基于贝叶斯阴阳机的2kb/s NMF-WI语音编码算法

2kb/s Bayesian Ying-Yang Waveform Interpolative Speech Coding Based on Non-Negative Matrix Factorization

本文提出了一种改进型的基于非负矩阵分解(Nonnegative Matrix Factorization,NMF)的特征波形(Charac-teristic Waveform,CW)分解算法,一方面应用惩罚次胜者竞争学习算法(Rival Penalized Competitive Learning,RPCL)和贝叶斯阴阳机(Bayesian Ying-Yang,BYY)和谐学习算法,来计算NMF分解阶数,在没有明显降低语音质量的前提下,降低了编码器的复杂度;另一方面根据CW的能量与编码矩阵的能量间的变化关系,提出了相位谱的混合自回归合成方法,提高了语音的自然度.最后,开发出一套改进型2kb/s NMF-WI低复杂度语音编码方法,采用基于K-L散度的NMF迭代算法和收敛速度更快的基矢量Mel刻度分带初始化方法,按照基音周期的统计分布将特征波形分为6类,在CW分解模块,复杂度下降了10MOPS,语音质量提高,与采用4bit散布矢量量化相位谱的2.16kb/s NMF-WI语音编码器的语音质量相当.

An improved charracteristic waveform decomposition based on nonnegative matrix factorization was proposed. Two methods based on Bayesian Ying-Yang（BYY）harmony learning and rival penalized competitive learning（ RPCL）to compute factorization rank of nonnegafive matrix factorization（NMF）were proposed. Computational complexity is decreased and speech quality is not decreased obviously.Mixed autoregressive model for construction of WI phase was proposed according to the energy of CW and coding matrix, which improves the naturalness. In the end, a low complexity NMF-WI speech coding at 2kb/s was developed. NMF based on Kullback-Leibler divergence and Mel scale band-partitioning initialization used for basis vectors were proposed, and CWs were classified into six based on pitch dislribution. In CW factorization, computational complexity dropped by 10 MOPS. Speech quality is increased,and equivalent to 2.16kb/s NMF-WI using 4bit phase VQ.