LibriBrain Experiments

This repository provides experimental setups and code accompanying the figures presented in the publication:

LibriBrain: Over 50 Hours of Within-Subject MEG to Improve Speech Decoding Methods at Scale

---

Installation

You can install the package in editable/development mode so that any local changes are immediately reflected:

pip install -e .

Experiment Configuration

Configuration files for the phoneme decoding experiments detailed in our paper can be found in:

libribrain_experiments/phoneme/configs

Important Configuration Notes:

Before running the project, make sure to update the configuration files with the correct local paths:

• data_path: Specify the paths for your training, validation, and testing datasets.
• output_path: Set this to the directory where output results (e.g., logs, predictions) will be saved.
• checkpoint_path: Define the location where model checkpoints should be stored.

---

Running an Experiment

Use the following command format to execute an experiment:

python libribrain_experiments/hpo.py \
    --config=libribrain_experiments/configs/phoneme/<config-name>/base-config.yaml \
    --search-space=libribrain_experiments/configs/phoneme/<config-name>/search-space.yaml \
    --run-name=<run-name> \
    --run-index=<run-id>

Replace <config-name>, <run-name>, and <run-id> with your own values—<config-name> selects which experiment folder under libribrain_experiments/phoneme/configs, <run-name> is the Weights & Biases run name, and <run-id> is the hyperparameter/seed configuration index.

---