Sound-based classification of objects using a robust fingerprinting approach

Abstract

Tangible Acoustic Interfaces (TAIs) are interaction devices that are able to localize the interaction point on a solid surface. Their advantages over traditional interaction devices (touch screens, touch pads, etc.) is in the fact that actual acoustic (vibrational) signals are acquired by contact sensors. This opens the way to interaction classification and recognition. With this application in mind, this paper approaches the problem of classifying the interaction object from the acquired sounds. We focus on continuous interaction noise, which we classify through a “fingerprinting” approach: features are extracted from the acquired signals and matched against pre-computed features. More sophisticated solutions can be devised for the problem of the classification of noiselike sounds but our approach has the advantage of being computationally simple and can be profitably implemented in real-time.