Wavelet band features

Wavelet band features are used to classify transient acoustic signals produced by impulses. Wavelet band features of one-dimensional sampled signals are formed by concatenating moments of squared moduli of each subband resulting from recursive decomposition of the signal using a low-pass conjugate quadrature filter. Joint time-frequency constant- Q (single octave) localization of the subband decomposition is shown to be a consequence of the vanishing moments and smoothness properties of wavelets associated with the filter. The moments of squared moduli constitute robust shift-invariant measures of the interband and intraband energy distribution that provide low-dimensional feature vectors for classifying signals. Classification results based on moment features are compared to results based on the short-time Fourier transform (STFT)