The long-term objective of this project is to use digital signal processing (DSP) techniques to take a detailed nonlinear auditory-nerve (AN) model and to implement a Real-time, Nonlinear AN Modeling Engine for humans. We envision this to be a modular processing core that may be customized for different applications by using different hardware and software wrappers around it. Further, the modular approach will allow easy updating of the DSP model as more accurate AN models become available in the future. Such a real-time model of the human ear has commercial and scientific applications as well as societal benefits: Better 'fitting' of hearing-aids that are highly customized for an individual - this is a $4B market Better speech recognition systems - biometrics, cell phones, interactive automotive systems, military applications, ergonomical human-machine interface - this market is poised to explode to $5B Facilitate physiological and psychophysical research by providing results in real-time, versus days, for accurate models of the healthy ear and ears affected by different impairments. The short-term focus is on the real-time human model (normal and impaired) for hearing-aid fitting. The AN model developed in the Carney Lab includes the compressive nonlinearity in the inner ear, and the appropriate frequency glides for low-, mid-, and high- characteristic frequency AN fibers - these properties make the model more accurate in predicting the behavior of the normal as well as the impaired ear. However, it does not run in real-time. The principals of Discrete Laboratories have been working with cutting edge audio technology and have developed numerous audio products using DSPs - all running in real-time. In this project, Dr. Carney's physiological modeling expertise will be merged with the experience of the digital audio specialists at Discrete Laboratories. In Phase I we will use our background in optimizing high-level code into real-time DSP assembly code. We will start by analyzing the model in terms of efficient DSP implementation. The three functional blocks of the model are the (1) Signal Path, (2) Control Path, and (3) Inner Hair Cell and Synapse. We will lay down specifications for each functional block; simulate and verify our approach using analytical tools, and write DSP assembly code according to the specifications. The final result will be benchmarks for the program, data and computational resources required for each functional block of the model. This feasibility study lays down a solid foundation for prototyping this real-time model in Phase II.
