A channel-weighting method for speech recognition using wavelet decompositions

A decomposition of signal into a set of frequency channels of equal bandwidth on a logarithmic scale, i.e., an analysis of the signal using constant Q filters, using wavelet and multiresolution analysis is used in this paper to derive cepstrum features of different spatial frequency bands. Based on the decompositions, each channel is modeled as a Bayesian subnetwork and each subnetwork is weighted by a weighting algorithm. The distortions for speech recognition between a reference model and the input vectors are then computed by summing the weighted scores of all decomposed channels. The experimental results show that the recognition rate of this method is superior to those non-weighting methods.