Glottal source models for speech coding and synthesis

An algorithm for simultaneously estimating the parameters of a glottal source model and the vocal tract filter is described. The five-parameter LF model is used for the glottal source, and the vocal tract filter is modeled as all-pole. The model parameters are assumed to be different in closed and open glottal phases to allow for the additional damping of the formants in the open phase. The analysis of steady vowel sounds from four speakers shows that the analysis method leads to an accurate estimate of accurate estimate of the glottal source waveform and the vocal tract formants.<>