An electrophysiological investigation of human motor cortex and its application to speech restoration

This dissertation provides a means for restoration of speech production for humans with profound paralysis through microelectrode recording and prediction of intended speech from correlated neural activity. The basis for this research comes from a human subject with locked-in syndrome implanted with a neurotrophic electrode (Kennedy and Bakay, 1998). Most locked-in patients have complete loss of voluntary movement, but have fully intact sensory and cognitive function. Therefore, it is possible, in this dissertation, to conduct scientific experimentation, as the subject is capable of perceiving and understanding instructions and provides debriefing information through limited eye movements.;The design and implementation of the neural prosthesis had three main parts. The first involved isolation of single units from an extracellular multiunit recording. A total of 56 neural units were identified from recordings of human speech motor cortex using the neurotrophic electrode during a range of neurological and cognitive states.;The second component of the neural prosthesis design involved the analysis of task-specific neural units. The methodology employed relates speech production to previous studies of motor execution and utilizes recent algorithms for prediction of motor behavior from extracellularly recorded action potentials. The preferred states (e.g. preferred direction and position) of the neural units were found to have broad, though not uniform, coverage over the range of movements of the vocal tract necessary for speech production.;The final element in the implementation of the neural prosthesis was the application of results from the first two stages to the creation of a real-time device for decoding speech from neural activity and subsequent computer speech synthesis. Single unit activity was decoded into continuous-state acoustic measures of speech (i.e. formant frequencies) in a real-time environment and synthesized. The subject participated in an acoustic center-out task (c.f. Georgopoulos et al., 1986) consisting of three different vowel-vowel utterances. Decoding performance steadily improved over a two month trial period. The subject continues to participate in the study utilizing the neural prosthesis for speech production.