语音多带激励模型基音搜索算法的一种改进

An improved pitch detection algorithm for the multiband excitation model

深入分析了ＭＢＥ模型分析算法的复杂度，认为原算法中的双闭环基音搜索结构有很大的改进余地，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进行了全面的对比分析．得出的结论是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．

Firstly, we discuss the complexity of the MBE model analysis algorithm. It is found that there is much to be improved in the double closed loop pitch detection framework of this algorithm. So, we present an improved MBE analysis algorithm based on the cardinal B Spline D yWT pitch detection theory. The key feature of this improved algorithm lies in the fact that we have changed the double closed loop framework to that of the open loop and closed loop. According to our analysis, this improvement is able to reduce greatly the complexity of pitch detection without weakening the other performance. By means of simulation, a comprehensive comparison in complexity and robustness between the original MBE algorithm and the improved algorithm is made, which verifies the results of our previous theoretical analysis.