Sergey Stavisky, an associate professor of neurological surgery at UC Davis, has won the Chen Institute and Science Prize for AI Accelerated Research for developing an AI-powered speech neuroprosthesis that restores communication for people who have lost the ability to speak.
Key Achievement
- An ALS patient who could no longer speak intelligibly received an implantable brain-computer interface (BCI).
- AI models trained on his brain signals decode neural activity into phonemes, words, sentences, and synthetic speech modeled on his pre-ALS voice.
- Results: Over 99% word accuracy, real-time performance (as fast as 30ms delay), ability to modulate intonation, and even sing.
- The patient has generated millions of words (2.7 million over two years), enabling rich conversations with family, independent computer use, and continued full-time employment.
How It Works
Modern BCIs generate massive neural data that traditional methods can’t handle effectively. Stavisky’s team uses multiple AI models: one decodes brain signals into sound units, while others (leveraging large language models) turn them into fluent text or voice output.
Background & Future
Stavisky shifted from motor BCIs to speech after realizing communication is patients’ top priority. His work builds on advances in intracortical recording and machine learning. Future goals include a natural-sounding “high-fidelity surrogate voice,” smaller fully-implanted devices, and broader clinical applications for conditions like stroke, aphasia, and cerebral palsy.
