A power-conserved real-time speech coder at low bit rate

A real-time speech coder based upon codebook exciting linear prediction (CELP) with a front voice activity detector at around 75 k/ps, which can operate acceptably in a mobile communication environment is presented. This coder possesses two digital signal processors arranged in parallel to implement the exhaustive codebook search simultaneously. The coder constructs a linear-shifted Gaussian noise codebook to reduce computational complexity. A pair of bidirectional recursive functions are derived based on the codebook structure to drive the dual processors to operate in parallel efficiently. An open-close loop combined search procedure is adopted to compute long-term predictive parameters, which permits an acceptable tradeoff between speech quality and computational complexity