Tehran University of Medical Sciences Auditory and Vestibular Research 2423-480X 0 0 2026 05 26 1541 1541 Amirhossein Sameti Audiology Department, School of Rehabilitation, Tehran, University of Medical Sciences, Tehran, Iran Nematollah Rouhbakhsh Audiology Department, School of Rehabilitation, Tehran, University of Medical Sciences, Tehran, Iran Amir Homayoun Jafari Department of Medical Physics and Biomedical Engineering, School of Medicine, Tehran University of Medical Sciences, Tehran, Iran; Research Center for Biomedical Technologies and Robotics, Tehran University of Medical Sciences, Tehran, Iran Zahra Shirzhiyan Computational Neuroscience, Institute of Medical Technology, Brandenburg University of Technology Cottbus-Senftenberg, Cottbus, Germany 2026 02 04 2026 04 28 Background and aim: The brain's ability to resolve rapid temporal cues such as voice-onset time (VOT) is essential for speech perception in challenging listening environments. We tested whether central auditory compensation for VOT in noise is reflected in the fidelity of cortical auditory evoked potentials (CAEPs) using a neural-network classifier and a cross-condition similarity metric. Methods: Electroencephalography (EEG) was recorded from 22 normal-hearing adults in response to /ka/ and /ga/ syllables with varying VOTs, presented in quiet and noise (+7 dB Signal-to-noise ratio). We measured CAEPs’ peak amplitude (N1-P2), employed a convolutional neural network (CNN) to classify CAEPs by syllable identity, and computed a cross-condition correlation (rcc) to quantify the similarity between responses in quiet and noise. Results: Background noise significantly reduced N1-P2 amplitude, behavioral performance, and CNN classification accuracy, confirming the degradation of phoneme-specific neural representations. Critically, inter-subject variability in behavioral speech in noise performance was significantly correlated by both rcc (r=0.443, p=0.02*) and CNN accuracy in noise (r=0.492, p=0.01*). Individuals with higher behavioral speech-in-noise (SIN) scores exhibited CAEPs in noise that were more similar to their clean-speech responses (higher rcc) and more discriminable by CNN. Scalp topography displayed the highest rcc values over fronto-central regions, with the strongest correlation between rcc and SIN performance. Conclusion: The convergence of our findings demonstrates that successful SIN perception relies on the brain's capacity to maintain a stable, noise-invariant cortical representation of speech, particularly in fronto-central auditory regions. These EEG-derived metrics may serve as a research tool for future clinical investigations. https://avr.tums.ac.ir/index.php/avr/article/view/1541 https://avr.tums.ac.ir/index.php/avr/article/download/1541/685