MEGState: Phoneme Decoding from Magnetoencephalography Signals

TL;DR

This paper introduces MEGState, a new neural network architecture that improves phoneme decoding from MEG signals, advancing non-invasive speech brain-computer interfaces by capturing detailed cortical responses despite challenges like low signal-to-noise ratio.

Contribution

MEGState is a novel architecture specifically designed for phoneme decoding from MEG signals, demonstrating superior performance over baseline models on the LibriBrain dataset.

Findings

MEGState outperforms baseline models on multiple metrics.

MEGState effectively captures fine-grained cortical responses.

Results support MEG-based phoneme decoding as a scalable BCI approach.

Abstract

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.