ALS patient achieves 97% accuracy in natural, unscripted conversation using BrainGate2 speech interface

BrainGate consortium/BrainGate2 clinical trial (speech neuroprosthesis)

A 45-year-old ALS patient (Casey Harrell) using the BrainGate2 speech-decoding system achieved 97% accuracy in real-time conversation without error correction. After eight months of recorded training data, Harrell was able to communicate fluidly both in response to prompts and in spontaneous conversation, with a vocabulary of more than 125,000 words. The system translates motor cortex signals to text and synthesized speech matching his pre-ALS voice.

• 97% decoding accuracy in natural, unscripted conversation represents a major milestone for speech BCIs.
• Casey Harrell able to communicate within minutes of turning the system on after training.
• Vocabulary access of 125,000+ words supports natural, complex dialogue.
• Breakthrough led by UC Davis researchers David Brandman and Sergey Stavisky in collaboration with Brown and Providence VA.