基于原子预选择的音频匹配追踪算法

Audio Matching Pursuit Algorithm Based on Atomic Preselection

匹配追踪是一种应用于信号处理的稀疏表达贪婪算法,该算法在原子选择中使用的是遍历匹配方式,其计算复杂度较高,匹配过程中需要已知完整待处理信号,难以满足实时需要。为此,提出一种新型音频匹配追踪算法。由于采用短时非完备字典对信号进行稀疏表达,使待处理信号不受长度限制,根据信号能量分布关系,在原子匹配之前预处理,以提高匹配过程的执行速度。实验结果表明,该算法在达到现有Krstulovic'快速算法信号表达效率的同时,能降低计算复杂度,提高运行速度。

The Matching Pursuit（MP）is a sparse expression greedy algorithm which is applied to signal processing.Its computation complexity is high due to traversalmatching in atom selection,and the matching process has to know the complete signal needed to be processed,which leads its limited application in real-time circumstance.To solve these problems,this paper presents a new audio matching pursuit algorithm,which adopts a short-term and non-complete dictionary to sparse by express the signal,so that the signal to be processed is free from length limit.Besides,according to the distribution relationship of signal energy,the atom is preprocessed before matching to improve the execution speed during the matching process.Experimental results show that the algorithm not only can compare the efficiency with Krstulovic′fast algorithm in signal expression,but also can reduce the computation complexity and improve the running speed.