Acoustical analysis of Emotions in Hindi speech and their Transformations using Prosodic Information

Abstract

Emotion is an important cue to express the feeling of one s expression. It is also an newlineimportant factor in expressive speech synthesis. If one wants to understand the meaning newlineof an utterance, interpretation of proper emotion must be carried out; otherwise, newlinesemantically utterance will give wrong result. The present thesis describes the newlinepreliminary study of some emotional states expressed in natural speech and their effect newlineon the pitch patterns. Studies have shown that Fundamental frequency (F0) is the newlineimportant acoustic parameter of many prosodic constructs to improve the naturalness newlineand intelligibility of computer generated speech. Efforts have been made to find newlinecorrelation between the emotional state and F0. The major contribution of this research is newlineto propose an Emotion Transformation methodology. Emotion Transformation is applied newlineto synthesizing/transforming emotional speech from standard neutral speech. The newlinetransformation adopts the concept of Linear Modification Model (LMM) model. The newlineparameters used in the study are F0, segmental duration and energy factor from the newlineacoustic distribution analysis. Experiments have been carried out to verify the results by newlinemodifying segmental duration and F0 contours of neutral emotion using the Time newlineDomain- Pitch Synchronous Overlap Add (TD-PSOLA) implementation. A newlinemethodology is proposed with two concepts on the basis of unit of speech- Phrase based newlineEmotion conversion (PECT) and Word based Emotion conversion (WECT).