People often have inner speech, which is the silent talk in their minds, like thinking words without saying them out loud. Scientists have now found brain activity linked to this inner speech and decoded it with up to 74 percent accuracy. The research was led by Stanford University.
This work is published in Cell. It could improve communication for those who cannot speak audibly by using brain-computer interfaces, or BCIs, which are devices that read brain signals and turn them into actions or words. These interfaces start translating thoughts when a user thinks a password in their head.
BCIs help people with disabilities by using sensors placed in the brain to catch signals related to movement and convert them into controls, like moving a robotic arm. Past BCIs decoded attempted speech, where users try to speak by moving muscles, even if no sound comes out, and turned that into text. However, this can be exhausting and slow for those with weak muscle control.
Advancing to silent communication
The scientists wondered if BCIs could read inner speech instead, which might be easier since it needs no muscle effort. They recorded brain signals from tiny electrodes in the motor cortex, a brain area that handles speaking, in four people paralyzed by amyotrophic lateral sclerosis, or ALS, a disease that weakens muscles over time, or by a brainstem stroke, which damages the brain area linking to the body.
The signals for attempted speech and inner speech overlap but inner speech is weaker. Using this data, they trained artificial intelligence (AI) models to interpret imagined words. In tests, the BCI decoded sentences from a large vocabulary with good accuracy and even caught unplanned thoughts, like counting items on a screen.
The system tells attempted speech from inner speech, so BCIs can ignore silent thoughts if needed. For privacy, a password like thinking "chitty chitty bang bang" unlocks decoding, recognized almost perfectly. Current BCIs make errors in free inner speech, but more sensors and better programs could fix that. This progress points to BCIs restoring natural, fluent talk for those who lost their voice.