A groundbreaking AI-powered speech neuroprosthesis has restored the ability to speak to a person with advanced ALS — and the achievement has just been recognized with one of science's most prestigious AI awards.
Sergey Stavisky, an associate professor of neurological surgery at the University of California, Davis, received the 2026 Chen Institute & Science Prize for AI Accelerated Research for developing a brain-computer interface (BCI) that translates neural activity into fluent, natural-sounding speech in real time.
The technology
The system uses a dual-stage deep-learning pipeline:
1. The Phonetic Layer — A first AI model decodes real-time brain activity into phonemes, the foundational sound units of speech. 2. The Linguistic Layer — A second model using large language model architectures instantly arranges those phonemes into clean words, phrases, and coherent sentences.
The system achieves greater than 99% word accuracy with a processing latency of just 30 milliseconds — fast enough to match natural human conversation.
A life transformed
One participant with ALS who could no longer speak intelligibly has been using the system for over two years. The results are remarkable:
- 2.7 million words generated purely through brain signaling - Full-time employment — he now participates in video meetings and writes work messages using only his brain signals - Expressive speech — the synthetic voice, modeled on his pre-ALS recordings, allows him to modulate tone, adjust intonation, and even sing - Daily independence — the BCI provides total communicative independence, enabling rich conversations with family and friends
From movement to speech
Stavisky's path to this breakthrough began with an unexpected insight. Early in his career, while working on motor BCIs for robotic arm control, he noticed a consistent pattern across patients: while moving objects was helpful, patients universally stated that restoring their own voice was their highest priority.
That realization, combined with rapidly accelerating advances in machine learning around 2018, prompted a mid-career pivot from motor prosthetics to speech neuroscience — at a time when speech decoding was considered one of the hardest problems in neuroprosthetics.
The future
Stavisky's long-term goal is a "high-fidelity surrogate voice" — a system so natural that if someone spoke through a standard phone line, the listener would not be able to tell the voice was synthetic.
Future developments focus on shrinking the system into a fully wireless, completely internal medical implant, while expanding access to patients with stroke-induced aphasia, cerebral palsy, and other language disorders. Companies are already beginning clinical trials for commercial speech BCIs.
The prize is awarded by the Chen Institute, Science magazine, and the American Association for the Advancement of Science (AAAS).
