A recursive estimation of ARMA parameters based on a robust time varying model for speech analysis
Masahiro Serizawa, Nobuhiro Miki, Nobuo Nagai · The Journal of the Acoustical Society of America · 1988
This paper presents a recursive estimation of ARMA parameters based on a robust time-varying model for speech analysis. This algorithm is basically similar to the recursive least-squares estimation (RLS), but it is different in that the time variation of the ARMA parameters is dependent on past ones. This is based on the assumption that the speech production process does not vary instantly. This method has two linear estimators: an input estimator and a parameter estimator for known input. The variation of the parameters is estimated by using the likelihood function. The proposed method is equivalent, under certain conditions, to the RLS with the forgetting factor. However, using the proposed method, this factor can be estimated as the value that represents the variation of the parameters. Finally, the proposed method was applied to a synthetic speech and real speech. The results show that the estimated spectra sufficiently represent the dynamic movement of formants without jitters or extreme estimation errors.