A modeling of pitch pattern of speech is important in the study of accent, synthesis and recognition of speech. The Fujisaki model is well known as a model of the pitch pattern of speech. The paper proposed a modeling of pitch pattern of speech based on a smoothing spline function which has a guarantee of uniqueness of solution and is more flexible compared with Fujisaki model, and clarified the efficiency of the proposed model. The paper also proposed the method of automatic selection of knots from the observed pitch data. The experimental results denoted that the pitch pattern was sufficiently represented by 3 order smoothing spline function and by similar number of parameters as Fujisaki model.
展开▼