Carried out by a group of informatics, neuroscience and medical researchers at Albany Medical Centre in the USA, the team managed to identify the brainwaves relating to speech by using electrocorticographic (ECoG) technology to monitor the frontal and temporal lobes of seven epileptic volunteers. This involves using needles to record signals directly from a person's neurons; it's an invasive procedure requiring an incision through the skull.
The participants then read aloud from a sample text while machine learning algorithms pulled out the most likely word sequence from the signals recorded by the EcOG. Existing speech-to-text tools then transcribed the continuously spoken speech directly from the brain activity.
Error rates were as low as 25 percent during the study, which means the potential for the system is pretty vast. The findings could offer locked-in and mute patients a valuable communication method but it also means humans could one day communicate directly with a computer without needing any intermediary equipment.
However, brain-to-text tech seems unlikely to make it into a consumer product unless the researchers can develop a way to access brainwaves without the need to expose the brain.