改进的LBG算法及浊音隶属度码本的训练

Improved LBG Algorithm and Training of Voiced Membership Codebook

为了改善MELP声码器解码端混合激励信号的性能,提出了五维浊音隶属度矢量的概念及其提取算法,用其提取并建立了浊音隶属度码本的训练样本集;用改进的LBG算法训练了浊音隶属度码本;将此码本运用到混合激励声码器中。仿真实验表明,解码语音的质量较之LPC声码器有显著的提高。

In order to improve the performance of MELP vocoder decoding end mixed excitation signal, the concept of a five dimensional voiced membership degree vector and its extraction algorithm is put forward, for the extraction and the establishment of the voiced membership codebook training set;with improved LBG algorithm training voiced membership codebook;this codebook to use mixed excitation vocoder. Simulation results show that,the quality of the decoded speech is significantly improved than that of LPC vocoder.