1983 年 39 巻 8 号 p. 512-520
The ARMA(autoregressive and moving-average)model provides an effective means for precise representation of the production process of speech signals, especially the nasals and fricative consonants. A stable and efficient estimation of ARMA parameters was accomplished directly from the speech signal using the SEARMA(simultaneous estimation of ARMA parameters)method proposed by the authors. In order to trace the temporal variation of the poles and zeros with a high accuracy, a method was also proposed for adaptive control of the location and the length of an analysis interval. The paper describes a theoretical relationship among the location and the length of an analysis interval and the convergence of estimated parameters, for stationary and non-stationary speech signals. The effects of adaptive control of an analysis interval are also investigated by analysis of both synthetic and natural speech signals.