fixed: it narrows across typical development, from infancy through middle childhood and adolescence, and its width varies with stimulus characteristics and task demands. In autistic children and adolescents, multiple studies have found the window to be enlarged relative to typically developing peers, and some work links this widening to symptom severity. Because the window is dynamic, researchers wondered whether it could be actively trained. In typically developing adults, a single brief session of simultaneity judgment training with feedback has been shown to narrow the window, and in some studies to improve word recognition in noise, with effects persisting at least seven days. Similar benefits have been reported in adults with schizophrenia. Whether the same plasticity exists in autistic youth was an open question, addressed previously by only two small studies with inconclusive results.

The Hannover team recruited 19 autistic children and adolescents aged 8 to 17, all with confirmed diagnoses under ICD-10 criteria verified by gold-standard instruments such as the ADOS-2 and ADI-R, along with normal or corrected-to-normal vision, normal hearing and fluent German. After exclusions for implausible or random response patterns, 16 participants entered the analysis at baseline and immediately after training, and 15 completed the follow-ups at seven and fourteen days. The training itself was elegantly simple: a modified simultaneity judgment task in which a white ring flashed on a screen and a brief 1850 Hz beep sounded, separated by time lags between 25 and 150 milliseconds. After each trial, participants received feedback on their answer, both as a written word and as a green happy or red sad emoji. Synchronous and asynchronous trials were presented in equal proportions to prevent response biases.

To measure the training’s impact, the researchers assessed each participant at four time points: before training, immediately after, seven days after and fourteen days after. The main probe was the simultaneity judgment task, in which stimulus onset asynchronies ranged from 0 to 250 milliseconds in 25-millisecond steps, with only visual-leading conditions to enhance ecological validity. The team fitted generalized linear mixed-effects models with a binomial distribution, treating age as a covariate, and derived psychometric functions from which they estimated the temporal binding window using a 50 percent response criterion, with participant-level bootstrap resampling to quantify uncertainty. The word recognition task, adapted from a paradigm by Ross and colleagues, served as the second probe: German disyllabic nouns were presented either as audio alone with a static face or as full audiovisual video clips, mixed with either no background noise or white noise at a demanding signal-to-noise ratio of minus 12 decibels.

The results were nuanced. The models revealed a significant interaction between stimulus onset asynchrony and point of measurement, indicating that the shape of the psychometric function changed across the study. The estimated temporal binding window shrank from 189.8 milliseconds at baseline to 153.5 milliseconds immediately after training and 145.2 milliseconds at seven days, before drifting back toward baseline at 161.2 milliseconds by day fourteen. Bootstrap confidence intervals confirmed that the reductions at the first two post-training points were statistically reliable, while the day-fourteen difference was not. Yet the immediate effect carried a warning sign: the slope of the psychometric function became shallower right after training, the opposite of what genuine perceptual sharpening would predict. Seven days later, the slope was steepest of all, a pattern more consistent with truly improved simultaneity perception.

This distinction matters, and the authors handle it with unusual candor. A shallower function immediately after training could reflect decreased sensitivity to timing differences, but it could equally reflect a shift in decision strategy or response criteria. Previous research suggests autistic children tend toward more conservative response behavior, answering no when unsure, and show heightened reactivity to errors and greater performance impairment after frustrating negative feedback. It is plausible that the emoji-and-text feedback in the training pushed participants toward reporting stimuli as asynchronous whenever they felt uncertain, temporarily flattening the curve without any change in perception itself. Because this flattening was transient and the expected steeper curve emerged at day seven alongside the narrowed window, the authors suggest the training may still have contributed to more precise simultaneity perception, but they emphasize that future studies must control for response criteria, for example by estimating individual decision thresholds, before drawing firm conclusions.

The speech results were more sobering. As expected, word recognition in noise was substantially better in the audiovisual condition than in the audio-only condition, with a difference of roughly 12.5 percentage points at minus 12 decibels, confirming that participants did benefit from seeing the speaker. But contrary to the hypotheses, and to findings in typically developing adults and schizophrenia patients, there was no significant improvement in word recognition at any point after training. The authors offer several candidate explanations. One is interindividual heterogeneity: prior work by Feldman and colleagues found that training effects generalized only in autistic youth with above-average nonverbal intelligence, suggesting subgroup-specific responsiveness. A deeper possibility comes from work by Crosse and colleagues indicating that linguistic and non-linguistic audiovisual stimuli follow different developmental trajectories in autism and engage distinct cortical networks; sharpening simultaneity perception for simple flashes and beeps may simply not be sufficient to improve the integration of complex speech, which recruits hierarchically organized brain networks and multiple integrative mechanisms.

The study also grapples with a conceptual wrinkle that may have muddied the entire field. Recent work argues that the temporal window of integration, within which stimuli are truly fused into one percept, must be distinguished from the temporal window of synchrony, within which stimuli are merely judged as simultaneous. The simultaneity judgment task used here, and in most prior training studies, measures only the latter, which is neither necessary nor sufficient for genuine integration. This may explain both the absence of speech gains in the present study and the contradictory findings across earlier literature, and it motivates the adoption of paradigms such as the McGurk illusion that can probe integration directly. The authors also acknowledge their limitations candidly: the sample was small, there was no typically developing or untrained control group, causal attribution is therefore impossible, and the paradigms were originally calibrated on adults, with the minus 12 decibel noise condition proving especially challenging for the young participants.

What remains is a genuinely useful map of the terrain. A single brief multisensory training session can measurably alter simultaneity judgments in autistic youth, with window narrowing persisting for about a week and fading by fourteen days, a time course that mirrors findings in typically developing adults and carries a practical implication: any future intervention would need repetition at least every two weeks, and probably sooner. The hoped-for transfer to speech comprehension in noise did not materialize, but the study clarifies why the question is worth pursuing. Autistic individuals show specific multisensory deficits when processing naturalistic speech in noise that cannot be explained by unisensory impairments, which keeps improved integration a promising target. The path forward, the authors argue, lies in tailored paradigms that account for altered multisensory development, individual response styles and heterogeneity across subgroups, potentially aided by eye-tracking to separate inattention to visual speech from true integration differences, and by neuroimaging to reveal how training changes the underlying cortical networks. Only then can perceptual training be engineered into interventions that genuinely improve the everyday communicative lives of autistic children and adolescents.

Subject of Research: Effects of multisensory perceptual training on audiovisual temporal binding and speech-in-noise perception in autistic children and adolescents

Article Title: Multimodal speech perception in noise after multisensory perceptual training in autistic children and adolescents

Article References: Möde, L., Ghaneirad, E., Szycik, G. R., Worthmann, H., Bleich, S., & Borgolte, A. (2026). Multimodal speech perception in noise after multisensory perceptual training in autistic children and adolescents. Scientific Reports, 16(1), Article 31873. https://doi.org/10.1038/s41598-026-75297-3

Image Credits: AI Generated

DOI: 10.1038/s41598-026-75297-3

Keywords: autism, multisensory integration, temporal binding window, audiovisual training, speech perception in noise, simultaneity judgment, perceptual plasticity, temporal window of synchrony, psychophysics, response criteria, child and adolescent development, Scientific Reports