Analysis And Modelling Of Vocal Tract And Emotional Speech In Indian Native Languages

Abstract

Human beings use verbal as well as non-verbal based communication during their interaction with the world. Speech is perhaps the best method of interaction which is often an unique signal convey linguistic as well as non-linguistic information with respect to time. Phonemes are articulated during the production of speech. Most of the languages incorporate about 40-50 phonemes which act as alphabets to develop various words which are built based on sequences of phonemes (syllables) during speech thereby forming sentences. The distinctiveness with respect to phonetic and the individual speaker among the speech sample collected by individual subjects (male and female speakers) are largely ingrained with respect to vocal tract shapes. It is observed that most of speech researches were carried out by considering the speech production related investigations with the assumption that distinct patterns of acoustic cues will be found to be associated with discrete emotional state. Most of the emotional speech analysis focussed on source related acoustic features or cues. newlineThe concept of important prosodic features is important features which identifies emotional state of intra and inter speakers. The current research includes identification of speech to be neutral or emotional while categorizing the emotional aspects of recorded speech signal. Emphasis was provided to the identification of emotional status of individual with the aid of feature extraction of speech and to model the vocal tract shape for different emotional states. The estimation of vocal tract area function of the related specific speech signal uses the Linear Predictive Coding (LPC), which is a dominant aspect for the assessment of vocal tract area function of the related specific speech signal based on the Auto-Regressive (AR) models. The research aims on the modelling and analysing the vocal tract shape under primary emotions like Sad, Happy, Angry and the Neutral states for acted emotional speech sentences of the south Indian speakers considering the native languages like Kannada and Telugu untrained database based on auto regressive modelling of vocal tract. Initially Prosodic features have been evaluated to analyse four emotional states namely ANGRY, NORMAL, HAPPY and SAD. We proposed an approach to analyse the role of vocal tract area variability as parameter in distinguishing the different emotional state speech processing using MATLAB newline