Efficient search procedures for selecting the optimum innovation in stochastic coders

The authors describe several procedures that simplify the search in stochastic coders, but do not put constraints on the structure of the codewords. The computational efficiency is achieved by moving the search for the best codeword from time domain to another domain where the convolution operation in the linear filter is reduced to a single multiplication operation. A computational reduction by a factor of approximately 20 is achieved, without producing any perceptual degradation of the synthetic speech. The advantages of search procedures in a transform domain are discussed, not only in terms of efficiency, but also to provide better insights into the function of stochastic coders