Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Cortical dynamics of perceiving creaky voice
Poster Session C, Thursday, October 1, 10:45 am - 12:45 pm, Wangari Maathai
This poster is part of the Sandbox Series.
Benjamin Lang1, Trevor McPherson1, Lauren Ostrowski1, Kate Urrutia1, Jeffrey Xing1, Travis Ramirez1, Mingxiong Huang1,2, Timothy Gentner1; 1University of California San Diego, 2VA San Diego Healthcare System
Previous research on the neurobiology of language has focused on numerous elements of processing speech segments and suprasegmental information (Bhaya-Grossman et al., 2025; Gnanateja et al., 2025; Mai et al., 2024; Gwilliams et al., 2022; Bhaya-Grossman and Chang, 2022; Li et al., 2021; Chang et al., 2010), but voice quality remains an underrepresented variable. Here, we focus on irregular phonation (or, “creaky voice”) as an acoustic cue to rapid and overlapping changes in segmental and suprasegmental information (Garellek, 2015; Garellek, 2022). As a phonetically multidimensional signal (Keating et al., 2015), characterizing the perception of voice quality allows us to test hypotheses about the temporal integration of segmental and suprasegmental percepts that arise from similar acoustic cues on different timescales. In this study, we aim to test whether or not perceiving different linguistic representations of creaky voice (segmental vs. suprasegmental) elicits different spatiotemporal dynamics in the brain. MEG data from 65 L1 American English listeners was acquired as they actively listened to ~56 speakers from the Buckeye Corpus (Pitt et al., 2007), and completed a word identification task focused on segmental and suprasegmental use of creaky voice (adapted from Garellek (2015)). MEG was preprocessed using MNE-Python (Gramfort et al., 2013) and speech-selective sensors were identified using a one-way ANOVA comparing the 1s post-stimulus window to a 500 ms pre-stimulus baseline. Log band power was extracted across six frequency bands: delta (2-4 Hz), theta (4-8 Hz), alpha (8-12 Hz), beta (15-30 Hz), low gamma (30-80 Hz), and high gamma (80-150 Hz). The probability of a neural response to speech was estimated in each frequency band using second-order maximum noise entropy (MNE) models (Mai et al., 2024; Kozlov and Gentner, 2016). MNE models allow us to estimate the additive contribution of discrete, linguistic elements to continuous speech stimuli that evoke a neural response with minimal assumptions about the stimulus-response relationship. MNE models were trained and tested (80% train / 20% held-out) by subject on each band using 1024ms unlabeled spectrographic windows of speech in order to estimate the receptive field of neural populations in response to speech. Unlabeled MNE models were then assessed for performance above chance by computing the correlation between recorded neural activity and predicted neural activity and comparing it against a null distribution of correlations between recorded activity and 100 shuffled MNE models of predicted activity. Preliminary results from one subject indicate speech-sensitive encoding can be recovered using MNE models (85/85 speech-sensitive channels with at least one significant band exceeding 95% of the recorded-shuffled distribution). These unlabeled models establish the baseline with which models including segmental and suprasegmental labels of creaky voice will be built. Planned analyses include model validation for the remaining 64 subjects, extension of analyses to source regions of interest (ROIs) such as bilateral early auditory cortex and auditory association cortex, and models of speech stimuli combined with time-varying, discrete labels of segmental and suprasegmental realizations of creaky voice in order to investigate the spatiotemporal contribution of these linguistic elements to stimulus-driven responses.
Topic Areas: Speech Perception, Prosody