Fast Methods for the CELP Speech Coding Algorithm

When the original CELP algorithm was introduced in 1983, it proveded good speech quality at intermediate bit rates, but required an extremely high level of computational effort. To allow the implementation of the algorithm on modest hardware (one or two general purpose digital signal processing devices) many fast versions of the algorithm have recently been developed. Here the computational and storage requirements of these methods are compared. The methods can be divided into methods which use modified codebooks, methods which use preselection of codebook entries, transform methods, the autocorrelation method, methods which replace computation with storage and off-line computations, and procedures using overlapping codebook vectors.