摘要
8kbit/s短延时语音编码算法LD-ACELP,采用了代数码本激励线性预测(ACELP)的编码方法,利用语音的帧间相关性对线谱对参数采用了分裂式矢量量化技术,并采用高效的码本结构、码本搜索技术和增益矢量量化技术来获得较高的语音合成质量和较短的算法延时。LD-ACELP的帧长为10ms,算法延时为15ms。通过信噪比及人耳主观听觉实验等性能测试表明,该算法具有与国际电联16kb/s短延时语音编码算法LD-CELP(G.728)相当的语音合成质量。
In order to satisfy the demands of speech quality and algorithm delay of speech coding algorithm in the digital voice communication system, we have studied a kind of speech coding algorithm with high quality and low delay at the bit rate of 8 kbit/s. The encoding algorithm is based on algebraic code excited linear prediction (ACELP). LSP quantization with split vector quantization (Split VQ) technique using interframe correlation, efficient codebook structure and codebook search strategies, along with efficient gain vector quantization procedures, have been developed to achieve high quality reconstructed speech and low algorithm delay. The frame length of LD ACELP is 10 ms and the algorithm delay is 15 ms. Through performance tests of SNR and subjective listening, we have verified that the proposed algorithm is mostly equivalent to the 16 kbit/s low delay speech coding algorithm LD CELP (Speech coding standard G.728 of ITU).
出处
《清华大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
1997年第9期100-103,共4页
Journal of Tsinghua University(Science and Technology)
基金
国家"八六三"高科技项目
关键词
短延时
矢量量化
语音编码
LD-ACELP
algebraic code excited linear prediction (ACELP)
low delay (LD)
vector quantization(VQ)