Developing a Freely Accessible Dataset for Training the Assessment of Connected Speech in Aphasia

Abstract

This paper outlines the steps required to create a freely accessible dataset for training students, clinicians, and researchers on how to use the Auditory-Perceptual Rating of Connected Speech in Aphasia (APROCSA) tool developed by Casilio et al. (2018) to assess connected speech in people with aphasia. We address the utility of collecting connected speech samples, the problems with analyzing them, and the reasons that an updated, freely accessible training protocol is required for APROCSA. We specify the primary need to obtain a freely accessible dataset containing video examples. We discuss the protocol we followed to collect connected speech samples from participants and the process for editing the videos to remove personally identifiable information so that they may be freely shared. We describe the team of expert raters that we assembled and outline the steps for how consensus ratings were achieved for all participants. Finally, we analyze our dataset by tabulating the interrater reliability for each participant and addressing reliability for APROCSA connected speech features.