Wavelet preprocessing of acoustic signals

The authors describe results on using the wavelet transform to preprocess acoustic broadband signals in a system that discriminates between different classes of acoustic bursts. This is motivated by the similarity between the proportional bandwidth filters provided by the wavelet transform and those found in biological hearing systems. The experiment involves comparing statistical pattern classifier effects of wavelet and FFT (fast Fourier transform) preprocessed acoustic signals. The data used are from the DARPA Phase I database, which consists of artificially generated signals with real ocean background. The result show that the wavelet transform did provide improved performance when classifying in a frame-by-frame basis. The DARPA Phase 1 database is well matched to proportional bandwidth filtering, i.e., signal classes that contain high frequencies compensate for the poor match of the wavelet transform for long-duration (high-frequency) signals