Classifying Simulated Event Cochlea Data Using an Artificial Neural Network

Neuromorphic engineering, a specialized engineering field inspired by the human brain, may remedy the fear of stagnation against Moore’s Law. This law states that the number of transistors on a chip doubles every two years. Following in the footsteps of R. F. Lyon and C. A. Mead, an electronic cochlea has been selected to explore audio processing similarly to a human ear. While we do not have an electronic hardware cochlea, we have created a simple software model, inspired by cochlear models, in the programming and simulation language MATLAB. To do this, we first explored preexisting base code options and described the process of creating wholly original code. Initial tests of audio frequency analysis were done via spectrograms with a brief consideration of spike conversion for analysis. Ultimately, the project was completed by processing input signals through a cascade of simulated bandpass filters. The magnitude sampled from each filtered frequency range was plotted to represent different musical notes’ ranges onto a 4x4 pixel image grid. These images were then prepped to be used in a neural network algorithm that mathematically uses them to progressively recognize musical chords.