# Brain-to-Text '25: Intracortical Speech Neuroprosthesis Recordings from BrainGate2 Participant T15 (Card et al., NEJM 2024, Dryad)

## Facts

- Provider: Nicholas S. Card, Maitreyee Wairagkar, Sergey D. Stavisky, David M. Brandman et al. (UC Davis Neuroprosthetics Lab); Hugging Face mirror
- License: CC0-1.0 (https://api.datacite.org/dois/10.5061/dryad.dncjsxm85)
- Size: 8 files
- Stored at: Hugging Face, https://huggingface.co/datasets/yw784/brain-to-text-25-competition
- Access: public, readable without an account
- DOI: https://doi.org/10.5061/dryad.dncjsxm85
- Category: Neuroscience
- Source: https://doi.org/10.5061/dryad.dncjsxm85
- Page: https://scigantic.com/archive/22595280-8e70-5edd-b494-6ae52ab7c697
- Keywords: Neuroscience, Hugging Face

## About

Data for An accurate and rapidly calibrating speech neuroprosthesis (Card, Wairagkar, Stavisky, Brandman and colleagues, UC Davis Neuroprosthetics Lab, New England Journal of Medicine 2024), the dataset of the Brain-to-Text 2025 decoding competition. Participant T15 of the BrainGate2 trial, implanted with intracortical microelectrode arrays (256 recording channels), attempted to speak cued sentences. t15_copyTask_neuralData.zip holds processed neural features for more than 11,000 Copy Task trials from 45 sessions over 20 months: threshold crossings at -4.5 RMS and spike band power for each channel (512 features) at 20 ms bins, z-scored, with sentence labels for train and validation trials; test labels are held out for the competition. t15_copyTask.pkl has decoded phonemes, words and RNN logits for 1,718 online trials and t15_personalUse.pkl summarizes 22,126 conversation sentences. Also included are a pretrained RNN baseline and n-gram language models. Use with the Neuroprosthetics-Lab nejm-brain-to-text code for phoneme decoding, brain-computer interface (BCI) speech decoding and neural sequence-to-text models. This is a mirror of Dryad DOI 10.5061/dryad.dncjsxm85, which is released under CC0 1.0. Cite Card et al., N Engl J Med, 2024.

## Load the data

```python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="yw784/brain-to-text-25-competition", filename="languageModel.tar.gz", repo_type="dataset")
print(path)
```

## Data structure

Example file: `languageModel.tar.gz`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/22595280-8e70-5edd-b494-6ae52ab7c697
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/22595280-8e70-5edd-b494-6ae52ab7c697
- Over MCP: `get_schema_card` with id `22595280-8e70-5edd-b494-6ae52ab7c697` at https://api.scigantic.com/mcp

## Related topics

- [Neuroscience datasets](https://scigantic.com/topics/neuroscience)
