摘要
在分组网络中传输语音时，因为存在流量过载的情况，所以需要考虑其引发的分组丢失对生成语音质量的影响．对于采用合成分析法的语音编码方式，一旦分组（帧）丢失，会引起译码器输出语音的质量明显降低．介绍了一种基于共轭代数码本激励线性预测编译码器丢失语音帧的重建机制，依照Ｇ．７２９标准的三种兼容程度（完全兼容、比特流兼容和不兼容）分别建立模型，重点在于外推出激励信号和综合滤波器参数．运用本机制模拟随机性和突发性的帧丢失两种情形，发现在帧丢失率达３％时语音质量几乎没有降低，在可同时修改编码器和译码器的情况下，当帧丢失率达到１０％时。
A
common issue of speech transmission through packetized networks is to consider the
discarded packets as a result of overload.For the CELP based speech coder,the missing frame
will cause significant quality degradation. To improve G.729 output speech quality, the three
cases were described:(1)compatible with G.729,(2)compatible with bit stream,and (3)change
both the encoder and decoder.The reconstruction model extrapolated the excitation signal and
the short term synthesis filter.These techniques were quite effective.Very little degradation was
observed at frame loss rate up to 3%.For case 3,the output speech was still intelligible at 10%
frame loss rate.
出处
《上海交通大学学报》
EI
CAS
CSCD
北大核心
1999年第1期119-121,共3页
Journal of Shanghai Jiaotong University
基金
美国贝尔实验室上海分部资助
