The Speech Accessibility Project’s second challenge has been accepted to the Neural Information Processing Systems 2026 Competition Track.

The NeurIPS Competition Track is highly competitive, and the project’s competition is one of 16 selected.

“Competitions play a unique role in advancing machine learning research by bringing together communities around well-defined, impactful challenges for fair comparison of methods,” NeurIPS said in a release. “They foster collaboration, encourage reproducible research, and provide shared benchmarks that accelerate progress across diverse domains.”

The Speech Accessibility Project’s competition allows researchers to contribute to large, significant improvements in automatic speech recognition for people with disabilities.

The project’s first challenge resulted in a word error rate reduction from 18% to 8% on the speech of people with Parkinson's disease and other disorders.

The new competition significantly expands the scope of the competition by offering a greater diversity of speech patterns and streaming ASR. A cash prize will be given to every team that places a system on the Pareto frontier of system latency and system accuracy. Exactly one non-streaming system will win, and at least one streaming system will win.

“Teams participating in this competition really have the chance to go down in the history books by creating, for the first time, accessible speech technology with latencies low enough for real-world on-device deployment,” said Speech Accessibility Project Leader Mark Hasegawa-Johnson, a professor of electrical and computer engineering at Illinois.