线谱频率帧内、帧间联合预测算法  

Joint Intraframe and Interframe Coding of Line Spectrum Frequency

作　　者：韩笑蕾[1] 赵晓群[1] 张楠[1] 

机构地区：[1]同济大学电子与信息工程学院,上海201804

出　　处：《系统仿真学报》2013年第5期1053-1059,共7页Journal of System Simulation

基　　金：水声通信与海洋信息技术教育部重点实验室(厦门大学)开放课题基金资助项目(UAC200902)

摘　　要：为充分利用线谱频率的帧内及帧间相关性降低语音的编码速率,设计了基于偏最小二乘及其简化算法的帧内、帧间联合预测模型。该模型可根据浊清音出现次序的不同,利用前后帧的线谱频率及当前帧的第i-1个已预测的线谱频率对当前帧的第i个线谱频率进行预测。结果表明,偏最小二乘模型及其简化模型均有效降低线谱频率的动态范围,其中基于偏最小二乘回归算法的预测模型最为精确,而简化模型的运算量及计算复杂度均优于偏最小二乘回归算法的预测模型,在采用(4,6)分组SVQ量化器对LSF参数预测误差进行量化时,每帧仅用7比特即使平均谱失真小于1 dB,较2.4 kbps下MELP编码标准中每帧对线谱频率的量化比特节省了18比特。In order to reduce speech coding bit rate, the advantages of intra-frame and inter-frame correlation of Line Spectrum Frequency （LSF） were taken to design a prediction model as well as a simplified prediction model based on partial least squares （PLS） algorithm. According to the U/V pattern of super-frame structure, the proposed algorithms could predict a LSF not only by some LSFs from the previous and next frame, but also by its exactly previous LSF in the same frame. Simulation results show that the PLS model and its simplified model can both effectively reduce the dynamic range of LSF. The PLS model is more accurate, while, the simplified model performs better in computational load and complexity. Using SVQ to split the prediction error vector into two sub-vectors of dimensions （4,6） can get an average spectrum distortion about 1 dB at 7 bits/frame, whose bit rate are less than 2.4 kbps MELP coding standard 18 bits/frame.

关 键 词：线谱频率(LsF) 偏最小二乘(PLS) 帧内 帧间相关性 EEDSVQ 多元线性回归(MLR) 

分 类 号：TN912.3[电子电信—通信与信息系统]