Superpositional HMM-based intonation synthesis using a functional F0 model

This paper addresses intonation synthesis combining statistical and functional approach with manipulation of fundamental frequency (F-0) contours in HMM-based speech synthesis. An F-0 contour is represented as a sum of micro, accent, and register components at the logarithmic scale, which is rooted in the Fujisaki model. Separated context-dependent (CD) HMMs are trained for each type of components extracted from a speech corpus based on a functional F-0 model. At the phase of synthesis, CDHMM-generated micro, accent, and register components are superimposed to form F-0 contours for input text. Objective and subjective evaluations are carried out on a Japanese speech corpus. Compared with the conventional approach, this method not only demonstrates the improved performance in naturalness of synthetic speech by achieving better global F-0 behaviors but also shows its flexibility for intonation manipulation through modifying the functional model parameters.