Normally Phonated Speech Recovery from Whispers by MELP

Abstract

In our study Mixed Excitation Linear Prediction (MELP) based system that reconstructs speech from whisper while preserving the individuality of a speaker's speech is developed. The proposed system can be used as speech prosthesis. To reconstruct speech from whisper, pitch generation by using the perceived pitch relationship with formant frequencies, formant and voicing modification steps were processed. To design and test the proposed system 20 speaker's whispered speech were recorded. In the analyzing phase the acoustic differences observed between normal and whispered speech were discussed. Acoustic parameters of the whispered speech modified in order to obtain a synthetic speech that is closer to normal speech and as the last part enhanced speech is synthesized by MELP.