摘要
在G.729的基本框架下,码本搜索采用次最优部分码本快速搜索法,知觉加权滤波器使用基于声学心理模型的知觉加权滤波器,使得8Kb/s共扼结构代数码激励线性预测语音编码在不降低语音质量的情况下降低计算复杂度。次最优部分码本快速搜索方法虽然降低搜索准确度,但是大大减小了搜索时的数据处理量;声学心理滤波器考虑人耳对不同频率信号的不同敏感度因而能获得更好的主观音质效果。仿真结果表明,该算法复杂度降低,并取得满意的合成语音质量。
On the base of G.729, a non-exhaustive codebook search method is used. And a Psycho-acoustic perceptual weighting filter is substituted for traditional perceptual weighting filter. The two methods decrease computing complexity of 8Kb/s CS-ACELP speech codec while do not reduce the synthetic speech quality. Though the non-exhaustive codebook search method debases search veracity, it scales down data processing quantity of searching. The Psycho-acoustic perceptual weighting filter is based on human acoustic structure. It takes the ears' different sensitivity to different frequency into account. So the new filter brings to the system better subjective speech effect. Simulation result shows that the new algorithm can reduce the complexity and acquire satisfied synthetic speech quality.
出处
《电声技术》
北大核心
2004年第1期56-59,共4页
Audio Engineering
