A high-performance speech neuroprosthesis

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Brain-computer interfaces (BCIs) offer a potential communication pathway for individuals with severe paralysis.
• Existing speech BCIs have limitations in accuracy and vocabulary size, hindering effective communication.
• Amyotrophic lateral sclerosis (ALS) progressively impairs speech, necessitating advanced assistive technologies.

Purpose of the Study:

• To develop and evaluate a high-performance speech-to-text BCI using intracortical microelectrode recordings.
• To assess the BCI's accuracy and speed in decoding attempted speech from a participant with ALS.
• To investigate neural coding principles relevant to speech BCI development.

Main Methods:

• Utilized intracortical microelectrode arrays to record spiking neural activity during attempted speech.
• Developed decoding algorithms to translate neural signals into text.
• Evaluated performance using word error rates across different vocabulary sizes and measured decoding speed.

Main Results:

• Achieved a 9.1% word error rate on a 50-word vocabulary, a 2.7-fold improvement over prior state-of-the-art.
• Demonstrated the first successful large-vocabulary (125,000 words) decoding with a 23.8% word error rate.
• Reached a decoding speed of 62 words per minute, 3.4 times faster than previous records.

Conclusions:

• High-resolution neural recordings enable accurate and rapid speech decoding via BCIs.
• The neural code for speech contains features that facilitate robust decoding even after prolonged paralysis.
• This BCI technology presents a viable approach to restoring communication for non-speaking individuals.