基于B-样条子波变换的MBE语音模型分析算法改进

An Improvement in MBE Speech Model Analysis Algorithm Using B-spline Wavelet Transform

首先对 ＭＢＥ模型分析算法的复杂度进行了深入的分析，认为原算法中的双闭环基音搜索结构有很大的改进余地．为此提出了一种基于Ｂ－样条二进离散子波变换（Ｂ—ＳＤｙＷＴ）基音检测法的改进ＭＢＥ模型分析算法。其关键在于利用Ｂ－ＳＤｙＷＴ基音检测原理将这种双闭环搜索算法改造成先开环后闭环的结构。理论分析可以说明这种改造可以大幅度地降低基音搜索的复杂度，而不降低其性能。通过仿真实验，我们对改进前、后的ＭＢＥ模型基音搜索算法的复杂度、准确度以及抗噪声能力等方面作了一个全面的对比分析。得出这样的结论：改进算法与原算法相比，各项性能均不低于原算法甚至有所改善，特别是证实了理论分析中的复杂度降低作用。

We discussed the complexity of MBE model analysis algorithm, and pointal out that there is much can be improved in the double cloud-loop pitch detection framework of this algorithm. We presents an improved MBE analysis algorithm based on cardinal B-spline DyWT pitch detection theory. The key feature of this improvement algorithm lay on that we changal the double closed-loop framework to that of open-loop and closed-loop. Accroding to our analysis, this improvement is able to reduces greatly the complexity of pitch detec- tion without weakening the other performances. By means of simulation, a comprehensive comparison, such as the case of the complexity and the robustness, between the original MBE algorithm and the improved algorithm is made. Through there comparisons, the results of our previous theory analysis are verified.