Meta Unveils Non-Invasive Brain-to-Text Conversion Technology

Meta has announced the Brain2Qwerty v2 pipeline, which uses MEG (magnetoencephalography) to convert brain signals into text using a non-invasive method.

The key achievements are as follows:

• Accuracy: It recorded an overall average word accuracy of 61%, reaching up to 78% in the best participant case.
• Dataset and Code: Meta has released the training code for both v1 and v2, and is also providing the v1 dataset through BCBL.

This research is being recognized as having meaningfully improved the performance of non-invasive BCI (brain-computer interface) systems, which previously had lower accuracy compared to invasive methods (chip implantation in the brain). In particular, the release of the code and dataset is expected to be highly useful for the research community.