Inner-speech decoding on command could help paralyzed people speak

A groundbreaking study demonstrates that the brain’s inner monologue can be decoded on command, a development that could one day help people with severe speech paralysis communicate more naturally. In a proof-of-concept trial, researchers recorded neural activity from four participants with paralysis and trained artificial-intelligence models to interpret imagined speech. The system achieved up to 74 percent accuracy on a vocabulary of 125,000 words, and researchers also introduced a password-based safeguard to prevent decoding unless a user activates it with a chosen thought. The work, conducted with implanted brain sensors in the motor cortex, signals a potential advance over existing brain-computer interfaces that rely on attempting to speak aloud, which can be physically exhausting for users with conditions like ALS or brainstem stroke.

The team’s approach hinges on comparing inner speech, the silent orchestration of thought before words appear, with attempted speech, where users try to vocalize despite impaired muscles. While both modes activate overlapping regions in the motor cortex, inner speech generally elicits weaker neural signals. By combining new inner-speech data...