Scientists are making significant strides forward in brain-computer interface (BCI) technology, and a newly developed system can translate our thoughts into text or sound.
It’s essentially an inner speech decoder, developed by researchers from institutions across the US. In tests on four volunteers with severe paralysis, the decoder hit an accuracy rate of up to 74 percent in translating thoughts into audible speech.
The potential here is for a BCI that can help those with speech or motor impairments to communicate more effectively than ever before, though there’s still work to be done improving how accurate and personalized the system is.
Previous BCIs have relied on brain activity that is created when a paralyzed person tries to speak or write, even though their bodies cannot carry out the action. This new technology gets a step closer to the source.
Related: Woman’s Brain Implant Turns Her Thoughts Into Speech in Real Time
“If you just have to think about speech instead of actually trying to speak, it’s potentially easier and faster for people,” says neuroscientist Benyamin Meschede-Krasa, from Stanford University in the US.
This new BCI is based on an implant designed to measure neural activity and spot patterns relating to units of speech called phonemes. These phonemes can then be built into sentences.
Machine learning was used to train the BCI to connect brain signals to words as the four participants thought about them, specifically in the motor cortex part of the brain responsible for movement (including speaking).
The researchers found that there was overlap between certain brain patterns when the volunteers tried to speak (which would involve signals connected to engaging muscles), and when they only imagined words and phrases (which wouldn’t).
Although there was overlap, the signals could be distinguished from each other. With some probability calculations thrown in, in terms of what phonemes and words usually go together, the new BCI can recognize up to 125,000 words using only inner speech.
“These patterns appeared to be a similar, but smaller, version of the activity patterns evoked by attempted speech,” says neuroscientist Frank Willett, from Stanford University.
“We found that we could decode these signals well enough to demonstrate a proof of principle, although still not as well as we could with attempted speech.”
There’s still a long way to go here, and the BCI often fell far short of that 74 maximum accuracy rating. However, by utilizing upgraded implant technology and mapping more of the brain for thought cues, the researchers are confident that the system can be quickly improved over the next few years.
Another issue to overcome is the potential for translating, logging, and speaking out inner monologues that are intended to be kept private – not something you want in a BCI. Safeguards like thinking of a special password to start and stop decoding could be implemented here, the researchers suggest. This concept was trialed in the experiments with 98 percent accuracy.
We’re now seeing plenty of progress made in this area of technologyL earlier this year another study was published on real-time thought decoding, though it was personalized to a single individual.
“The future of BCIs is bright,” says Willett. “This work gives real hope that speech BCIs can one day restore communication that is as fluent, natural, and comfortable as conversational speech.”
The research has been published in Cell.
