Towards detection of perceptually similar sounds: investigating self-organizing maps
Christian Spevak, Richard Polfreman, Martin J. Loomes · ePrints Soton (University of Southampton) · 2001
This paper outlines a system for the detection of perceptually similar sounds ('sound spotting'), reports on a series of preliminary experiments and discusses their results. The sound spotting system pursues a frame-based approach and consists of three main stages, an auditory model, a self-organizing map and a pattern matching algorithm. The experiments described examine how different types of self-organizing maps classify a set of test sounds preprocessed by an auditory model and evaluate their performance by means of visualizations and quality measures. With these outcomes in mind we suggest directions for the further development of the sound spotting system.