Events Calendar

Title: "Extension and Application of the Cortically Inspired Spatial Processing Algorithm" Committee: Prof. Kamal Sen, BME (Advisor, Chair) Prof. Steve Colburn, BME Prof. Oded Ghitza, BME Prof. Matt Goupell, University of Maryland - Hearing and Speech Sciences Abstract: The inability to selectively attend to a sound stream in midst of many competing sounds plagues hearing impaired listeners, making it difficult to communicate in social situations. Machine hearing systems have attempted to solve this problem with an engineering approach (e.g. using statistical methods and machine learning) but still performs poorly in acoustically complex settings. In contrast, normal hearing listeners can easily solve this problem, suggesting that the solution exists somewhere in the brain. Recently, we developed a cortically-inspired algorithm (CA) to tackle this problem. The CA transforms sounds into an abstract neural space using directionally sensitive model neurons, then processes the resultant spikes via a lateral inhibition network, effectively suppressing sounds belonging to unattended directions. One limitation of the algorithm is the low quality of reconstructed sounds. In this proposal, I plan to address this limitation by encoding sounds with a population of directionally sensitive neurons, then decoding the processed spikes with a bank of linear filters. This work also aims to apply the decoder of the CA to search for optimal parameters for cochlear implants. Finally, this work aims to apply the framework of CA to automatic speech recognition, which in theory is more robust to noise and competing sound streams when performed in the neural space. If successful, this work will contribute to the field of machine hearing as well as treatments for the hearing impaired population, enabling better hearing in natural settings.