Figure 3: Block diagram of the proposed AKF-based SEA. AKF-SMBT [19, 20] do not address the conditions that have time varying amplitudes. To cope with this problem, in this paper, we first es- timate noise PSD, P,(l,m) from each noisy speech frame using an SPP method [26] (described in section 3 .2). Then employ inverse Fourier transform to P, (l,m), yields an es- n~ timate of the noise autocorrelation matrix, R,,(7), where T is the autocorrelation lag. By solving R,,(7) (q = 40). As in [19], to reduce bias in the estima a2.) for each noisy speech frame, we compute t the corresponding pre-whitened speech, y(n, /) autocorrelation method [21, Chapter 8]. The rame, the estimates of ({a;}, 02,) address condi have time-varying amplitudes. using the Levinson-Durbin recursion [21, Chapter 8], gives ({b;}, 72) ted ({ai}, hem from using the ramewise Yw(n,l) is obtained by employing a whitening filter, H(z) to y(n,l). With estimated {b;}; H.,(z) is constructed as in eq. (48). Unlike AKF-RMBT and AKF-SMBT [19, 20], since H,,(z) is constructed with {b,;} for each noisy speech tions that