A Novel Brain–Machine Interface to Restore Speech

Project: Research project

Project Details

Description

People who become completely paralyzed, or “locked-in,” from diseases such as stroke, ALS or cere-bral palsy are unable to speak or communicate. For these patients, the ability to restore communication is paramount. Brain machine interfaces (BMIs) can restore basic communication by allowing patients to slowly spell out individual letters by decoding non-speech related evoked potentials from the scalp in re-sponse to a stimulus. This paradigm requires intense concentration and a great deal of time (several minutes) to communicate a simple sentence. The resultant communication delays are a critical impedi-ment – consider the frustration engendered by even just a few seconds of delay in conversations.
A far more user-friendly, and far higher throughput, paradigm would be to decode the patient’s in-tended speech directly from the speech motor cortical signals. The patient would attempt to speak and the BMI would directly decode entire words with minimal delays. Words are composed of groups of sounds, or phonemes. Signals recorded from the surface of the motor cortex (using electrocorticogra-phy, or ECoG) contain information about phonemes based on the movements of articulator muscles (e.g., tongue vs. lips [1]). However, all prior attempts to decode phonemes using ECoG have been lim-ited to a small set of phonemes. Further previous attempts all used algorithms that required prior knowledge of the timing of the spoken words, which is not practical for use by paralyzed patients. We recently published evidence that we can decode individual phonemes using ECoG signals {Mugler 2014}. We hypothesize that applying more sophisticated decoding algorithms developed for speech recognition will enable us to decode speech continuously without having to determine the exact onset times of each phoneme. We hypothesize that using these algorithms, as well as prior linguistic knowledge of the interdependencies among phonemes within words, will markedly improve our ability to decode words from ECoG.
StatusFinished
Effective start/end date1/1/1512/31/16

Funding

  • Northwestern Memorial Hospital (Agmt Signed 3/9/15)

Fingerprint

Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.