Date of Award

Summer 2015

Document Type


Degree Name

Doctor of Philosophy (PhD)


Electrical & Computer Engineering

Committee Director

Dean J. Krusienski

Committee Member

Shuiwang Ji

Committee Member

Jiang Li

Committee Member

Dimitrie C. Popescu


Millions of people worldwide suffer from various neuromuscular disorders such as amyotrophic lateral sclerosis (ALS), brainstem stroke, muscular dystrophy, cerebral palsy, and others, which adversely affect the neural control of muscles or the muscles themselves. The patients who are the most severely affected lose all voluntary muscle control and are completely locked-in," i.e., they are unable to communicate with the outside world in any manner. In the direction of developing neuro-rehabilitation techniques for these patients, several studies have used brain signals related to mental imagery and attention in order to control an external device, a technology known as a brain-computer interface (BCI). Some recent studies have also attempted to decode various aspects of spoken language, imagined language, or perceived speech directly from brain signals. In order to extend research in this direction, this dissertation aims to characterize and decode various speech representations popularly used in speech recognition systems directly from brain activity, specifically the electrocorticogram (ECoG). The speech representations studied in this dissertation range from simple features such as the speech power and the fundamental frequency (pitch), to complex representations such as the linear prediction coding and mel frequency cepstral coefficients. These decoded speech representations may eventually be used to enhance existing speech recognition systems or to reconstruct intended or imagined speech directly from brain activity. This research will ultimately pave the way for an ECoG-based neural speech prosthesis, which will offer a more natural communication channel for individuals who have lost the ability to speak normally.