Noninvasive decoding of typed sentences from human brain activity.

Brain2Qwerty achieves an average character error rate of 29% with magnetoencephalography, significantly outperforming electroencephalography in decoding sentences from brain activity.

• Why it matters: Decoding speech from brain signals noninvasively is crucial for developing safe brain-computer interfaces, especially for patients unable to communicate, reducing risks associated with invasive implants.
• What they did: Researchers trained a deep learning model on data from 35 healthy volunteers using electro- and magnetoencephalography while they typed memorized sentences on a QWERTY keyboard, achieving notable decoding accuracy.
• The result: The method enables accurate sentence decoding with as low as 18% error in top participants, paving the way for noninvasive communication devices for noncommunicating patients.