Scientists have recently created an artificial intelligence that is capable of turning the brain activity into text output. The system is working on neural patterns currently that get detected once a person speaks aloud. Experts have stated that it could later help patients who are suffering from locked-in syndrome and others who cannot speak or type for communicating.

The research’s co-author from the University of California, Dr. Joseph Makin said that the system has not reached that point yet, but it could stand as the basis for speech prosthesis. Makin and his colleagues have been noting down the about the process of the development of the system by recruiting about four participants who had their brains implanted with electrode arrays to monitor epileptic seizures.

Citing reliable sources, the participants had been asked to read about 50 sentences aloud, multiple times. This includes a sentence like ‘Tina Turner is a pop singer’. While the candidates spoke, the team had been tracking their neural activity. The data was further fed in a machine learning algorithm which happens to be an artificial intelligence system capable of converting brain activity data of every spoken line into various strings of numbers.

The system also compared all the sounds that had been predicted form the smaller parts of the brain activity data with a genuine recorded audio, to ensure the numbers were only related to an aspect of the speech. The string of numbers was further fed in the second part of the system that later converts itself into a word sequence.

After numerous trials, the team recognized that the accuracy level of the new system was much higher than the approaches before. While it varied in different people, one candidate needed only 3% of correcting on average for each sentence, which is higher than the rate of word error of 5% for all the human transcribers. The team emphasized that the algorithm handles only a small number of sentences.

Source Credit- https://www.theguardian.com/science/2020/mar/30/scientists-develop-ai-that-can-turn-brain-activity-into-text