Research On Key Points For Achieving Fast Search Of Fixed Codebook In G.729

Digital Speech Communication is the most important and common service in current Telcommunication networks. Now that frequency resouce is becoming more and more valuable with the developement of information technology, how to reduce the transmission rate of speech signals is always the target people are pursuing. To achieve so, speech coding plays a vital role in the process. Among those standards for low-bit rate speech coding, G.729, which is based on Conjugate Structured-Algebraic Code Excited Linear Predictive (CS-ACELP), becomes one of the most prevailing standards by right of its low bit rate and high speech quality. And how to search the fixed codebook which best match the input vector quickly and correctly is one key technology in G.729. Though being lower than G.729, the search complexity of traditional algorithms is still very high.Inspired by hierarchization and replacement, a fast search algorithm, Hierarchical Single Pulse Replacement (HSPR), is proposed to further reduce the search complexity. Different from the traditional ones, this algorithm focuses on hierarchizing the pulses in each track prior to pulse replacement by computing the contribution of a single pulse to the codebook. By doing so the candidate pulses for replacement are greatly reduced in the sense that pulses in LPMG (Low Probability Member Group) in each track will be tested for replacement if and ONLY if there's NO pulse in HPMG (High Probability Member Group) meeting the algorithm demand. The following simulation and analysis show that obviously HSPR outperforms other corresponding algorithms in terms of the great reduction of the search complexity.