Researchers Discover Scaling Law in Neural Data, Achieve 50% Accuracy in Speech Decoding Using 175 Hours of EEG

Jul 11, 2024, 05:00 AM

Researchers have discovered a scaling law in neural data that significantly improves the accuracy of speech decoding using non-invasive EEG recordings. The study, conducted by Motoshige Sato, K Tomeoka, I Horiguchi, K Arulkumaran, R Kanai, and S Sasai from Araya Inc, involved collecting 175 hours of EEG data from a single participant. The findings demonstrate that speech decoding accuracy can reach approximately 50% in a 512-phrase classification task. This breakthrough is particularly notable given the day-to-day variability in neural data and the challenges associated with non-invasive methods.