LibriBrain100 Competition Advances Non-Invasive Speech Decoding via MEG
September 2, 2026
The 2026 PNPL competition focuses on cross-subject generalization for brain-computer interfaces using the LibriBrain dataset. While previous winning submissions achieved 95.6% F1-macro on phoneme classification within single subjects, the current challenge targets decoding accuracy across new users with minimal data.
HOW THIS AFFECTS YOU
●
researcherThis shifts the focus from within-subject performance to the harder problem of cross-user generalization.
●
healthSuccess in this area is a critical step toward practical, non-invasive neural speech interfaces.