Enhancement of VSELP speech coding for North American digital cellular

We consider enhancement of the North American IS-54/136 vector sum excited linear prediction (VSELP) speech coding in the sense of reducing complexity or improving speech quality. For complexity reduction, we find a few techniques which can reduce, respectively, the pitch and excitation search complexities by approximately 40-50% at a moderate SSNR loss of roughly 0.5 dB. For quality improvement, we consider tree-structured pitch and excitation search. Aside from demonstrating a roughly 0.3-1 dB SSNR gain, we present a technique which can significantly reduce tree-search complexity. And we analyze the simulation results to obtain rule-of-thumb formulas for quasi-optimal determination of the tree parameters under different complexity and performance conditions