Speech Accessibility Project challenge competition
The project's challenge competition seeks to rapidly advance the state of the art in dysarthric speech recognition. It is hosting a competition on eval.ai from Nov. 1 through Jan. 31, 2025.
Submissions will be evaluated based on their ability to correctly transcribe dysarthric speech in two sequestered test sets (Test1 and Test2), with a distribution across etiologies similar to that in Train and Dev sets available to any organization that signs a data use agreement.
Prizes
The competition has two evaluation metrics: word error rate and semantic score. Semantic score is a weighted combination of BERTscore, phonetic distance, and natural language inference probability.
Scores on Test1 will be published immediately upon submission. Scores on Test2 will be published Feb. 4, 2025. The team with the lowest WER on Test2, and the team with the highest Semantic Score on Test2, will each receive a prize of $5,000 U.S. dollars.
Download the full challenge description (loads as a PDF) for more information about the competition.