一种快速的代数码书搜索方法

A fast algebraic codebook search method

代数码激励线性预测(ACELP)编码算法在各类语音编码标准中得到广泛采用。在ACELP码书搜索时,要得到全局最优的码向量,所需运算量很大,难以在硬件系统上实时实现。为降低计算复杂度,提出了一种基于脉冲预选和替换步骤的快速码书搜索方法。该方法采用预选技术来减少可能的脉冲组合,利用两级搜索与判断机制来改善脉冲替换搜索的效率。实验表明,该方法可以有效地降低复杂度,并维持较高的合成语音品质。

The algebraic-code-excited linear prediction （ACELP） coding algorithm has been widely adopted in various standardized speech coders. In ACELP coders, the global optimization of codevectors requires a huge amount of computational load which brings the bottleneck in real-time implementation on hardware systems. For reducing the computational complexity, a fast codebook search method based on pulse pre-selection and replacement procedure was proposed in this paper. With possible pulse combinations reduced first by using a pulse position prediction technique, the two-level searching strategy and judging mechanism were then employed to enhance the search efficiency in pulse replacing. The experimental results show that the proposed method maintains acceptable speech quality with complexity considerably reduced.