Lattice quantizers in speech coding

Based on the theory of spherically invariant random processes, which are excellent models for speech signals, an algorithm is deduced that maps the signal into vectors with uniformly distributed components. Thus for the first time lattice quantizers can be used for optimal encoding of speech signals. Experimental results are presented, taken from modeled as well as real speech signals. The mapping effect is illustrated, and the quantizers are rated with regard to their performance.<<ETX>>