带有帧间级间预测的线谱频率参数多级矢量量化  

High efficiency MSVQ for prediction linear spectrum frequency parameters with inter-frame and inter-stage prediction

作　　者：李晔[1] 彭坦[1] 许明[1] 计哲[1] 崔慧娟[1] 唐昆[1] 

机构地区：[1]清华大学电子工程系,北京100084

出　　处：《清华大学学报（自然科学版）》2009年第7期981-983,共3页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(60572081)

摘　　要：为在极低速率下实现高质量的语音编码,提出一种高效的带有帧间及级间预测的线谱频率参数多级码本矢量量化(IFP-MSVQ-ISP)算法。算法利用多级矢量量化中上一级码本的选定码矢对残差矢量进行预测,对去除预测分量的残差矢量再进行下一级矢量量化。测试结果表明,这种带有多级码本级间预测的算法与无级间预测的算法相比,能够有效降低线谱频率参数的量化误差,使谱失真降低0.1 dB以上,合成语音客观MOS提高0.02以上。该算法的实现对极低速率下语音压缩编码算法的研究具有重要的参考价值。A high efficiency multi stage vector quantization algorithm with inter-frame prediction and inter-stage prediction （IFP-MSVQ-ISP） was developed to predict the linear spectrum frequency （LSF） parameters. The residual vector of each quantization stage is first predicted by utilizing the selected vector from the pre-stage analysis to calculate the remaining vector. Simulations show that the algorithm greatly reduces speech spectral distortion by 0. 1 dB while enhancing the mean opinion score of the reconstructed speech by 0. 02 compared with the classic MSVQ without inter-stage prediction. Thus, the algorithm is very useful for very-low-bit-rate speech coding.

关 键 词：语音编码 多级矢量量化 级间预测 

分 类 号：TN912.3[电子电信—通信与信息系统]