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
田春环 1姜占才1
作者信息
- 1. 青海师范大学物理系，西宁 810008
- 折叠
摘要
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。
Abstract
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.
关键词
声码器/矢量量化/LBG算法/浊音隶属度Key words
vecoder/vector quantization(VQ)/LBG algorithm/voiced membership引用本文复制引用
出版年
2015
NETL
NSTL
维普
万方数据