Efficient statistics-based algebraic codebook search algorithms derived from RCM for an ACELP speech coder

Abstract

Applied to analgebraic codebook search conducted on an algebraic code-excitedlinear-prediction (ACELP) speech coder, two improved versions of reducedcandidate mechanism (RCM), designated as Fixed-G1-RCM and Fixed-2Track-RCM, arepresented in this study for further search performance improvement. It ismainly derived from two major research findings in a piece of our prior work.The first finding is that a pulse with a high contribution in a track is morelikely to serve as the optimal pulse in the optimal vector pertaining to thetrack, and the second is that the speech quality can be well maintained at asearch accuracy above 50%. In this proposal, the trade off can be tuned betweenthe search accuracy and the search complexity so as to reach a nearlyconsistent speech quality. It is experimentally demonstrated that either Fixed-G1-RCM(N = 3), or Fixed-2Track-RCM (M = 2, N = 3, or M = 3, N = 4) can provide ahighly superior search performance relative to a global pulse replacementmethod (iteration = 2) and an iteration-freepulse replacement method.

DOI: http://dx.doi.org/10.5755/j01.itc.44.4.9847