Speech compression using line spectrum pair frequencies and wavelet transform

A simple way to compress speech using the line spectrum pair (LSP) frequencies and wavelet transform is described. The method does not involve any training or large database. Redundancy in a speech signal is removed by passing it through the LSP frequency analysis filter. The output of the filter, termed the residual error signal, that has less dynamic range than original signal is passed through a four-level wavelet transform filter. Only the approximation and two of the details are quantized. The reconstructed speech is of high quality at 0.51 bits per sample.