Efficient Search Procedures for Selecting the optimum Innovation in Stochastic Coders

Stochastic linear predictive coders are presently being regarded as potential candidates for producing high quality synthetic speech at bit rates as low as 4.8 kbits/sec. Their inherent computational complexity is, however, one of the drawbacks. In fact, determining the optimum innovation sequence in stochastic coders has been done by exhaustive search of a random codebook in which each of the codewords must be individually filtered. As a result, these coders require more than 500 million multiply-add operators per second - a capability much beyond the reach of present digital hardware.