摘要
提出了一种新的 6 .3kbit/sCELP语音编码方案 ,其语音质量接近ITU TG .72 9的合成语音质量 ,而编解码复杂度约为 10MIPS。它在CELP的 3个主要方面作了改进 :LSP参数的矢量量化 ;开环基音预测 ,长时预测滤波器系数的求解和量化一体化 ;代数码本激励。这些措施在保证合成语音质量的前提下大大降低了计算量 ,从而能够在单片TMS32 0C30系统上实时实现。
This paper presents a novel 6.3 kbit/s CELP speech coder.Its speech quality approaches that of ITU T G.729 while its coding complexity is only 10 MIPS.The improvements of CELP include the vector quantization(VQ) of LSP parameters,open loop pitch prediction with combination of LTP(long term prediction)coefficients solution and quantization,and algebraic codebook excitation.All these improvements can reduce computation overhead greatly while keeping speech quality so that this scheme can be realtime realized on mono chip TMS320C30 system.
出处
《南京邮电学院学报》
2000年第1期30-33,共4页
Journal of Nanjing University of Posts and Telecommunications(Natural Science)
关键词
码激励线性预测
矢量量化
语音编码
Code excited linear prediction
Linear spectrum pairs
Vector ouantization
Algebraic codebook