MEGState: Phoneme Decoding from Magnetoencephalography Signals
This work addresses the problem of non-invasive neural speech decoding for brain-computer interfaces, representing an incremental improvement in a domain-specific area.
The paper tackled the challenge of decoding phonemes from magnetoencephalography (MEG) signals, which have low signal-to-noise ratio and high temporal dimensionality, and introduced MEGState, a novel architecture that consistently surpassed baseline models across multiple evaluation metrics on the LibriBrain dataset.
Decoding linguistically meaningful representations from non-invasive neural recordings remains a central challenge in neural speech decoding. Among available neuroimaging modalities, magnetoencephalography (MEG) provides a safe and repeatable means of mapping speech-related cortical dynamics, yet its low signal-to-noise ratio and high temporal dimensionality continue to hinder robust decoding. In this work, we introduce MEGState, a novel architecture for phoneme decoding from MEG signals that captures fine-grained cortical responses evoked by auditory stimuli. Extensive experiments on the LibriBrain dataset demonstrate that MEGState consistently surpasses baseline model across multiple evaluation metrics. These findings highlight the potential of MEG-based phoneme decoding as a scalable pathway toward non-invasive brain-computer interfaces for speech.