Speech disorders can affect people's ability to communicate with others effectively, as it limits a listener's ability to decode messages effectively. Particularly, we focus on a speech disorder type called dysarthria, where speech motor muscles weaken, resulting in imprecise articulation, slurred or slow speech, or changes in voice characteristics.

Previous work has seen a positive effect on intelligibility when listeners are previously exposed to a specific speaker with dysarthria, i.e. short-term perceptual adaptation. In these studies, listeners were presented with an explicit familiarisation phase prior to the transcription task for the same speaker. During the familiarisation phase, listeners listen or read-while-listen to stimuli. However, it is unclear how much exposure time is necessary for an improvement in the transcription task, as previous studies have proposed different lengths of exposure (Borrie et al., 2012). To the best of our knowledge, no study has controlled for different lengths of exposure in the familiarisation phase to study its effects in listener short-term perceptual adaptation for dysarthric speech. We hypothesise that exposing listeners to different amounts of stimuli (up to 5 minutes of recordings) prior to the transcription task will improve listeners' transcription accuracy rates, but that this improvement will not be linear, i.e. improvements will plateau as exposure time increases.

To investigate our hypothesis, we propose a between-subjects study where listeners are assigned to different exposure conditions. Conditions are divided by speaker (speaker A or B), and length of the familiarisation phase prior to testing (100 sentences, 50 sentences, 25 sentences, 0 sentences). After exposure, listeners complete a transcription task on unseen sentences from the same speaker as in the familiarisation phase. Speakers A and B are selected from the Speech Accessibility Project (Hasegawa et al., 2024), a dataset created to improve automatic speech recognition systems for those with a speech impairment. Speakers were selected based on their overall intelligibility, aimed to be between 40-60%, to avoid selecting speakers that might be too challenging to understand (and therefore we might see very minimal effects from exposure), or too intelligible (and therefore we would see ceiling effects). We will present our results for the aforementioned study, and discuss possible future directions for this research, such as replicating it with more challenging speakers (i.e. those with lower overall intelligibility).

References:

• Borrie, S. A., McAuliffe, M. J., & Liss, J. M. (2012). Perceptual learning of dysarthric speech: A review of experimental studies. Journal of Speech, Language, and Hearing Research, 55(1), 290-305.
• Hasegawa-Johnson, M., Zheng, X., Kim, H., Mendes, C., Dickinson, M., Hege, E., ... & MacDonald, B. (2024). Community-supported shared infrastructure in support of speech accessibility. Journal of Speech, Language, and Hearing Research, 67(11), 4162-4175.