This presentation describes our progress and lessons learned from a year-long data collection of speech and video recordings of people with Amyotrophic Lateral Sclerosis (ALS) and healthy controls via an easy-to-use cloud-based conversational agent, Tina. The research community can use this data to accelerate the development of biomarkers, diagnostics, therapies, and fundamental scientific understanding of ALS. 

ALS is a heterogeneous, neurodegenerative disorder of the motor neurons that affects speech and swallowing, among other motor functions. Research has shown that speech- and face-based measures of patients from conversational speaking tasks have the potential to serve as endpoints in clinical trials and in diagnostics when clinically monitoring disease progression. 

As of October 2021, we have recruited 300 participants through the EverythingALS advocacy portal. Of these, 91 were people with ALS (47 females, 44 males, mean age: 59.26 years) while 209 were healthy controls (149 females, 60 males, mean age: 46.65 years). Each  participant was encouraged to engage in weekly conversations with Tina, for about ten minutes per session, which produced video recordings of the subject’s face, full-duplex audio recordings, audio and facial measures, measurement of progression using self-reported ALSFRS-R and ROADS, and demographic information. Participants did a mean of 11 sessions (standard deviation: 10 sessions). The total number of overall sessions was 3265 (out of which 1187 were from people with ALS).

We will present the lessons learned in collecting our large-scale audio-visual corpus for ALS research, one year into the project. We will also report on audio and visual biomarkers that show significant promise for early diagnosis and progress monitoring, along with a longitudinal analysis of fast vs slow progressors. 

We find that cloud-based conversational technology and associated speech and facial biomarkers offer significant utility to the ALS community for remote patient assessment and progress monitoring.