We’re excited to announce that Phase 1 of the LibriBrain Competition is now live and accepting submissions.
What’s Available Now#
Starting today, participants can submit their solutions for the Speech Detection task in both:
- Standard Track: Using our own “LibriBrain” dataset
- Extended Track: Leverage any and all additional data sources
Key Details#
- Submission Deadline: July 31st, 2025
- Evaluation Metric: F1 Macro Score - the unweighted mean of F1 scores across both classes: $\text{F1}_{\text{macro}} = \frac{\text{F1}_{\text{speech}} + \text{F1}_{\text{non-speech}}}{2}$
- Leaderboard: Competition updates available on our leaderboard page
- Starter Kit: Get up and running in under an hour with our tutorials
Getting Started#
- Install the PNPL package:
pip install pnpl - Download the dataset automatically with the PNPL package: Follow our participation guide
- Train your first mode using our tutorials: Use our provided notebooks - or build something on your own
- Submit your results: Upload via our submission portal on Eval.ai
Dataset Features#
This competition features the deepest MEG dataset to date - 5–50× deeper than most existing datasets in the field. This unprecedented depth enables researchers to explore neural patterns of speech processing at a level of detail not previously possible. This year, we’ve taken care of all preprocessing and provide simple Pytorch dataloaders.
Questions?#
Join our Discord community for real-time support and discussions with fellow participants and organizers.
Good luck to all participants. We look forward to seeing the innovative approaches you’ll develop.