Acoustic Source Separation and Localization

The Phase I research project will develop a bio-inspired acoustic source separation, localization and tracking algorithm called PreservEar. Separation of speech signals from an auditory scene, often referred to as the cocktail party problem, has been studied for decades. Current source separation solutions suffer from six major problems: (1) source localization and tracking is not very robust; (2) intensive signal processing needs; (3) many separation methods require at least as many sensors as sources; (4) many require the number of signals within the recorded mixtures be known in advance; (5) do not perform real-time under real life conditions well; (6) large power consumption needs. This limits their deployment on small embedded devices. The intellectual merits of PreservEar are that it addresses these issues by providing a solution that fuses novel computational auditory scene analysis algorithm with robust localization-tracking developed with singular focus of having a single package solution that is effective, compact, light, practical and easy to use. System developed will have low memory, low computational and low power need; use only two microphones, and run in real-time on off-the shelf devices like Bluetooth headsets. Continued objective is to also use device for hearing loss restoration and hearing loss prevention.