Comprehension of synthetic speech with three text-to-speech systems using a sentence verification paradigm

Abstract
The comprehensibility of three text-to-speech synthesizers, namely DECtalk 2.0 (Perfect Paul), male voice of lnfovox SA-201, and SmoothTalker 3.0, was studied using a sentence verification task. Adult listeners heard true and false sentences of two different lengths. They first verified the truth value of a sentence and then they transcribed it. There were significant differences between DECtalk and lnfovox synthesizers in transcription accuracy and between infovox and the other two synthesizers in response latency.