Imagine a world in which authors can write books in days, not months, using only the power of their minds. This hands-free future could be around the corner: scientists have created software that hooks up to your brainwaves and transcribes whatever you're thinking.
Brain-to-Text is the software behind this futuristic, sci-fi-style concept. It has the potential to transform the lives of those who have lost the ability to communicate effectively. Stephen Hawking, for example, often has to scroll through letters of the alphabet one at a time while typing out messages. As you can imagine, the process is slow and laborious. Software like Brain-to-Text could therefore be life-changing.
The study designed to test this new concept, published in Frontiers of Neuroscience, required participants who already had electrodes fitted in their brains. This is because external, brainwave-reading caps, which record electrical activity across the scalp, are not sensitive enough to pick up the sharp signals needed to identify individual letters. The skull blurs this sensitive information.
This limited the number of people who could participate in the trial to seven, all of whom suffered from epilepsy and already had electrodes implanted in their brain to treat it. Unfortunately for the researchers, the electrodes were only put in the regions of the brain that required rewiring, and thus were not evenly distributed everywhere.
With no way around this limitation, the participants were asked to read different passages of text aloud while their neural data was read by a computer. The passages read included JFK's inaugural speech, Humpty Dumpty, and even Charmed fanfiction.
As the individuals spoke the words, the computer had to learn to recognize the individual sounds they were making and match it to the corresponding brain wave. Eventually, the computer was able to pick up different brain patterns and match them to sounds.
The results were encouraging. The Brain-to-Text software was consistently more accurate at classifying phonetics than a randomized model.
“This is just the beginning,” said Peter Brunner, a coauthor of the study. “The prospects of this are really endless.” The paper comments that traditional speech-recognition software has thousands of hours of acoustic data to model and refine the software, whereas Brain-to-Text has just two or three samples from seven people. With more trials and tweaking, the software can only get more accurate.
The technology can't easily be made commercially available because when it comes to brains, one size definitely does not fit all. The brain waves that transmit phonetic data are so sensitive that every brain will need to be assessed individually. Also, there is the issue of inserting a network of electrodes directly into the user's brain. The increase in quality of life therefore needs to be greater than the risk of brain damage or surgical complications.