Tasks#
The LibriBrain Competition features two core tasks in decoding language from MEG recordings of brain activity:
- Speech Detection
Train a model to distinguish speech vs. silence based on brain activity measured by MEG during a listening session.
- Phoneme Classification
Build a classifier that maps MEG data to the specific phonemes being heard.
Tracks#
We offer two tracks per task to balance resource constraints with open exploration:
| Track | Training Data Allowed | Motivation |
|---|---|---|
| Standard | LibriBrain only | Level playing field — innovate on methods/efficiency |
| Extended | Any data | Embrace scale — see how far teams with resources can go |
Regardless of training data, all tracks will be evaluated on a shared competition holdout set in order to best compare approaches and measure progress.
You’re welcome to enter any and all tracks. To encourage participation, prize money will be awarded to each team only once: if you place in multiple tasks/tracks, then the lower prize you would win rolls down to the next eligible team.
We believe these tasks will spark new breakthroughs in language decoding from brain activity. Ready to get started? Check out the participation guide!