Pitch contour modelling and modification for expressive Marathi speech synthesis

In this paper, we have measured and analyzed features of speech signal such as fundamental frequency, jitter and shimmer its statistical modeling for Marathi. These models can be used for modifying prosody of the neutral speech further. Jitter and shimmer are measures of cycle-to-cycle variations of fundamental frequency and amplitude respectively. It characterizes the emotion and differs in values as emotion varies. An emotion or target model mentioned here is in the form of interrogate. A pitch target model is developed to model and modify the prosody of the Marathi words. The study comprises the study of existing pitch contour of words whose prosody is to be modified and target pitch contour. Its statistical analysis is done. At the end Gaussian normalization is employed to modify the prosody with help of analyzed data. Result of the subjective experiments satisfies the native listeners.