改进的基于GMM谱包络拟合的语音编码器  

New vocoder based on spectral envelope fitting using improved GMM

作　　者：王蓉蓉[1] 曾毓敏[1] 周挺挺 李平[2] 

机构地区：[1]南京师范大学物理科学与技术学院,南京210023 [2]泰州职业技术学院信息工程技术学院,江苏泰州225300

出　　处：《中国科技论文》2016年第20期2289-2295,共7页China Sciencepaper

基　　金：江苏省科技支撑计划项目(BE2014139);江苏省自然科学基金资助项目(BK2010546)

摘　　要：提出了1种改进的基于多项式拟合和高斯混合模型的甚低码率语音编码器(polynomial fitting and Gaussian mixture model,PGMM)。该编码器将固定数量的语音帧划分为1个片段,利用相邻帧间的相关性对片段内的高斯混合模型参数进行多项式拟合,并用PGMM模型拟合语音谱包络。仿真结果表明,PGMM算法极大地降低了编码器的编码速率,在0.86kb/s时仍然可以获得理想的解码语音。A vocoder is proposed to improve low bit-rate speech, based on polynomial fitting and Gaussian mixture model （PGMM）. In the vocoder, several frames are collected into a segment according to the correlation between neighboring frames. The polynomial trajectory is used to fit the parameters of Gaussian in a segment and bring in the PGMM model. The new model is used to fit the speech spectrum envelope. The results show that the bit rate of PGMM vocoder is further reduced meanwhile it can get an intelligible decoding speech at 0. 86 kb/s.

关 键 词：语音编码 高斯混合模型 多项式拟合 范德蒙矩阵 

分 类 号：TN912.3[电子电信—通信与信息系统]