Mean square error optimal weighting for multitaper cepstrum estimation

EURASIP Journal on Advances in Signal Processing

Table 3 Cepstrum ξ _c for simulated AR processes, where the AR model is estimated from ‘A’ of hallo

ξ _c(K,M)	M ₁(49)	M ₂(12)	M ₃(14)	F ₁(39)	F ₂(12)	F ₃(43)
OPT098₀₀₂	0.546 (6)	0.323 (6)	0.323 (6)	0.583 (6)	0.322 (6)	0.554 (6)
OPT098₀₀₄	0.532 (11)	0.294 (11)	0.290 (11)	0.582 (11)	0.290 (11)	0.531 (11)
OPT098₀₀₈	0.529 (21)	0.259 (21)	0.257 (21)	0.590 (21)	0.245 (21)	0.522 (21)
OPT093₀₀₂	0.703 (6)	0.208 (6)	0.223 (6)	0.734 (6)	0.202 (6)	0.746 (6)
OPT093₀₀₄	0.693 (11)	0.176 (11)	0.194 (11)	0.724 (11)	0.158 (11)	0.689 (11)
OPT093₀₀₈	0.673 (21)	0.182 (21)	0.191 (21)	0.716 (21)	0.156 (21)	0.663 (21)
SIN_opt	0.630 (3)	0.193 (8)	0.216 (7)	0.643 (4)	0.179 (8)	0.629 (3)
THOM_opt	0.661 (3)	0.198 (8)	0.224 (7)	0.671 (3)	0.186 (8)	0.661 (3)
WELCH_opt	0.590 (4)	0.186 (8)	0.205 (8)	0.633 (4)	0.167 (9)	0.608 (4)
HAMM	1.69 (1)	1.64 (1)	1.65 (1)	1.71 (1)	1.63 (1)	1.70 (1)
AR_opt	0.964 (49)	0.165 (12)	0.140 (14)	0.362 (39)	0.281 (12)	0.611 (43)

There were six different speakers (three males and three females). The true model orders are noted for different speakers. The number of multiple windows K is also given after the value of ξ _c for the different methods. For the AR estimator, the estimated model order M for the minimum error is presented.