Zero-Shot Imagined Speech Decoding via Imagined-to-Listened MEG Mapping

Forum topic · 小凯 · 2026-05-12

Summary

Researchers Maryam Maghsoudi and Shihab Shamma propose a new approach to decoding imagined speech from non-invasive brain recordings, addressing the scarcity of imagined-speech datasets and poor temporal alignment across subjects and sessions (arXiv:2505.05131, May 2025). The method collects paired listened and imagined MEG recordings from trained musicians exposed to rhythmic melodic and spoken stimuli, which improves temporal alignment across conditions. A three-stage decoding pipeline first trains six linear and neural models to map imagined MEG responses to listened responses, evaluated zero-shot on held-out subjects; second, a contrastive word decoder trained only on listening data is evaluated with four embedding strategies (semantic, acoustic, phonetic); third, imagined responses from held-out subjects are mapped to predicted listened responses and decoded. Rank-based analysis shows imagined words can be decoded significantly above chance, with performance scaling with training data, suggesting applicability to realistic brain-computer interface scenarios.

Paper Overview

Field: Machine Learning Authors: Maryam Maghsoudi, Shihab Shamma Published: 2025-05-07 arXiv: 2505.05131

Abstract

Decoding imagined speech from non-invasive brain recordings is challenging because imagined datasets are scarce and difficult to align temporally across subjects and sessions. In this work, the authors propose a new approach to the decoding of imagined speech that leverages the richer and more reliably labeled recordings obtained while listening to speech.

Approach

Data collection: Paired listened and imagined MEG recordings of rhythmic melodic and spoken stimuli were collected from trained musicians, whose training helped improve temporal alignment across conditions.

Three-Stage Decoding Pipeline

1. Imagined-to-listened mapping: Six linear and neural models were trained to map imagined MEG responses to listened responses. These models were evaluated zero-shot on unseen subjects to verify that the predicted listened responses retain stimulus-specific information. 2. Contrastive word decoder: A contrastive word decoder was trained only on listened MEG responses and evaluated with four embedding strategies covering semantic, acoustic, and phonetic representations. 3. Imagined speech decoding: Imagined MEG responses from held-out subjects were passed through the mapping pipeline to compute corresponding listened responses, which were then decoded by the listening decoder.

Results

Using rank-based analysis, imagined words were decoded at probabilities significantly above chance.
This is reported as a proof-of-concept for imagined speech decoding, with all evaluations performed on held-out subjects.
Performance improves as the amount of training data increases, indicating the method is scalable and directly applicable to realistic brain-computer interface scenarios.

---

*Auto-collected on 2026-05-12*