ROBUST RECURSIVE AR SPEECH ANALYSIS

In this paper a new robust recursive method of estimating the linear p rediction parameters of an auto-regressive speech signal model using w eighted least squares with variable forgetting factors (VVFs) is descr ibed. The proposed robust recursive least-squares (RRLS) method differ s from the conventional recursive least-squares (RLS) method by the in sertion of a suitably chosen nonlinear transformation of the predictio n residuals. The RRLS algorithm takes into account the contaminated Ga ussian nature of the excitation for voiced speech, and the effect of n onlinearity is to assign less weight to the small portions of large re siduals so that the spiky excitation will not greatly influence the fi nal AR parameter estimates, while giving unity weight to the bulk of s mall to moderate residuals generated by the nominal Gaussian distribut ion. In addition, the VFF is adapted to a nonstationary speech signal by a generalized likelihood ratio algorithm, which accounts for the no nstationarity of a speech signal. The proposed method has a good adapt ability to the nonstationary parts of a speech signal, and gives low b ias and low variance at the stationary signal segments. The feasibilit y of the robust approach is demonstrated with both synthesized and nat ural speech.