Using speech in noise to improve understandability for elderly listeners

Abstract

This paper describes work designed to improve understandability of spoken output, specifically for the elderly, by using a speaking style employed by people to improve their understandability when speaking in poor channel conditions. We describe an experiment that shows the understandability gains that are possible using naturally-produced examples of this style. Additionally, we describe how to model this style, and evaluate the differences in understandability for speech synthesis produced using those models