Speech coding by wavelet representation of residual signal

We propose and analyze a speech coder which utilizes a wavelet representation of the residual signal. We first present a brief description of a few wavelet-based speech coders reported in the literature. We then introduce the discrete wavelet transform and discuss its potential in data compression. Next, we briefly describe some speech coding tools including adaptive spectral flattening filter, line spectral pairs, voiced/unvoiced classifier, perceptual weighting filter and adaptive postfilter. Subsequently, we provide a detailed description of the proposed wavelet-based coder, which is capable of producing high quality speech at 1O kbits/s. Some experimental results are also presented. Finally, we discuss the strengths and weaknesses of the proposed coder, and proposed some possible improvements.<
>