English listening comprehension test using high-quality text-to-speech synthesis

Abstract

Due to the declaration of a state of emergency in the wake of the 2020 coronavirus pandemic, recording speech sounds for listening comprehension was impossible for two months. With this in mind, we investigated the feasibility of using text-to-speech (TTS) based on deep learning as an alternative method in an emergency. We created a test by synthesizing English speech using Google's Tacotron 2 based on the script for the listening section of The Common Test for University Admissions. In the experiment, 246 first-year students at national and public universities in the Tokyo metropolitan area answered a listening test with mixed synthesized speech. Along with the results of the experiment, we will also consider the advantages of using synthesized speech, which is not just a replacement for the recording process but also has the potential to be used for mass production of audio items.