语音清浊音分类及浊音谐波提取算法——三阶累积量基于正弦语音模型的应用  

A Speech Unvoiced/Voiced Classification and Voiced Harmonic Extraction Algorithm--Using Third-order Cumulant Based on Sinusoidal Speech Model

作　　者：应娜[1] 赵晓晖[1] 

机构地区：[1]吉林大学通信工程学院信息科学实验室,长春130012

出　　处：《计算机工程与应用》2006年第1期64-67,76,共5页Computer Engineering and Applications

摘　　要：在低信噪比和非平稳噪声干扰下,语音信号的清浊音检测是语音信号处理中的一个重要研究问题。论文基于语音正弦模型,提出了一种清浊音分类和浊音谐波提取算法。该方法在分析了语音的三阶累积量谱后,用子谐波-谐波方法取得基音,并计算出谐波参数和高低频能量比值。它利用谱包络估计器得到谱包络及尖峰信号,结合最小均方估计准则下的迭代算法计算语音谐波的信噪比;通过对上面各计算结果的综合评价得出语音帧的浊音度,从而得到语音清浊音的分类和浊音谐波数。仿真结果表明,该算法在复杂噪声背景下,能有效进行语音分类,准确得到浊音度。同时该算法还具有实时性好、语音参数分析精度高的特点。Unvoiced/volced detection of speech is a challenging problem especially in low signal-to-noise ratio or the non-white-stationary noise environment.Based on speech sinusoidal model,an algorithm for the classification of unvoiced/voiced speech and the extraction of pitch in voiced speech is proposed.After the analysis for speech signal with of three-order cumulant spectrum,this algorithm through using subharmonic-harmonic rate method can compute harmonic parameters and high-low frequency energy ratio.It utilizes spectral envelope estimation to obtain envelope and peak signal and calculates signal-to-noise ratio of speech harmonics in iterate form under least square estimate criterion. Through evaluation of the above calculation resuhs,voicing probability in the analyzed frame,the classification of the unvoiced/voiced speech and the voiced harmonic number are obtained.Simulation results indicate that the proposed algorithm,under the complicated background noises,especially Gaussian noise,can effectively classify speech in high accuracy for voicing probability,and is easy for real time implementation with high precision of analyzed speech parameters.

关 键 词：正弦语音模型 三阶累积量 清浊音分离 谐波提取 

分 类 号：TN911[电子电信—通信与信息系统]