一种基于MELP模型600bps声码器的设计  

A 600bps Speech Coder Based on MELP

作　　者：石乔林[1] 韦凯[1] 吴辉[1] 

机构地区：[1]中国电子科技集团公司第58研究所,江苏无锡214035

出　　处：《电子与封装》2012年第10期28-30,共3页Electronics & Packaging

摘　　要：基于增强型混合激励线性预测(MELPe)模型,设计了一款600bps低速率语音编码器。该编码器在保持MELPe算法特征的同时,利用相邻帧的帧间冗余,把连续的三帧构成一个超帧,对超帧采用多模式预测和多级矩阵量化技术进行联合量化。同时针对超帧的不同模式,通过预测系数对相邻超帧的模式转换进行处理,实现线谱对参数(LSF)的矢量量化。最后对基音周期与增益参数进行联合量化,进一步提高量化效率,完成一款在600bps下仍具有较好合成语音质量的语音编码器的设计。The paper describes a 600bps speech coder based on MELP （enhanced mixed excitation linear prediction） algorithm. Consecutive three speech frames are grouped into super-frame and are jointly quantized by utilizing inter-frame redundancy in coder. The LSF vector is quantized with multi-mode predictive and multistage matrix quantization that handle mode transition by predictive coefficient and different mode in super-frames. The efficiency of the quantization is improved by joint quantization of pitch and gain. All of that make the quality of the synthetic voice better ever at 600bps.

关 键 词：MELP 语音编码 基音周期 矢量量化 

分 类 号：TN402[电子电信—微电子学与固体电子学]