Robust Sound Recognition: A Neuromorphic Approach

Humans perform remarkably well at sound classification that is used as cues to support high-level cognitive functions. Inspired by the anatomical structure of human cochlea and auditory attention mechanism, we present a novel neuromorphic sound recognition system that integrates an event-driven auditory front-end and a biologically plausible spiking neural network classifier (SNN) for robust sound and speech recognition. Due to its event-driven nature, the SNN classifier is several orders of magnitude more energy efficient than deep learning classifier, therefore, it is suitable for many applications in wearable devices.