Brain-to-Text ’25: Decoding Speech from Neural Activity

Brain-to-Text ’25 was featured in the 2025 Machine Learning Marathon (MLM25). This Kaggle competition challenges participants to decode intracortical neural activity during attempted speech into text – aiming to restore communication for people with paralysis.

Challenge design

• Task: Decode neural recordings from speech-related brain regions into the words a participant is attempting to say.
• Domain: Brain-computer interfaces and neural speech decoding.
• Data: A new intracortical speech neuroscience dataset provided for the competition.
• Methods: The 2024 edition’s top approaches used RNN ensembles merged with fine-tuned large language models. The baseline achieved 9.7% word error rate; the top entrant reached 5.8%.

Links

• Kaggle challenge: Brain-to-Text ’25

Related resources

• Workshop: Intro to Deep Learning with PyTorch: Learn RNNs and sequence modeling fundamentals in PyTorch — key building blocks for neural speech decoding.
• Workshop: Intro to Natural Language Processing (NLP): Brush up on text processing and language model basics relevant to the text decoding target.
• Library: Hugging Face: Top solutions fine-tuned LLMs from Hugging Face — learn how to discover and use open-source models.

Questions

If you have any lingering questions about this project, please feel free to post to the Nexus Q&A on GitHub. We will improve materials on this website as additional questions come in.