Esophageal speech enhancement using source synthesis and formant patterns modification

Abstract

This paper deals with esophageal speech, which is a voice ofsubstitution used by alaryngeal persons in order to be able tocommunicate with others. This voice, characterized by a low intensityand poor intelligibility, is hard to understand. In this paper, wepropose ideas to enhance this kind of voice. More precisely, we enhancethe source excitation signal and the formant structure of the speechvocal track. We modify pitch values by those of natural speech and wereplace the source by a synthetic one based on LF model. We alsoenhance the formant structure by enlarging formant bandwidth and weamplify their amplitudes without increasing the background noise. Then,we englobe all modification in the same scheme including allimprovements.