Data for: An accurate and rapidly calibrating speech neuroprosthesis
Data files
Oct 09, 2024 version files 58.92 MB
-
README.md
1.90 KB
-
t15_copyTask.pkl
57.79 MB
-
t15_personalUse.pkl
1.13 MB
Jul 03, 2025 version files 11.59 GB
-
README.md
3.18 KB
-
t15_copyTask_neuralData.zip
11.05 GB
-
t15_copyTask.pkl
57.79 MB
-
t15_personalUse.pkl
1.13 MB
-
t15_pretrained_rnn_baseline.zip
484.94 MB
Abstract
Brain-computer interfaces can enable communication for people with paralysis by transforming cortical activity associated with attempted speech into text on a computer screen. Communication with brain-computer interfaces has been restricted by extensive training requirements and limited accuracy. A 45-year-old man with amyotrophic lateral sclerosis (ALS) with tetraparesis and severe dysarthria underwent surgical implantation of four microelectrode arrays into his left ventral precentral gyrus 5 years after the onset of the illness; these arrays recorded neural activity from 256 intracortical electrodes. We report the results of decoding his cortical neural activity as he attempted to speak in both prompted and unstructured conversational contexts. Decoded words were displayed on a screen and then vocalized with the use of text-to-speech software designed to sound like his pre-ALS voice. On the first day of use (25 days after surgery), the neuroprosthesis achieved 99.6% accuracy with a 50-word vocabulary. Calibration of the neuroprosthesis required 30 minutes of cortical recordings while the participant attempted to speak, followed by subsequent processing. On the second day, after 1.4 additional hours of system training, the neuroprosthesis achieved 90.2% accuracy using a 125,000-word vocabulary. With further training data, the neuroprosthesis sustained 97.5% accuracy over a period of 8.4 months after surgical implantation, and the participant used it to communicate in self-paced conversations at a rate of approximately 32 words per minute for more than 248 cumulative hours. In a person with ALS and severe dysarthria, an intracortical speech neuroprosthesis reached a level of performance suitable to restore conversational communication after brief training.
Dryad DOI: https://doi.org/10.5061/dryad.dncjsxm85
Nicholas S. Card, Maitreyee Wairagkar, Carrina Iacobacci, Xianda Hou, Tyler Singer-Clark, Francis R. Willett, Erin M. Kunz, Chaofei Fan, Maryam Vahdati Nia, Darrel R. Deo, Aparna Srinivasan, Eun Young Choi, Matthew F. Glasser, Leigh R. Hochberg, Jaimie M. Henderson, Kiarash Shahlaie, Sergey D. Stavisky, and David M. Brandman.
- “*” denotes co-senior authors
Overview
This repository contains the data necessary to reproduce the results of the paper “An Accurate and Rapidly Calibrating Speech Neuroprosthesis” by Card et al. (2024), N Eng J Med.
The code is written primarily in Python and is hosted on GitHub.
The data can be downloaded from this Dryad repository. Please download this data and place it in the data
directory of the GitHub code. All included data has been anonymized and does not include any identifiable information.
Files:
t15_copyTask.pkl
- Data from Copy Task trials during evaluation blocks (1,718 total trials) are necessary for reproducing the online decoding performance plots (Figure 2).
- Copy Task data includes, for each trial: cue sentence, decoded phonemes and words, trial duration, and RNN-predicted logits.
t15_personalUse.pkl
- Data from Conversation Mode (22,126 total sentences) is necessary for reproducing Figure 4.
- Conversation Mode data includes, for each trial, the number of decoded words, the sentence duration, and the participant’s rating of how correct the decoded sentence was.
- Specific decoded sentences from Conversation Mode are not included to protect the participant’s privacy.
t15_copyTask_neuralData.zip
- Processed neural data and sentence labels during 11,000+ Copy Task trials, from 45 data collection sessions spanning 20 months.
- Processed neural data is threshold crossings (-4.5 RMS threshold) and spike band power for each of the 256 recording channels (512 features total) at 20 ms resolution. Data are normalized (z-scored) based on the preceding 20 trials.
- Trials are split into “train”, “val”, and “test” trials. “test” trials do not include the ground truth sentence label as they will be used for the Brain-to-Text 2025 challenge.
- Refer to the GitHub repo for specific instructions on how to load and use this data, and refer to the Brain-to-Text ‘25 competition page for details on how to submit your own results.
t15_pretrained_rnn_baseline.zip
- An RNN model that has been pretrained on the T15 copyTask neural dataset.
- Refer to the GitHub repo for specific instructions on how to load and do inference with this model.
Change log
3 July 2025: Added two zip files. One is fully de-identified processed neural data. The other is the weights for a pretrained RNN model. No personal or identifiable information is included in any of the data included here.
- t15_copyTask_neuralData.zip
- t15_pretrained_rnn_baseline.zip