A dynamic tonal perception model for optimal pitch stylization

Automatic pitch stylization is an important resource for researchers working both on prosody and speech technologies. In order to be useful, the stylized F0 curve should contain the fewest possible number of control points while remaining, at the same time, close to the original curve from a perceptual point of view. Here, a pitch stylization algorithm aimed at finding the optimal balance between the number of employed control points and perceptual equality with respect to the original curve is presented. Rather than being defined by means of statistical closeness to the original F0 curve, the quality of the stylized curve is defined on the basis of a dynamic tonal perception model.