Brian Stasak, Julien Epps, Mark Larsen, Helen Christensen
Thursday, December 15th, 2022, 2.30pm – 3pm
Abstract
While large amounts of recorded speech audio data are collected for medical analysis, there is not a compact guideline available that outlines which human-rated annotations are important to consider when analyzing speech from individuals with potential mental illness. Herein, an annotation guideline is proposed that highlights fifty-two different speech-audio recording transcription considerations, including several new ones related to voice accent, prosodic, intelligibility, quality, auxiliary behavior, task compliance, disfluency, and noise factors. Further, a free, new software tool for recording recommended speech-based mental health annotations is provided to help scientists who may be unfamiliar with speechaudio data collection.
SST2022 