MEGConformer: Conformer-Based MEG Decoder for Robust Speech and Phoneme Classification

TL;DR

This paper introduces a Conformer-based MEG decoder that effectively classifies speech and phonemes from MEG signals, achieving top leaderboard performance and demonstrating robustness with novel augmentation and normalization techniques.

Contribution

It presents a compact Conformer architecture tailored for MEG signals, incorporating MEG-specific augmentation, normalization, and class handling methods, setting new benchmarks on the LibriBrain 2025 PNPL benchmark.

Findings

Achieved 88.9% Speech Detection accuracy.

Achieved 65.8% Phoneme Classification F1-macro.

Won the Phoneme Classification Standard track.

Abstract

Decoding speech-related information from non-invasive MEG is a key step toward scalable brain-computer interfaces. We present compact Conformer-based decoders on the LibriBrain 2025 PNPL benchmark for two core tasks: Speech Detection and Phoneme Classification. Our approach adapts a compact Conformer to raw 306-channel MEG signals, with a lightweight convolutional projection layer and task-specific heads. For Speech Detection, a MEG-oriented SpecAugment provided a first exploration of MEG-specific augmentation. For Phoneme Classification, we used inverse-square-root class weighting and a dynamic grouping loader to handle 100-sample averaged examples. In addition, a simple instance-level normalization proved critical to mitigate distribution shifts on the holdout split. Using the official Standard track splits and F1-macro for model selection, our best systems achieved 88.9% (Speech) and 65.8% (Phoneme) on the leaderboard, winning the Phoneme Classification Standard track. For further implementation details, the technical documentation, source code, and checkpoints are available at https://github.com/neural2speech/libribrain-experiments.