Automatic creation of TTS intelligibility tests

Abstract

This work presents a new method to automatize the creation and the analysis of subjective intelligibility tests for Text To Speech systems. In order to reach this goal, two main ideas have been adopted: the employment of multiple-choice answers and the use of algorithms able to automatically create the appropriate test set for intelligibility tasks. The reasons that led to the design of this new kind of intelligibility tests and the choices regarding the design of the algorithms are described. A practical example of the application of this new methodology has been experimented in the context of an online subjective test for intelligibility evaluation of two TTS voices.