Fumio Sugiyama

An LPC vocoder for efficient implementation

Linear prediction has been widely used in speech coding. However, a large amount of calculation is necessary for parameter extraction by the autocorrelation method, so it is not suited for hardware implementation. This paper proposes a new analysis method, involving iterative method in the lattice circuit for successive K parameter approximations. In the iterative method, the improved hybrid algorithm was found best, from the viewpoint of high speed convergence for speech. For pitch extraction, partly divided average magnitude difference function (AMDF) method was used. Using these two methods, a vocoder was developed which requires only a small amount of calculation and is suited for hardware implementation (e.g. LSI). Good speech quality was obtained.