摘要
给出一种高质量的4Kb/s更新式码激励线性预测(RCELP)语音编码算法。该算法的编码器帧长为20ms,主要特点是使用了从自适应激励信号中分析得到的码本作为固定码本,采用预测式两级分裂矢量量化器量化线谱对(LSP)参数。主观试听表明,该算法的MOS值为3.67,其语音质量与32Kb/sADPCM基本相当。
This paper proposes a4Kb/s speech algorithm called renewal code excited linear prediction(RCELP).This coder operates on speech frames of20ms.Its main features are as follows:the fixed codebook is generated from adaptive codebook,and the predictive two stage split-vector quantizer is used to quantize line spectrum pair(LSP).Subjective testing indicates that RCELP scores3.67in MOS and the speech quality produced by RCELP is equivalent to that of32Kb/s ADPCM.[
出处
《电声技术》
北大核心
2004年第4期54-56,60,共4页
Audio Engineering