Spectral Estimation of Voiced Speech Based on System Identification