Contactless optical decoding of cortical language responses via region-transferable speckle dynamics

Abstract Remote decoding of human brain activity without surgery or physical contact remains a major challenge for brain-computer interfaces. Using laser speckle dynamics, we show that cortical responses during language processing can be decoded contactlessly. Across two independent experimental paradigms involving 17 healthy participants, 23 sessions, and 462 recordings, a classifier distinguished responses to intelligible versus incomprehensible speech in Wernicke’s area with a mean accuracy of 95.69% and an AUC of 0.98, using 40-ms input segments and requiring less than 1 minute of calibration (20s of labelled data per category). Importantly, the decoding representations were derived from a self-supervised model trained on a different task, inner speech, and from a different cortical region, Broca’s area, demonstrating transfer across both language tasks and anatomically distinct cortical areas. These findings support the feasibility of contactless optical decoding of cortical language responses and indicate that speckle-derived representations of cortical hemodynamics capture response structure that generalizes across regions, substantially reducing calibration requirements and providing a foundation for practical contactless cortical decoding.