Speaker based vocal tract shape estimation for kannada vowels

Abstract

Speech is the most convenient way of human communication. This paper describes the method of estimating basic acoustic features and vocal tract area function directly from the acoustics of speech wave form for estimation of vocal tract area function Linear Predictive Coding (LPC) based Auto regressive modelling of speech signal is used. LPC is one of power full speech analysis technique. This method predominant in speech researches to estimates basic acoustic features and Vocal Tract (VT) area function and to represent the speech for low bit rate transmission and storage applications. The vocal tract shape (area) estimation carried out by the approximation of acoustic tube model describes the phonetic distinctiveness and the individual speaker (gender) from the phonetics (vowels). The paper focused on the speaker specific dynamic changes of the area function. Paper presents the VT shape estimation for kannada vowels based on LPC, by considering the speaker individually among the vowels uttered by the different male and female speakers. The vowel utterances are recorded from the south Indian speakers. The process is implemented using MATLAB for inter speakers. The VT area functions for different speakers are investigated and the role of vocal tract variability as a parameter on speech analysis is carried out.