Identification of nonstationary components by complex spectral zero analysis

In some applications, such as voiced speech, the observed time series may be nonstationary due to roots on the unit circle. If there is a single root at 1 or at -1, then simple differencing will remove the nonstationary component. However, determining an operator which would remove a nonstationary component having a more complex distribution of zeros on the unit circle is a much more complicated problem. In this paper we propose a method for finding operators for seasonal type components, which have an implied placement of zeros on the circle. This is applied to short time speech signals.