摘要
为了解决高质量散布脉冲码激励线性预测语音编码算法中的代数码书搜索复杂度高的问题,通过在候选的脉冲位置上保持目标信号和脉冲响应的相对最大相关,提出了一种不降低主观话音质量的多路径散布脉冲搜索算法,计算负载从19.66 MOPS减少到0.069 4:MOPS,大幅度减轻了散布脉冲码书的计算负载.主观和客观听力测试表明,算法的改进前后的合成语音质量没有感性差别.
In order to reduce high complexity for algebraic codebook search in high quality dispersed-pulse CELP speech coding, a multi-path algorithm was proposed in DP-CELP coder at 4 kb/s by keeping relatively maximum correlation between target signal and impulse response in the possible position of pulse. The complexity of the algebraic codebook search is reduced to 0.069 4 MOPS from 19.66 MOPS by using this method. Subjective and objective hearing test indicates that the reconstructed speech quality has no perceptual difference from the original one.
出处
《北京工业大学学报》
CAS
CSCD
北大核心
2004年第2期161-165,共5页
Journal of Beijing University of Technology
基金
Supported by the Science and Technology Project of Beijing Municipal Education Commission (KM200310005024)
Xinxing Project of Beijing Municipal Science and Technology Commission (953811300).
关键词
语音编码
码激励线性预测
多路径码书搜索
speech coding
code-excited linear prediction (CELP)
multi-path codebook search