Speech samples from over 1000 individuals with impaired speech havebeen submitted for Project Euphonia, aimed at improving automated speechrecognition systems for disordered speech. We provide an overview ofthe corpus, which recently passed 1 million utterances (>1300 hours),and review key lessons learned from this project. The reasoning behinddecisions such as phrase set composition, prompted vs extemporaneousspeech, metadata and data quality efforts are explained based on findingsfrom both technical and user-facing research.
@inproceedings{macdonald21_interspeech, title = {Disordered Speech Data Collection: Lessons Learned at 1 Million Utterances from Project Euphonia}, author = {Robert L. MacDonald and Pan-Pan Jiang and Julie Cattiau and Rus Heywood and Richard Cave and Katie Seaver and Marilyn A. Ladewig and Jimmy Tobin and Michael P. Brenner and Philip C. Nelson and Jordan R. Green and Katrin Tomanek}, year = {2021}, booktitle = {Interspeech 2021}, pages = {4833--4837}, doi = {10.21437/Interspeech.2021-697}, issn = {2958-1796},}Cite as:MacDonald, R.L., Jiang, P.-P., Cattiau, J., Heywood, R., Cave, R., Seaver, K., Ladewig, M.A., Tobin, J., Brenner, M.P., Nelson, P.C., Green, J.R., Tomanek, K. (2021) Disordered Speech Data Collection: Lessons Learned at 1 Million Utterances from Project Euphonia. Proc. Interspeech 2021, 4833-4837, doi: 10.21437/Interspeech.2021-697