Speech compression using wavelet transform

A great deal of mathematical and engineering analysis depends on methods for representing a complex phenomenon in terms of elementary, well-understood phenomena. Recently wavelet theory has provided a new method for decomposing a function or signal. Wavelet transforms of real signals tend to create low-value coefficients at the finer scale. This property makes it possible to devise compression schemes that quantise the fine-scale coefficients more severely than coefficients at other scales. In this paper, two schemes for speech compression based on wavelet transform are introduced. In the first one, short time segments of speech samples are wavelet transformed and quantised using different types of quantisers. In the second scheme, a linear prediction is carried out using the autocorrelation method and tapped delay line transversal filter, then the prediction residual is wavelet transformed, quantised, and transmitted. A good quality speech reproduction is obtained at compression ratios of 50%-90.36%. The system may find its applications in cellular and satellite communications.