Joint estimation of glottal source and vocal tract for vocal synthesis using Kalman smoothing and EM algorithm

In this paper, a joint parameter estimation of the derivative glottal source waveform and the vocal tract filter is presented where aspiration noise and observation noise are taken into account within a state-space model. The Rosenberg-Klatt glottal model is used in conjunction with an all-pole filter to model voice production. The EM algorithm is employed to iteratively estimate the model parameters in a maximum-likelihood sense, utilizing a Kalman smoother in the expectation step. The model and estimator allow for improved estimates of model parameters for resynthesis, yielding an output which sounds natural and remains flexible for modification, a desirable property for expressive vocal synthesis.