一种新的语音信号共振峰提取的算法

A New Calculation Method of Extracting Formant

本文提出了一种新的语音信号共振峰的提取方法。在LPC幅度谱上搜寻最大的极大值点所对应的频率,并将它作为构成声道参数的某一谐振腔所对应的共轭复根的角度,再通过LPC系数的相—频特性的一次导数和三次导数相结合的方法求出这对共轭复根的幅度,从而确定了该谐振腔,也就得到了该谐振腔的共振峰。然后,用LPC的多项式对该谐振腔所对应的多项式做多项式除法,得到新的LPC系数,接着重复前面的步骤,可以较好地求出在LPC谱中对应幅度最大的两个共振峰。

This paper presents a new method for extracting formants. We search for the frequency on which the LPC spectrum has the maximum max. , considering it as the angle of a complex root pairs derived from one resonator which makes up of the vocal tract. Then the magnitude of the complex root pairs will be extracted by adopting the first and third derivatives of LPC phase-frequency .characteristics. After the root location has been decided,we get the formant of the resonator. As the next step, we divide the LPC polynomial by the resonator polynomial. The result polynomial would be used for the new down-order LPC coefficients. Repeating the steps shown above,we can extract the first two formants.