Read speech protocol criteria for speech-based health screening applications

Brian Stasak, Julien Epps

Thursday, December 15th, 2022, 3pm – 3.30pm


Automatic speech-based processing using machine learning is expanding in digital healthcare, bolstering potential as a non-invasive, remote medical screening tool. There is currently a need for deeper understanding of read speech protocols and applying systematic measurements to help tailor new protocols with deliberate attribute criteria. This study investigates eight read speech protocols commonly utilized to study speech behaviours and proposes two new protocols with greater criteria extremes. An investigation of text, phonetic, linguistic, and affective proposed criteria automatically extracted from these read speech protocols reveals important merits and limitations for use in speech-based digital health screening applications.