Modeling Auditory Pattern Recognition and Learning with Gradient Frequency Neural Oscillator Networks

This Small Business Technology Transfer research project addresses the perception and learning of complex sound patterns within complex auditory scenes. The objective is to model auditory signal processing, pattern recognition and learning in the human auditory system. Our novel approach simulates the nonlinear signal processing that has been observed in auditory physiology. By mimicking functionally important nonlinearities, this technology has the potential to simulate many human perceptual capabilities. Our specific goal is to enable the recognition and learning of sound patterns in complex acoustic environments in real time. During Phase I, we plan to 1) simulate complex pattern recognition amidst background noise, 2) simulate complex pattern learning in the presence of noise and multiple targets, and 3) investigate hardware acceleration using GPU and FPGA methods to provide significant speedup by the end of Phase I. A detailed report will be delivered along with a plan for achieving real-time pattern recognition and learning amidst background noise by the end of Phase II. The success of the model will inform fundamental scientific research by further elucidating the role of nonlinear processing in the auditory system. BENEFIT: A technology that can successfully recognize complex sound patterns in natural environments would have significant implications for almost every application, military and civilian, that processes sound. Existing systems would improve, deployment in new environments would be enabled, and new applications would become possible. Military applications would include audio surveillance, biometric security, and sonar. Civilian applications would include hearing technologies, speech technologies, and music applications. Hearing impairment is expected to affect 700 million people worldwide by 2015 and such technology would greatly benefit the users of hearing devices. Improvements to speech technologies would include better recognition rates and noise tolerance for speech recognition systems, and improved cell phone clarity in a range of environments. Music applications would be able to automatically segregate polyphonic recordings, yielding new and improved tools for several million musicians, for hundreds of millions of music consumers who enjoy music recommendation systems, and for copyright identification and management systems.