Analysis and mathematical modelling of Bangla sound units for synthetic voice generation

Abstract

This dissertation presents an analysis and mathematical modelling of Bangia sound units, which would enable to generate synthetic Banga speech. In this work, the spectral features and parameters of a speaker are investigated to determine their contribution to the occurrence of BangIa speech. The linear predictive coding (LPC) technique has been used to extract the spectral features and parameters, viz, pitch, gain, voiced I unvoiced decision etc., of Bangia speech. Use has been made of Hamming window in. this feature extraction. To develop the mathematical models of Bangia sound units, they (i. e., the sound units)are recorded for a particular person using a sound card. Then the spectral features and parameters of the sound units are extracted, and later on they are used to drive an all-pole digital filter The output of this filter is the desired synthetic speech .. Several software routines have been written for the analysis and synthesis of the Bangia sound units. These routines have been written in 'PASCAL' programming language. Finally, the modelled sound units are compared with the corresponding recorded speech. The play back facility of the sound card has been used for this purpose. Some of the pos9ible applications of the synthetic Bangia sound units are also explored ..