Work

Perceptual learning and neural processing of noise-vocoded speech under divided attention

Year: 2024
Type: dissertation
Abstract: Human listeners can understand acoustically degraded speech and their perception of such speech can improve with practice or exposure. Both perception and perceptual learning of degraded speech are thought to rely on on attention and theoretical accounts like predictive coding suggest a key role for attention supporting these processes. However, the nature of the attentional processes involved in perceptual learning was largely unknown—it was unclear whether undivided attention is necessary for learning and whether learning relies on language-specific or domain-general processes. Furthermore, the neural underpinnings of allocating attentional resources between degraded speech and concurrent sensory inputs are unresolved. In three studies, I created an online listening task to demonstrate the rapid perceptual learning of noise-vocoded speech (Study 1) and probed into the nature of the attentional resources for speech perceptual learning in dual tasks (Study 2). I also used fMRI and machine learning to explore how brain dispenses attentional resources for degraded speech processing while listeners are engaged in a concurrent task (Study 3). The results showed that the learnability of degraded speech was not restricted by either the difficulty or the processes (i.e., visual, phonological, or lexical) of a concurrent task. Neurobiologically, degraded speech processing under divided attention was associated with elevated responses in regions related to effortful listening and attentional control. Machine learning models further delineated a group of frontotemporal brain regions governing robust speech processing and dynamically dispensing attentional resources. Taken together, these results clarify current theoretical accounts by demonstrating that undivided attention is not required for rapid perceptual learning of speech. The work also resolves the neural underpinnings for the interaction between acoustic degradation and divided attention and developes a robust and explainable pipeline to applying machine learning to sophisticated analysis of task-based fMRI data that has a limited sample size.
Author: Wang, Han
Language: English
Field: Psychology
Open Access status: green
OpenAlex ID: W7023569733
Cites: 196
Cited by:
Related to: 10