Analysis and synthesis of laughter

Abstract

There is much enthusiasm in the text-to-speech community for synthesis of emotional and natural speech. One idea being proposed is to include emotion dependent paralinguistic cues during synthesis to convey emotions effectively. This requires modeling and synthesis techniques of various cues for different emotions. Motivated by this, a technique to synthesize human laughter is proposed. Laughter is a complex mechanism of expression and has high variability in terms of types and usage in human–human communication. People have their own characteristic way of laughing. Laughter can be seen as a controlled/uncontrolled physiological process of a person resulting from an initial excitation in context. A parametric model based on damped simple harmonic motion to effectively capture these diversities and also maintain the individuals characteristics is developed here. Limited laughter/speech data from actual humans and synthesis ease are the constraints imposed on the accuracy of the model. Analysis techniques are also developed to determine the parameters of the model for a given individual or laughter type. Finally, the effectiveness of the model to capture the individual characteristics and naturalness compared to real human laughter has been analyzed. Through this the factors involved in individual human laughter and their importance can be better understood.