An F0 contour fitting model for singing synthesis

A flexible F0 contour fitting model that controls F0 fluctuations and generates natural F0 contours of singing-voices is shown. By considering the smoothness of the F0 contour and the closeness of the notes, a curve with natural overshoot and preparation is generated. Vibrato and fine fluctuations are easy to add. The generated F0 contour gives the voice a more “human” feel. This method has the features of approximation and flexibility. A Mandarin singing synthesis system is built. The system concatenates and modifies units from an inventory of 414 syllables, originally recorded in reading style, to form arbitrary lyrical singing voice.