一种降低ACELP码书搜索复杂度的高效算法

Efficient complexity reduction in codebook search for ACELP

为了降低代数码激励线性预测(algebraic code-excited linear prediction,ACELP)语音编码算法的复杂度,以便更好地实时实现,提出了一种有效的改进算法。在自适应码书搜索上提出了不连续的开环基音搜索算法,利用时间抽取因子对不同时延段语音样点进行不连续抽取;在代数码书的搜索上提出了一致脉冲替换法,采用脉冲位置预选和循环判断机制控制码书搜索的次数。以G.729A为实验平台进行仿真,仿真结果表明,改进的算法在保证语音质量的情况下,有效降低了ACELP码书搜索的复杂度。

In order to reduce the complexity of ACELP （ algebraic code excited linear prediction）, and achieve a better operation in real-time system, this paper proposed an effective algorithm. First, it proposed the discontinuous open-loop pitch estimation in adaptive codebook search, used the time decimation factor in different delays, and then proposed the unified pulse-replacement search algotithm in stochastic codebook search, used the prediction technique and judging mechanism to control the number of codebook search. Finally, it simulated the improved algorithm in G. 729A. The result shows that the proposed algorithm not only acquires complexity reduction but also get a high speech quality.