Acoustic Model-Based Filter Structure for Synthesizing Speech Signals

This paper proposes a filter structure suitable for speech synthesis applications. We first derive the lossy pole-zero model by employing the wave digital filter(WDF) adaptor formula, and by converting the fixed termination value - 1 into a loss factor $\mu$c$\in$(-1, 1). Then we discuss how to determine the reflection We employ the Durbin's method in estimating the numerator polynomial of the lossy pole-zero transfer function from the given speech sound, and then apply the step-down algorithm on the numerator to extract the reflection coefficients of the closed-termination tract. For determining the reflection coefficients of the other parts we employ a pre-calculated pole-estimator polynomial.