Dissecting Challenging Speech Perception: Investigating How Stimulus Challenges and Task Demands Affect the Processing of Speech in Noise
Abstract
In this dissertation, I investigate how linguistic, acoustic, and task-based challenges impact speech perception. In the first experiment, participants (n=162) completed a sentence transcription task, indexing intelligibility, with varying levels of background babble noise. These sentences were high or low in semantic ambiguity (HA: two or more homophones; LA: no homophones). A subset of these participants (n=53) also completed a brief cognitive battery. General cognitive ability (g) predicted sentence intelligibility, especially in lower signal-to-noise ratios (SNR). After controlling for g, working memory was a significant predictor of performance in HA sentences only. This suggests that working memory may be particularly important for overcoming semantic ambiguity.
In the second experiment, we utilized anterior temporal lobe (ATL) optimized functional magnetic resonance imaging (fMRI) to investigate neural activation in response to the same acoustic (SNR) and semantic (ambiguity) challenges. Participants (n=30) listened to the same HA and LA sentences with and without background babble noise. All speech conditions showed activation in core speech areas, including bilateral superior temporal gyrus (STG), left inferior frontal gyrus (LIFG), and left ventral ATL. Sentences in noise, compared to clean speech, elicited greater activation in multiple-demand and core speech areas. HA sentences, compared to LA, elicited greater activation only in the LIFG. This study is one of the first to show this dissociation based on challenge type within the same participants.
The third experiment used fMRI to investigate how task demands – whether people are listening to understand, or to report the words they can hear – affect neural activation in response to the same speech stimuli. Participants (n=29) listened to sentences in four different levels of background babble noise while completing either an intelligibility or comprehension task. Both tasks activated core speech areas. As the SNR decreased, both tasks activated multiple-demand areas. When the two tasks were directly compared, activation in the left STG correlated with intelligibility, and activation in several clusters in frontal, parietal, and temporal regions correlated with comprehension.
Together, the results from these studies highlight key differences in cognition and neural activity underlying different types of listening challenges, based on specific stimulus and task demands.