Simulating stochastic behaviors in intonation control

This paper presents the extension of the quantitative target approximation model for simulating not only intonations but also stochastic behaviors in speech production. The simulation method first samples the initial F₀ state from the initial F₀ distribution. Then, to implement the communicative function, it predicts the pitch target parameters from the joint probability distribution of the initial F₀ state the possible pitch target parameters. The stochastic simulation can generate highly correlated pitch targets from the distribution.