Adaptive orthogonal transform coding systems for speech signals

In transform coding systems a block of speech samples is transformed into a set of spectral coefficients; these coefficients are then quantized independently and transmitted. It has been shown that nonadaptive transform coding systems are not very efficient for encoding of speech. Therefore the possibilities of using adaptive strategies to match the transform coding system to the changing statistics of speech signals are studied. The main results are: there is no need for changing the transform matrix. The discrete cosine transform is nearly optimum for most speech sounds. By using an adaptive bit assignment technique significant coding efficiencies can be achieved. A coding system based on a time-invariant cosine transform has been developed.