Optimimizing Noise-based Enhancement of Speech Recognition by Cochlear Implant Patients

Duke University, Durham NC

Investigators

Abstract

0085370 Collins It is well established that cochlear implants restore some level of functional hearing to most deaf individuals. However, speech recognition abilities vary widely across subjects and the mechanisms responsible for this variability are poorly understood. One factor that may impede speech recognition by cochlear implant subjects is that electrically stimulated nerves respond with a much higher level of synchrony than what is normally observed in acoustically stimulated nerves. Recently, some researchers have suggested that adding noise to a speech signal may decrease the synchronicity of the neural response observed under electrical stimulation, and thus, might restore a more normal response pattern. By generating more natural patterns, it may be possible to improve speech recognition for cochlear implant patients. Preliminary theoretical results from our lab indicate that more normal neural response patterns can be induced when small amounts of additive noise are added to periodic electrical signals. In addition, preliminary experimental results, again from our lab, indicate that speech recognition may also be improved using this approach. In the literature, the phenomenon whereby additive noise, when presented at an optimal level, improves signal transmission in nonlinear systems is known as stochastic resonance (SR). The goal of this research is to investigate and optimize a novel speech processing approach for cochlear implant patients based on the theory of stochastic resonance. To date, SR research has focused on the addition of noise to a weak signal within the context of a nonlinear systems. This research will instead consider the theoretical basis for driving a complex system to respond in a more chaotic fashion, and thus better mimic the responses observed in the normal auditory system. A series of theoretical and laboratory experiments has been designed to address the fundamental role of additive noise under electrical stimulation. A computational model of the neural response to electrical stimulation will be employed to develop the theoretical results, and results will be verified in psychophysical as well as neurophysiological experiments. Although optimizing the additive noise process for weak signals under "normal" acoustic neural stimulation has been considered in traditional SR research, this issue has not been addressed for neural systems subject to electrical stimulation. The specific questions that are proposed involve both generating a SR phenomenon and optimizing the phenomenon under electrical stimulation of the auditory system. This work will form an important theoretical basis for driving the auditory system to respond in a more natural, albeit chaotic fashion. Construction of the computer models will improve understanding of the neural response driven by electrical stimulation and assist in the design of new electrical stimulation paradigms that improve the representation of speech within the profoundly impaired auditory system. A collaboration with a neurophysiologist will ensure that the theoretical predictions are validated in a human model via psychophysical experiments and in neurophysiological data. In addition, the interdisciplinary scope of this work will provide a unique venue for the training of biomedical engineers.