id: 00916416
dt: a
an: 00916416
au: Miyanaga, Yoshikazu
ti: Adaptive identification of ARMA model and model identification system.
so: Nagai, Nobuo (ed.), Linear circuits, systems, and signal processing:
    Advanced theory and applications. New York, NY: Marcel Dekker. Electr.
    Eng. Electron. 62, 243-280 (1990).
py: 1990
pu: New York, NY: Marcel Dekker
la: EN
cc: 
ut: speech analysis; ARMA model; adaptive identification; input estimation;
    model reduction
ci: 
li: 
ab: This is chapter 9 of the volume under review. Speech analysis and synthesis
    are problems of constant interest in many domains of communication.
    Speech analysis requires the determination of the discrete transfer
    function of the vocal tract, i.e. $H(z)$. Now it is an accepted fact
    that $H(z)$ must be a rational function whose coefficients are to be
    found. The resonances of the transfer function correspond to formants
    whereas the antiresonances to antiformants. The simplest form of $H(z)$
    is the one called the all-poles form and it corresponds to the
    so-called auto-regressive (AR) model. The poles and zeros form
    corresponds to the so-called auto-regressive moving average (ARMA)
    model. It is clear that the AR model cannot accurately describe the
    speech signal so that the ARMA model should be taken into account. In
    the paper, it is shown that the choice of the order of the ARMA model
    is more delicate than that of the AR model. Due to the fact that the
    speech signal is not a truly stationary stochastic signal, adaptive
    methods are of high value in applying the ARMA method. This
    contribution presents a new method for the application of the ARMA
    model for speech signals. It is called model identification system
    (MIS) and contains three algorithms: adaptive identification, input
    estimation, and model reduction (this last one requires high
    calculation cost). It is shown that the MIS must be modified according
    to the fact whether the input signal is a white noise or an aperiodic
    impulse train. Also, it is emphasized that the order estimate is first
    established by the AR method and then it is accurately determined by an
    ARMA model approach. This contribution constitutes an excellent
    introduction to new and recent approaches in speech analysis.
rv: D.Stanomir (Bucureşti)