基于听觉模型的基音提取

PITCH EXTRACTION BASSED ON AUDITORY MODEL

本文提出了一种新型有效的基于听觉模型的基音提取方法.它主要是模拟人类听觉系统音调感知功能,在跨通道累加自相关处理方法的基础上,增加了神经系统在感知时的时间连续性模拟,由于对空间和时间分布的信息的综合累加作用,使所提的方法不仅能提取出淹没在各种噪音下的语言信号的基音信息,而且能够判断所处理信号是否由重叠语言信号构成,并进一步提取出叠在一起的独立的基音信息.初步的实验结果证明了所提模型的有效性.

A novel and effective pitch extractor based on auditory model is described in this paper. It mainal simulates the pitch perception function of auditory system of human being. This important process is achieved by using the simulation of temporal continuity in speech perception of human auditory system on the basis of the cross channel summation autocorrelation method. Due to its composite function of the spatial and temproal information, it can not only extract pitch information of speech signals from the various interfering noises. but also judge whether the processing information is composed of two or more overlapping speech signals and further extract the independent pitch information for overlapping ones. The preliminary experimental results have proven its effectiveness.