作　　者： ; ; ;

机构地区： 华南师范大学数学科学学院

出　　处： 《计算机学报》 2006年第1期106-115,共10页

摘　　要： 利用Hilbert-Huang变换对语言信号处理中基于事件的基音周期检测问题提出了一种新的检测方法.该方法利用Huang等人1998年提出的具有高时频分辨能力的H ilbert-Huang变换分析语音信号,并提取其瞬时能量,通过精确定位声门脉冲发生的时刻,从而精确地跟踪基音周期的变化,达到精确检测基音周期的目的.与传统方法相比,其优点主要表现在:(1)不需要对语音信号作短时平稳性假设;(2)检测精度高,适应范围广;(3)具有跟踪基音周期变化的能力;(4)能精确区分清浊音;(5)与传统方法相比,帧长大大增加,因而,在提取连续语音信号的基音轮廓时,用于分帧和拼合的开销大大减少,帧间拼合痕迹小.仿真数据和实际语音信号检测实验均获得了相当精确的检测结果.最后,需要指出的是,H ilbert-Huang变换作为一种新的信号分析方法,被成功地用于提取语音信号的基音周期,这本身是一个有意义的探索,它为拓展H ilbert-Huang变换理论的应用给出了一个新的尝试. In this paper, a novel algorithm of detecting pitch period from a speech signal based on Hilbert-Huang transform is proposed. Due to its high time-frequency local character and being applicable to nonlinear and non-stationary process, Hilbert-Huang transform is employed to analyze a speech signal and calculate its instantaneous energy, based on which the glottal pulses can be located accurately and the variation of the pitch period can be traced. As a result, the pitch period can be detected accurately. Comparing with the existing methods, the algorithm has advantages as following： It is unnecessary to assume that the pitch period is stationary within any segment; A high accuracy and a wide applicability can be received; It can be used to trace the variation of the pitch period; It is easy to discriminate unvoiced and voiced; The length of a frame is lengthened, which saves the consumption used to put together frames as being done in some existing methods when a continuous conversational speech signal is processed. Experiments on both synthesized data and real speech signals show encouraging detection results. It is a significative attempt to apply the new theory of Hilbert-Huang transform to detect pitch period from a speech signal.

关 键 词： 经验模型分解 变换 基音周期 基音检测

领　　域： [自动化与计算机技术] [自动化与计算机技术]