Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neural encoding of phonemes in continuous speech: Individual differences in categorical vs. gradient speech perception
Poster Session F, Friday, October 2, 2:45 - 4:45 pm, Wangari Maathai
Shengyue Xiong1, Bharath Chandrasekaran1; 1Northwestern University
Speech perception requires listeners to encode highly variable acoustic input across different talkers and contexts and ultimately mapping this variability onto stable phoneme representations. Traditional accounts emphasize that listeners perceive speech sounds categorically by discarding within-category variation. Emerging evidence, however, suggests that listeners retain sensitivity to fine-grained acoustic details and differ in how categorically or gradiently they map continuous acoustic cues onto phonemic categories. Fueling these advances are behavioral paradigms, particularly the Visual Analog Scale (VAS) task, which has emerged as a sensitive and reliable tool for capturing differences by allowing listeners to indicate where a speech sound falls along a continuous scale between category endpoints. However, it remains unclear whether these individual differences are reflected in phoneme-level neural encoding during naturalistic speech comprehension. Here, we combined a behavioral VAS task with EEG recorded during audiobook listening to test whether individual differences in speech perception suggest neural separability of phoneme categories. Thirty-six native English-speaking adults with normal hearing completed a VAS task using a 7-step /ba/–/pa/ voice onset time (VOT) continuum. VOT was manipulated from 10 to 70 ms in equal 10-ms increments. Responses were modeled using four-parameter logistic functions, and the slope parameter was used as an index of speech perception gradiency, with steeper slopes indicating more categorical perception and shallower slopes indicating more gradient perception. Participants then listened to 15 one-minute segments from Alice’s Adventures in Wonderland while 32-channel EEG was recorded. Phoneme-related potentials (PRPs) were extracted by averaging EEG responses time-locked to naturally occurring phoneme onsets within a 0–500 ms window. Analyses focused on 31 phonemes grouped into four manner-of-articulation classes: vowels, nasals/approximants, fricatives, and stops. Neural separability of phonemes was quantified using F-statistics derived from PRPs, calculated as the ratio of between-class to within-class variance across seven frontocentral electrodes (AF3, AF4, Fz, F3, F4, FC1, FC2). Pearson correlations were computed between VAS slope and neural separability across time windows. VAS slopes varied substantially across listeners (range = 0.75–8.19, M = 3.16, SD = 1.81), indicating pronounced individual differences in perceptual gradiency. PRPs showed reliable separability across phoneme manner classes, with prominent F-statistic peaks at approximately 55 ms, 117 ms, and 188 ms after phoneme onset (R1, R2, and R3, respectively). Critically, VAS slope showed a significant positive association with neural separability averaged across time (r=.35, p=.039) and particularly within the first 200 ms following phoneme onset (r = .66, p < .001). This relationship was driven primarily by the R1 (r = .66, p < .001) and R2 (r = .52, p = .003) peaks, with no significant association at R3 (r = −.15, p = .378). These findings suggest that individual differences in categorical versus gradient speech perception are reflected in early stages of cortical speech processing involved in the emergence and differentiation of phoneme representations. More categorical listeners exhibited more robust neural separability of phonemes, whereas more gradient listeners showed less distinct phoneme differentiation. Together, these results link behavioral gradiency in speech categorization to neural differentiation of phonological structure during continuous speech comprehension.
Topic Areas: Speech Perception, Methods