We have previously proposed a statistical model of speech F0 contours, which is based on the discrete-time version of the Fujisaki model. One advantage of this model is that it allows us to introduce statistical methods to learn the Fujisaki-model parameters from speech F0 contours. This paper proposes several modifications to our previous model and parameter inference algorithm, and quantitatively evaluates the performance of our modified parameter inference algorithm.
Index Terms: Speech F0 contours, statistical model, Fujisaki model, hidden Markov model, EM algorithm