一种新的峰值提取方法及其在语音基频提取中的应用  

A Novel Peak Picking Algorithm and Its Application in Extraction of Fundamental Frequency of Speech Signal

作　　者：张红[1] 宋俊寿[1] 黄泰翼[1] 

机构地区：[1]西南交通大学,中国科学院自动化研究所

出　　处：《铁道学报》1998年第6期68-73,共6页Journal of the China Railway Society

基　　金：模式识别国家重点实验室的资助

摘　　要：提出了一种基于二值化的侧抑制网络的峰值提取方法。二值化侧抑制网络具有很强的锐化作用，并具有很高的计算效率。本文将这种简单有效的峰值提取方法应用在语音基频提取算法中，对语音的短时谱进行峰值提取，形成了一种新型基频提取算法。该基频提取算法在ＣＯＳＤＩＣ数据库上进行了测试。实验数据表明利用了新的峰值提取方法的基频算法，比传统的自相关基频提取法和倒谱基频提取法有更好的性能，并验证了这种峰值提取方法的实用性。This paper proposes a novel peak picking algorithm with binary lateral inhibition network.The binary lateral inhibition network has strong sharpen effect to peaks,while it has amazing low computation load.We apply this effective peak picking method into extraction of fundamental frequency(F0),in which it serves to pick harmonic peaks from the short term spectrum of speech signal.The new F0 extraction algorithm is evaluated on COSDIC speech database.The results show that the new algorithm performance is better than two traditional algorithms——cepstrum and autocorrelation based F0 extraction algorithm,and show that the new peak picking method is practical.

关 键 词：峰值提取 语音 基频提取 二值化 侧抑制网络 

分 类 号：TN912[电子电信—通信与信息系统]