Acoustic signature analysis / pattern recognition is used for categorizing, classifying and characterizing sources, objects or sound/noise-events.
Humans do this intuitively; if we hear a train coming, we know immediately that this is a train, and not a dog nor a car – and we can even decide if the train is approaching or moving away. The sound processing capability of the human ear combined with the cognitive power of the brain; make us good pattern recognisers in many situations.
We try to mimic this capability by making systems that analyze acoustic signals, in order to make some judgements about the origin of the sound. The acoustic signal might be generated by the source itself. Alternatively the signal might be generated by humans, either by exciting the source in some way directly (knocking for instance), or by transmitting an acoustic signal that in turn results in an echo. The source could be a car, a human, a fracture, just simply anything that is able to generate or reflect audible or non-audible sound. By analyzing the signal with respect to time and frequency, it is possible to train a system that recognizes specific sounds or properties.
At SINTEF ICT we have addressed this problem in different applications:
Within the health field we work with a system whose aim is to support the diagnostics related to snoring. By analyzing the sound of the snoring, we aim to find the positions where the sound generation occurs.
For traffic monitoring we have developed systems that can measure whether the car has tyres with or without studs.
For Nordan (a window manufacturer) we have developed criteria and specifications for a sensor that detects break-ins.
Tone Berg, phone: +47 99735949, e-mail: Beskyttet adresse