Brain-to-text decoding model achieves 39% word error rate using MEG

Researchers have developed Brain2Qwerty v2, a novel model capable of decoding natural sentences from non-invasive brain recordings using magnetoencephalography (MEG). The model achieved an average word error rate of 39% by analyzing 22,000 sentences typed by nine subjects over 10 hours. This advancement, driven by deep learning and large language models, demonstrates that non-invasive brain-to-text decoding is approaching the accuracy levels previously only achievable with surgical implants, with performance expected to improve further with increased data volume. AI

IMPACT Advances non-invasive brain-computer interfaces, potentially restoring communication for individuals with speech and movement loss.

RANK_REASON Academic paper detailing a new model and benchmark results. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• Brain2Qwerty v2
• CatalyzeX
• DagsHub
• deep learning
• Gotit.pub
• Hugging Face
• large language models
• magnetoencephalography
• ScienceCast

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →