Latest thoughts and discoveries from the lab
Progress in thought-to-text with speech decoding BCIs is becoming increasingly difficult to compare. How can we standardise scores and measure progress?
Linguistics-inspired strategies for leveraging phonetic features in LibriBrain MEG data, covering feature taxonomies, diphthong handling, and feature-to-phoneme conversion pipelines.
Neuroscience-informed approaches to speech detection for the LibriBrain competition, including STG sensor analysis, spatial and temporal strategies, and architectural recommendations.
Exploring the Speech Detection task and the reference model architecture used in the LibriBrain competition, including insights into our 'Start Simple' approach and future research directions.
Exploring the motivation behind the 2025 PNPL Competition and how we're building towards non-invasive speech brain-computer interfaces through deep MEG datasets and collaborative research.