基于非负矩阵分解的1kbit/s波形内插语音编码算法  

1kbit/s waveform interpolation speech coding based on non-negative matrix factorization

作　　者：汤一彬[1,2] 吴海洋[1,2] 吴镇扬[1,2] 

机构地区：[1]东南大学信息科学与工程学院,南京210096 [2]东南大学水声信号处理教育部重点实验室(B类筹),南京210096

出　　处：《东南大学学报（自然科学版）》2010年第4期670-675,共6页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(60971098)

摘　　要：为了进一步降低编码速率,提出了一种基于非负矩阵分解的1kbit/s波形内插语音编码算法.该算法对特征波表面的幅度矩阵进行非负矩阵分解,以获得局部特征矩阵,并对该局部特征矩阵进行约束和改进,使优化后局部特征更加突出.对应的基矢量进一步稀疏,从而有利于对权矢量的量化,以实现对特征波表面的高效编码.该算法同时加入清浊音标志,对特征波表面的相位谱进行估计,以更好地提高合成语音质量.实验表明,该算法能够在1kbit/s的低编码速率条件下,获得与1.2kbit/s混合激励线性预测语音编码算法相近的合成语音质量,取得了较好的效果.A 1kbit/s waveform interpolation speech coding is proposed based on non-negative matrix factorization to achieve the lower encoding bit-rate.The new coder decomposes the magnitude matrix of the characteristic waveform surface with the non-negative matrix factorization to obtain the local characteristic matrix.Then the local characteristic matrix is improved with some constraints to enhance the local characters and make the corresponding basis vectors sparser,which is propitious to the quantization of the weighted vectors and can encode the character waveform surface more effectively.The surd/sonant flag is also introduced to estimate the phase spectrum of the character waveform surface,which can improve the synthetic speech quality better.The results show that the new algorithm can make the synthetic speech quality nearly the same as the speech quality from 1.2kbit/s mixed excitation linear prediction speech coding algorithm in the case of low encoding bit-rate of 1kbit/s,and achieves a good performance.

关 键 词：非负矩阵分解 波形内插 特征波表面 

分 类 号：TN912.3[电子电信—通信与信息系统]