Two speech neuroprostheses decode attempted speech at conversational speed
Stanford and UCSF teams separately decoded attempted speech from cortical implants in people who could no longer speak intelligibly, reaching 62 and 78 words per minute as text, with UCSF also driving a synthesized voice and a digital avatar.
What happened
In the Stanford study, a participant with ALS had intracortical microelectrode arrays implanted in speech-related cortex. Neural activity during attempted speech was decoded into text at 62 words per minute, with a 9.1% word error rate on a 50-word vocabulary and 23.8% on a 125,000-word vocabulary. In the UCSF study, a participant left with severe paralysis by a brainstem stroke had a high-density surface electrode array placed on the cortex. Her attempted speech was decoded into text at a median of 78 words per minute, and also into synthesized speech and the facial movements of a digital avatar.
Why it matters
Earlier communication BCIs were limited to spelling or cursor control at a fraction of these speeds. Both papers showed that decoding attempted speech directly from the brain can approach the pace of natural conversation, roughly 150 words per minute.
Caveats
Each study had one participant, error rates on large vocabularies are still high, and both systems needed wired connections and research staff. Durability at home over years has not yet been shown.