An efficient complexity reduction algorithm for G.729 speech codec

Abstract

Having been widely used in a VoIP system, the ITU-T G.729 standard is of high computational complexity, a heavy load for a real-time operation on a low-end smart phone or other embedded devices even with the G.729A standard. In this paper, there are three methods proposed for an efficient complexity reduction of G.729A, i.e., a combined local optimal pulse method, employed to search an algebraic codebook, a discontinuous pitch estimation, used to analyze the open-loop pitch, and the hybrid two-stage vector quantization (HTVQ), assigned to quantize the line spectral frequency (LSF) coefficients. It is validated by experimental results that the improved version of G.729A can be easily implemented on a smart phone (312 MHz CPU, 64 MB RAM) in real-time operation. In total 80.94% of computational load can be thus eliminated in comparison with the G.729A and a satisfactory speech quality is maintained as intended, an excellent efficiency demonstration by the methods proposed.