一种语音信号对数幅度谱包络的无偏算法

LPC谱估计算法摆脱不了全极点模型的束缚,通过提高模型阶数来获得更好的频谱包络,但其在谱的谷值处有很大偏差的缺点.文中对倒谱法通过对对数频谱密度函数定义和对残差的迭代更新,得出了一种对数幅度谱包络的无偏算法.与LPC和MEL-LPC法比较具有较多优点,为语音信号零极模型的估计和对语音特征研究提供依据.

An Agonic algorithm of Speech Logarithm Spectrum Envelope

LPC spectrum estimation is the main method at present,but it can not get rid of the chain of the pole zero moldel of signals.It has a better spectrum envelope through increasing the power of the models.There are some disadvantages such as difference in the valley of the spectrum.In this paper,the conventional methods of spectrum are studied,and an agonic arithmetic of the logarithm spectrum envelope by defining the desity function of the logarithm Spectrum and updating residual are given.The experiment shows that this method is superior to LPC and MEL LPC in estimating spectrum envelope.It can give an ideal estimation for the mixed pole zero moldel of speech signals and is helpful for the further research of the format of speech signal.