Stanford researchers have created a brain-computer interface that translates imagined speech into text, enabling communication for individuals with severe paralysis
Stanford researchers have developed a groundbreaking brain-computer interface that decodes inner speech, silent, imagined words into text. This advancement holds the potential to significantly improve the quality of life for individuals with severe paralysis, offering them a means of communication without the need for physical movement.
Decoding inner speech
Neurosurgery Assistant Professor Frank Willett, PhD, and his fellow researchers have used brain-computer interfaces (BCIs) to help people with paralysis speak again, using brain signals and inner speech.
The researchers meticulously explored whether a BCI could function based solely on neural activity evoked by ‘inner monologues’. Their previous demonstrations have shown that a BCI can accurately detect brain signals when people with paralysis attempt to speak or write by hand, converting these signals into words.
The brain’s motor cortex contains regions that control movement, including the muscular movements that produce speech. The researchers surgically implanted a BCI, a technique that uses tiny arrays of microelectrodes, to record neural activity patterns directly from the brain. A cable hookup then feeds these signals to a computer algorithm that translates them into actions such as speech.
The researchers then used a machine learning tool to decode the neural activity picked up by the arrays into words the patient wanted to say. They trained the computer to recognise repeatable patterns of neural activity associated with each “phoneme” – the tiniest units of speech – then stitch the phonemes into sentences.
Did the BCIs decode inner speech for patients with paralysis?
The researchers studied four people with severe speech and motor impairments who had BCIs placed in motor areas of their brains. They found that inner speech evoked clear and robust patterns of activity in these brain regions. The patterns were similar, but smaller, to the activity patterns evoked by attempted speech.
The researchers found that they could decode these signals well enough to demonstrate a proof of principle, giving hope that future systems could provide fluent, rapid, and comfortable speech to people with paralysis via inner speech alone. This is important as people with paralysis often find that attempting to speak can be fatiguing and frustrating. The potential impact of this research is immense, as it could significantly improve the quality of life for individuals with severe paralysis, enabling them to communicate more effectively and independently.
One privacy concern is that BCIs could end up decoding something that the user intended only to think, not say out loud. While this could introduce errors in current BCI systems aimed at decoding attempted speech, these systems still lack the resolution and fidelity required to accurately interpret rapid, unconstrained inner speech, which may lead to garbled output. However, the researchers are working to address the possibility of accidental inner speech decoding and have promising solutions.
For current BCIs decoding attempted speech, the researchers developed a training method that helps the system ignore inner speech. This method involves training the system to recognise specific brain signals associated with intended speech, thereby reducing the likelihood of unintended speech being captured. For next-generation BCIs designed to decode inner speech directly, we implemented a password-protection system that only decodes inner speech after the user imagines a unique password. This system ensures that the BCI only interprets inner speech when the user consciously intends to communicate, enhancing privacy and user control.