SpeechBrain-MOABB: An open-source Python library for benchmarking deep neural networks applied to EEG signals.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Deep learning significantly advances electroencephalography (EEG) decoding.
• Existing open-source tools lack comprehensive neural network support and standardized protocols, hindering reproducibility.
• Current libraries like MOABB and braindecode have limitations in hyperparameter search and are sensitive to random initialization.

Purpose of the Study:

• Introduce SpeechBrain-MOABB, a novel open-source toolkit for deep learning-based EEG decoding pipelines.
• Address the limitations of existing tools by providing standardized experimental protocols and robust evaluation methods.
• Facilitate the development of replicable and trustworthy EEG decoding pipelines.

Main Methods:

• Developed SpeechBrain-MOABB, an open-source toolkit integrating deep learning for EEG decoding.
• Implemented a complete experimental protocol standardizing hyperparameter search and model evaluation.
• Incorporated multi-step hyperparameter search and multi-seed training/evaluation for robust performance estimation.

Main Results:

• SpeechBrain-MOABB demonstrated superior performance compared to MOABB and braindecode.
• Achieved average accuracy improvements of 14.9% over MOABB and 25.2% over braindecode.
• Ensured performance estimates are robust to random initialization variability.

Conclusions:

• SpeechBrain-MOABB enables the creation of comprehensive, reproducible, and trustworthy EEG decoding pipelines.
• The toolkit facilitates the use of deep learning for EEG decoding by neuroscientists.
• Standardized protocols and robust evaluation methods improve the reliability of EEG decoding research.