Frank Klassner, Victor Lesser, Hamid Nawab
When dealing with signals from complex environments, where multiple time-dependent signal signatures can interfere with each other in stochastically unpredictable ways, traditional perceptual systems tend to fall back on a strategy of always performing finely-detailed, costly analysis of the signal with a comprehensive front end set of signal processing algorithms (SPAS), whether or not the current scenario requires the extra detail. Approximate SPAS (ASPAs) - algorithms whose processing time can be limited in order to trade off precision in their outputs for reduced execution time - can play a role in producing adaptive, less-costly front ends, but their outputs tend to require context-dependent analysis for use as evidence in interpretation. This paper examines the IPUS (Integrated Processing and Understanding of Signals) architecture’s ability to serve as a support framework for applying ASPAs in interpretation problems. Specifically, our work shows that it is feasible to include an approximate version of the Short-Time Fourier Transform in an IPUS-based sound-understanding testbed.